A CHERN-CALABI FLOW ON HERMITIAN MANIFOLDS

XI SISI SHEN

Abstract. We study an analogue of the Calabi flow in the non-Kähler setting for com-
pact Hermitian manifolds with vanishing first Bott-Chern class. We prove a priori esti-
mates for the evolving metric along the flow given a uniform bound on the Chern scalar
curvature. From this, we show that the flow exists as long as the Chern scalar curvature
arXiv:2011.09683v1 [math.DG] 19 Nov 2020

remains bounded. If the Chern scalar curvature remains uniformly bounded for all time,
we show that the flow converges smoothly to the unique Chern-Ricci-flat metric in the
¯
∂ ∂−class of the initial metric.

1. Introduction
The Calabi flow was introduced by Calabi for Kähler metrics in [5, 6] and is defined
by
∂ω √
= −1∂ ∂¯ R, ω(0) = ω0 ,
∂t
√
where R = trω Ric(ω) = trω −1∂ ∂¯ log det g is the scalar curvature of a Kähler metric g,
with associated Kähler
√ form ω, and the flow preserves∞the Kähler classR of then metric. If
¯
we let ωϕ (t) = ω0 + −1∂ ∂ϕ(t) and normalize ϕ ∈ C (X) such that X ϕωϕ = 0, then
the flow can be represented in terms of ϕ by
∂ϕ
= Rϕ −R,
∂t
R
Rϕ ω n
where Rϕ is the scalar curvature of ωϕ and R = XR ωn ϕ is the average scalar curvature of
X ϕ
ωϕ on X which is independent of t. Short-time existence of the Calabi flow follows from
the fact that it is a fourth order quasilinear parabolic equation as shown by Chen-He [8].
They also proved the global existence of the flow under the assumption of a uniform Ricci
curvature bound [8]. It was shown by Székelyhidi in [23] that if the curvature tensor is
uniformly bounded along the Calabi flow and the Mabuchi energy is proper, then the flow
converges to a constant scalar curvature Kähler (cscK) metric. Chen-Sun prove in [11]
that if the Calabi flow starts at an initial metric that is sufficiently close to a cscK metric,
then the flow exists and converges uniformly to the cscK metric. Additional literature on
the Calabi flow can be found in [3, 7, 12, 13, 14, 16, 17, 18, 26].
¯
We define an analogue of the Calabi flow on the ∂ ∂-class of Hermitian metrics when
the first Bott-Chern class vanishes. Let (X, ω0 ) be a Hermitian manifold and let g0 be
the associated Hermitian metric to the real (1, 1)-form ω0 . The real (1, 1) Bott-Chern
cohomology is defined as
1,1 {d-closed real (1,1)-forms}
HBC (X, R) = √ ¯ ψ ∈ C ∞ (X, R)}
{ −1∂ ∂ψ,
and the first Bott-Chern class, denoted cBC ¯
1 (X), is given by the ∂ ∂−class of the Chern-
Ricci form,
√
Ric(ω) = − −1∂ ∂¯ log det g,

1
for any metric ω on X. Now, let us define the space of metrics
√
H = {ωϕ : ωϕ = ω0 + −1∂ ∂ϕ ¯ > 0, ϕ ∈ C ∞ (X, R)}.
In the setting of vanishing first Bott-Chern class, cBC 1 (X) = 0, Tosatti-Weinkove [28]
observed that one can define the Mabuchi energy Mabω0 (ωϕ ) : H → R by
1 det gϕ 1
Z   Z
Mabω0 (ωϕ ) = log − F ωϕn + F ω0n ,
V X det g0 V X
√
whereR F is the Chern-Ricci potential of ω 0 , that is Ric(ω 0 ) = ¯ , normalized so
−1∂ ∂F
F n n
R
that X e ω0 = X ω0 . This definition of Mabuchi energy agrees with the formula in the
Kähler setting [24] (see also Section 9 of [28]). We remark that similar to how there are
several generalizations of the Kähler-Ricci flow in the non-Kähler setting [15, 22, 29, 30],
the Chern-Calabi flow we consider in this paper may not be the only generalization of the
Calabi flow evolving by the Chern scalar curvature. Other flows of Hermitian metrics in
the non-Kähler setting have been studied by [1, 2, 4, 19].
Assume ¯ k
√ that our Hermitian metric ω0 satisfies ∂ ∂ω0 = 0 for R k = 1, 2 and let ωϕ =
ω0 + −1∂ ∂ϕ ¯ for a smooth function ϕ normalized so that ϕω n = 0. We consider
X ϕ
a gradient flow of the Mabuchi energy defined above, starting at ω0 . This flow can be
expressed in terms of the potential ϕ by
∂ϕ det gϕ
(1) = Rϕ +2 Reh(tr Tϕ ), ∂(log − F )iϕ , ϕ(0) = 0,
∂t det g0
where Rϕ = trωϕ Ric(ωϕ ) is the Chern scalar curvature of ωϕ and tr Tϕ = (Tϕ )pp• is the
trace of the torsion of ωϕ . We note that R = 0 since cBC 1 (X) = 0. When the metric is
Kähler, this flow agrees with the Calabi flow. We note that we immediately obtain short-
time existence of this flow since its leading order term is a strictly elliptic fourth-order
operator and so the equation is a fourth order quasilinear parabolic equation, following
the same line of reasoning as in [8]. Given that the Mabuchi energy is decreasing along
this flow, we prove a new estimate, in addition to a result from [21], to obtain higher order
estimates on ϕ. In this paper, we prove

Theorem 1. Let (X, ω0 ) be a Hermitian manifold with cBC ¯ k

1 (X) = 0 and ∂ ∂ω0 = 0 for
k = 1, 2. A solution to the flow given by Equation (1) starting at ω0 exists as long as the
Chern scalar curvature remains bounded along the flow. In addition, if the Chern scalar
curvature remains uniformly bounded for all time, we have smooth
√ convergence of the flow
to the unique Chern-Ricci-flat metric of the form ω∞ = ω0 + −1∂ ∂ϕ ¯ ∞ for some smooth
function ϕ∞ on X.

Our assumption that ω0 satisfy ∂ ∂ω ¯ k = 0 for

¯ k = 0 for k = 1, 2, is in fact equivalent to ∂ ∂ω
0 0
n
R
k = 1, 2, . . . , n − 1, and allows us to ensure that the volume V = X ωϕ remains unchanged
along the flow and to obtain the estimates needed to prove long-time existence of the flow
under a Chern scalar curvature bound.
This paper is structured as follows:
• In Section 2, we cover the notation and basic properties of Hermitian metrics that
we will need in the next sections.
• In Section 3, we discuss properties of this flow. We first show that this flow is
indeed a gradient flow of the Mabuchi energy. We then prove that its fixed points
are precisely the constant Chern scalar curvature metrics, which when cBC
1 (X) = 0,
are in fact Chern-Ricci-flat metrics. Lastly, we discuss short-time existence of the
flow.

2
 • In Section 4, we first prove a priori estimates needed in order to show long-time
existence of the flow following the methods of Chen-Cheng in [10] and previous
work by the author in [21]. Finally, using these estimates we prove Theorem 1.

2. Preliminaries
In this section, we include several well-known identities that will be needed for computa-
tions in the next sections (see also Section 2 of [29]).
Let X be a compact complex manifold of complex dimension n. We will work in complex
coordinates z 1 , . . . , z n and write tensors in terms of this coordinate system. Let g = gij̄
√
be a Hermitian metric on X with associated (1, 1)-form ω = −1gij̄ dz i ∧ dz j where all
repeated indices are being summed from 1 to n. We will also refer to ω as a Hermitian
metric.
Let ∇ be the Chern connection associated to ω, defined for a (1, 0)-form a = ak dz k
by
(2) ∇i ak = ∂i ak − Γjik aj , ∇i ak = ∂i ak
and for a vector field X = X k ∂k by
∇i X k = ∂i X k + Γkij X j , ∇i X k = ∂i X k
where Γkij = gkp̄ ∂i gj p̄ is the Christoffel symbol of g and g kp̄ gip̄ = δik . The Chern connection
is compatible with the metric g, that is, ∇k gij̄ = 0 ∀i, j, k. We define the trace of a real
(1, 1)-form α = αij̄ dz i ∧ dz j with respect to ω by
nω n−1 ∧α
trω α = gij̄ αij̄ = ωn .
√
¯ be another Hermitian
The torsion of g is defined by Tijk = Γkij − Γkji . Let ωϕ = ω + −1∂ ∂ϕ
metric on X. From this definition, it is clear that
(∂ω)jkℓ̄ = (∂ωϕ )jkℓ̄
where (∂ω)jkℓ̄ = ∂j gkℓ̄ − ∂k gj ℓ̄ . Denoting the torsion of ωϕ by T̃ , it follows that
p q
(3) Tjk gpℓ̄ = (∂ω)jkℓ̄ = (∂ωϕ )jkℓ̄ = T̃jk g̃qℓ̄ ,
where g̃ij̄ is the metric in coordinates for ωϕ . For simplicity, we will use the notation
q p
Tjkℓ̄ = Tjk gqℓ̄ , T̃jkℓ̄ = T̃jk g̃pℓ̄
and so the first equality in Equation (3) can be rewritten as Tjkℓ̄ = T̃jkℓ̄ . In addition, we
p
will let (tr T )j denote Tpj . The curvature tensor is defined by
Rij̄k p = −∂j̄ Γpik , Rij̄kℓ̄ = gpℓ̄ Rij̄k p

where we note that Rij̄kℓ̄ = Rj īℓk̄ . We can commute the indices of the curvature tensor as
follows:
(4) Rij̄k p − Rkj̄i p = ∂j̄ Γpki − ∂j̄ Γpik = ∂j̄ Tki
p
.

The Chern-Ricci curvature of ω is defined by

Rij̄ = gkℓ̄ Rij̄kℓ̄ = −∂i ∂j̄ log det g,
its associated form by
√
Ric(ω) = −1Rij̄ dz i ∧ dz j

3
and its Chern scalar curvature by
R(ω) = g ij̄ Rij̄ = trω Ric(ω).

The following commutation formulae will be useful to us in the later sections. For a
(1, 0)-form a = ak dz k ,

[∇i , ∇j̄ ]ak = −Rij̄k ℓ aℓ , [∇i , ∇j̄ ]al = Rij̄ k̄ℓ̄ ak

(5)
[∇i , ∇j ]ak = −Tijr ∇r ak , [∇ī , ∇j̄ ]ak = −Tijr ∇r̄ ak
and for a scalar function f , we have
(6) [∇i , ∇j ]f = −Tijr ∇r f.

The Chern Laplacian with respect to g of a function f is defined by

√
∆f = trω −1∂ ∂f ¯ = gij̄ ∂i ∂ f = gij̄ ∇i ∇ f.
j̄ j̄

For a complex manifold, if we assume that ∂ ∂ω ¯ k = 0 for k = 1, 2, then in fact ∂ ∂ω

¯
vanishes for all k = 1, . . . , n − 1, which follows from a straightforward computation. Under
this assumption, the volume of the metric remains unchanged up to addition of ∂ ∂¯ of a
smooth function, that is,
√
Z Z
(7) ¯
(ω + −1∂ ∂ψ) = n
ωn,
X X
√
for all ψ such that ω + ¯ > 0.
−1∂ ∂ψ
¯ n−1 = 0, ensures the vanishing of the integrals of Chern
The Gauduchon condition, ∂ ∂ω
Laplacians of functions:
√ √
Z Z Z
n
∆f ω = n ¯
−1∂ ∂f ∧ ω n−1
=n f −1∂ ∂ω¯ n−1 = 0.
X X X

We will need the following divergence theorem in the non-Kähler setting (see Lemma 1 of
[20]):

Lemma 1. For any Hermitian metric ω and V ∈ Γ(X, T 1,0 , X), we have that
Z Z
∇i V i ω n = (tr T )i V i ω n ,
X X
p
where ∇ is the Chern connection with respect to ω and (tr T )i = Tpi is the trace of the
torsion of ω.

Using this divergence theorem, we can show the following:

Lemma 2. For any metric g with associated (1, 1)-form ω, under the assumption that ω
¯ n−1 = 0, we have that
is Gauduchon, i.e. ∂ ∂ω

gj k̄ ∇k̄ (tr T )j − (tr T )k (tr T )j = 0,

where ∇ is the Chern connection with respect to ω and T is the torsion of the Chern
connection with respect to ω.

4
Proof. This identity follows directly from the Gauduchon condition. One way to see this
simply is that for any smooth function u on X, it follows from the Gauduchon condition
and Lemma 1 that
Z Z Z
2 n n
0= ∆u ω = 2 u∆uω + 2 g j k̄ ∇j u∇k̄ uω n
X X X
Z Z Z
j k̄ n
= −2 g ∇j u∇k̄ uω + 2 g (tr T )j u∇k̄ uω n + 2
j k̄
gj k̄ ∇j u∇k̄ uω n
Z X X X
j k̄ n
=2 g (tr T )j u∇k̄ uω
Z X
= gj k̄ (tr T )j ∇k̄ u2 ω n
X
Z
= − (g j k̄ (∇k̄ (tr T )j − (tr T )k (tr T )j )u2 ω n .
X

Since this holds for arbitrary u, it follows that gj k̄ (∇k̄ (tr T )j − (tr T )k (tr T )j ) = 0. We
note that if X were not compact, this identity would still hold since we could take the
function u to be compactly supported. 

This outlines the key identities and formulae that we will need for the computations in
this paper. Note that throughout this paper, the constants may vary from one line to
another.

3. Properties of the flow

In this section, we prove that the flow defined in Equation (1) is indeed a gradient flow
of the Mabuchi energy, that fixed points of the flow are exactly those metrics that are
Chern-Ricci-flat, and that we have short-time existence of the flow.
¯ k = 0 for k = 1, 2. We will assume that ω0 is nor-
Let ω0 be Hermitian onRX satisfying ∂ ∂ω
n
malized such that V = X ω0 = 1 and we note that this integral remains unchanged along
the flow, see Equation (7). Using the assumption that cBC
1 (X) = 0, we have that
√
¯
Ric(ω0 ) = −1∂ ∂F,
for
R aF smooth R function F which we call the Chern-Ricci potential, normalized so that
e ω n = ω n . Let us define
X 0 X 0
√
¯ > 0, ϕ ∈ C ∞ (X)}.
H = {ωϕ : ωϕ = ω0 + −1∂ ∂ϕ
We will show that the flow defined in Equation (1) is a gradient flow of the Mabuchi energy
Mabω0 : H → R defined by
ωϕn
Z   Z
n
Mabω0 (ωϕ ) = log n − F ωϕ + F ω0n .
X ω 0 X
One can check that this definition of Mabuchi energy agrees with the formula in the Kähler
setting [24] (see also Section 9 of [28]).
To set some notation, let Ω := eF ω0n be a volume form on X which satisfies
√
−1∂ ∂¯ log Ω = 0
by the fact that F is the Chern-Ricci potential for ω0 . In this way, we see that the Chern
scalar curvature of ωϕ can be expressed as
ωϕn
Rϕ = −∆ϕ log .
Ω

5
Lemma 3. The Mabuchi energy is decreasing along the flow defined in Equation (1).

Proof. Taking the time derivative of Mabuchi energy, we have that

ωϕn
Z Z 
∂ n

Mabω0 (ωϕ ) = ∆ϕ ϕ̇ωϕ + log n − F ∆ϕ ϕ̇ωϕn
∂t X X ω0
Z n
ωϕ
= log F n ∆ϕ ϕ̇ωϕn
X e ω0

˜ be the Chern connection with respect to ωϕ . Substituting Ω = eF ω n and integrating

Let ∇ 0
by parts using Lemma 1, we have that
ωϕn
Z
∂
Mabω0 (ωϕ ) = log ∆ϕ ϕ̇ωϕn
∂t X Ω
ωϕn ωϕn
Z Z
j k̄ ˜ ˜ n ˜ ϕ̇ω n
= − (gϕ ) ∇j log ∇k̄ ϕ̇ωϕ + (gϕ )j k̄ (tr Tϕ )j log ∇
X Ω X Ω k̄ ϕ
ωn ωϕn n
Z Z
= ˜ j∇
(gϕ )j k̄ ∇ ˜ log ϕ ϕ̇ωϕn − (g ϕ ) j k̄
(tr T ) ∇
ϕ k j
˜ log ϕ̇ωϕ
k̄
X Ω X Ω
ωϕn ωϕn n
Z Z
j k̄ ˜ n
− (gϕ ) ∇k̄ ((tr Tϕ )j log )ϕ̇ωϕ + (gϕ )j k̄ (tr Tϕ )k (tr Tϕ )j log ϕ̇ωϕ
X Ω X Ω
ωn 
Z 
=− ϕ̇ Rϕ + 2 Re((gϕ )j k̄ (tr Tϕ )j ∇ ˜ log ϕ ) ωϕn ,
k̄
X Ω
√
where the second last step follows from the fact that −1∂ ∂¯ log Ω = 0 and the last step
uses Lemma 2. Since this flow is defined by the following evolution equation
ωϕn
ϕ̇ = Rϕ + 2 Re((gϕ )j k̄ (tr Tϕ )j ∂k̄ log
),
Ω
we immediately have that the Mabuchi energy is decreasing along this flow. 

We now check that all fixed points of this flow are of constant Chern scalar curvature
which, in the setting that cBC
1 (X) = 0, is equivalent to being Chern-Ricci-flat.

Lemma 4. A metric is a fixed point of the flow defined by Equation (1) if and only if it
is Chern-Ricci-flat.

Proof. Assume that ωϕ is Chern-Ricci-flat. Then we have that

ωϕn
0 = Rϕ = −∆ϕ log ,
Ω
ωn
which implies that log Ωϕ = const. since we are working on a compact manifold and
integration by parts gives
Z Z Z Z
2 n n j k̄ n
(8) 0 = ∆u ω = 2 u∆uω + 2 g ∇j u∇k̄ uω = 2 |∇u|2g ω n ⇒ u = const.,
X X X X
ωn
where we used that ω is Gauduchon in the first equality. Using that log Ωϕ = const, we
have that
ωϕn
ϕ̇ = Rϕ + 2 Re((gϕ )j k̄ (tr Tϕ )j ∂k̄ log ) = 0.
Ω
Thus, we have showed that a Chern-Ricci-flat metric is a fixed point of this flow.

6
In order to show the reverse direction, let us now assume that ωϕ is a fixed point of the
ωn ˜ be the
flow, that is, ϕ̇ = 0. This means that Rϕ + 2 Re((gϕ )j k̄ (tr Tϕ )j ∂k̄ log Ωϕ ) = 0. Let ∇
Chern connection with respect to ωϕ . Integrating against an arbitary smooth function u
and using Lemma 1 and Lemma 2, we have that
ωn
Z
0= u(Rϕ + 2 Re((gϕ )j k̄ (tr Tϕ )j ∇ ˜ log ϕ ))ωϕn
k̄
X Ω
Z
ω n Z
ωϕn n
=− ˜ j∇
u(gϕ )j k̄ ∇ ˜ log ϕ ωϕn + 2 u Re((g ϕ )j k̄
(tr T ) ∇˜
ϕ j k̄ log )ω
X
k̄
Ω X Ω ϕ
ωϕn n ωn
Z Z
= ˜ j k̄ ˜
∇j u(gϕ ) ∇k̄ log ωϕ + u(gϕ )j k̄ (tr Tϕ )k ∇ ˜ j log ϕ ωϕn
X Ω X Ω
ωϕn n ω n ωϕn n
Z Z Z
=− ∆ϕ u log ωϕ + ˜ j u(tr Tϕ )k log ϕ ω n +
(gϕ )j k̄ ∇ u(gϕ )j k̄
(tr T ) ˜
∇
ϕ k j log ω
X Ω X Ω ϕ X Ω ϕ
ωϕn n
Z
=− ∆ϕ u log ω .
X Ω ϕ
n
ωϕ
Choosing u = log Ω ,
we have by Equation (8) that
ωϕn 2 n
Z
0= ˜
|∇(log )| ω ≥ 0
X Ω ϕ ϕ
ωn √ n
ωϕ
and so log Ωϕ = const. which implies that Ric(ωϕ ) = − −1∂ ∂¯ log Ω = 0 and so ωϕ is
Chern-Ricci-flat. 

We now discuss short-time existence of this flow, following from [8].

Lemma 5. There exists a unique solution ϕ(t) satisfying Equation (1) for t ∈ [0, T ).

Proof. The flow evolves the potential function ϕ by

∂ϕ det gϕ
= Rϕ +2 Reh(tr Tϕ ), ∂ log F iϕ .
∂t e det g0
We see that this flow differs from the Calabi flow only by addition of lower order terms in
ϕ. Thus, the short-time existence of this flow follows the same way as for the Calabi flow,
using standard parabolic theory, see [8] and references therein. 

4. Proof of the main theorem

Our goal in this section is to prove a priori estimates for ϕ along this flow, conditional on
a bound on the Chern scalar curvature of the evolving metric, and then use this to prove
Theorem 1.
Let us begin with showing the a priori estimates:

Theorem 2. Let (X n , ω) be a compact Hermitian √ manifold of dimension n such that ω

¯ k
satisfies ∂ ∂ω = 0 for k = 1, 2. If ωϕ = ω + −1∂ ∂ϕ ¯ for a smooth potential function ϕ
on X, then for all k, there exists C(k) depending only on (X, ω), || Rϕ ||0 , and an upper
bound for Mabω (ωϕ ) such that ||ϕ||C k (X,ω) ≤ C(k).

7
Proof. The entropy quantity Ent(ω, ωϕ ) used in [21] is implied by a bound on the Mabuchi
energy Mabω (ωϕ ) since
ωϕn n
Z Z Z
n
Ent(ω, ωϕ ) := log n ωϕ = Mabω (ωϕ ) − F ωϕ + F ωn
X ω X
Z X
≤ Mabω (ωϕ ) + sup |F | ωϕn + C
X X
≤ Mabω (ωϕ ) + C,
where the inequalities follow from the fact that F is the Chern-Ricci potential of the fixed
metric ω. In what follows, we will pass the dependencies of the constants on Ent(ω, ωϕ )
to Mabω (ωϕ ).
The proof follows in almost exactly the same manner as in Chen-Cheng [10] using what
has already been proven in the non-Kähler setting in [21]. We include a proof here to
show how we account for the torsion terms that result from handling the derivative terms
of Rϕ in the L∞ estimate that are no longer assumed to vanish. We consider the coupled
second order equations
n
ωϕ
F = log ωn
(9)
∆ϕ F = −Rϕ + trωϕ Ric(ω).

For the C 0 , C 1 and Lp estimates, lifting the assumption that Rϕ = const. and noting
the fact that there is never an instance where we need to differentiate Rϕ , the estimates
work out exactly the same way as in [21] where the new bounds will additionally depend
on ||Rϕ ||0 . It is only in the proof of the L∞ estimate that we need to differentiate Rϕ
and here is where the proof will somewhat differ from the constant Chern scalar curvature
case (see Section 4 of [21]). Specifically, the term of the derivative of ∆ϕ F arising from
the computation of ∆ϕ |∂F |2ωϕ must be handled using integration by parts, and the torsion
integral arising from this integration by parts will need to be controlled.

Lemma 6. Let (ϕ, F ) be a smooth solution to (9). Then there exists a constant C de-
pending only on (X, ω), ||Rϕ ||0 and Mabω (ωϕ ), such that

max(trω ωϕ ) + max |∂F |2ωϕ ≤ C.

X X

˜ R̃ and T̃ denote, respectively, the Chern

For the purposes of simplifying notation, let ∇,
connection, curvature tensor and torsion with respect to g̃, the associate metric to ωϕ .
Commuting derivatives using the identities in (5) and (6), we have the following:

∆ϕ (|∂F |2ωϕ ) = g̃ ij̄ g̃pq̄ (∇˜ p∇˜ i∇˜ F + T̃ r ∇ ˜ r ˜ ˜ r

j̄ pi r Fj̄ )Fq̄ + T̃qj ∇i Fr̄ Fp + ∇i T̃qj Fr̄ Fp

+ (∇ ˜ q̄ ∇
˜ i∇
˜ j̄ F + R̃ ℓ̄ F )Fp + |∇


˜ ∇F
˜ |2 + |∇ ¯˜ |2
˜ ∇F
iq̄ j̄ ℓ̄ ωϕ ωϕ
r ˜ ˜ i T̃ r Fr̄ Fp
= g̃pq̄ ∆ϕ F )p Fq̄ + 2Re(g̃ij̄ g̃pq̄ T̃pi ∇r Fj̄ Fq̄ ) + g̃ij̄ g̃pq̄ ∇ qj
(10) pq̄ ij̄ pq̄
+ g̃ (∆ F ) F + g̃ g̃ R̃ F F + |∇ ℓ̄ ˜ ∇F
˜ | + |∇
2 ¯˜ |2
˜ ∇F
ϕ q̄ p iq̄ j̄ ℓ̄ p ωϕ ωϕ
˜ r Fj̄ Fq̄ )
= g̃pq̄ ∆ϕ F )p Fq̄ + 2Re(g̃ij̄ g̃pq̄ g̃rk̄ T̃pik̄ ∇
˜ i (T̃qj t̄ )Fr̄ Fp + g̃ pq̄ (∆ϕ F )q̄ Fp + g̃ ij̄ g̃ pq̄ g̃ kℓ̄ R̃iq̄kj̄ F Fp
+ g̃ij̄ g̃ pq̄ g̃ tr̄ ∇ ℓ̄
˜ ˜
+ |∇∇F | + |∇∇F | 2 ˜ ¯
˜ 2
ωϕ ωϕ

8
We now commute indices of the curvature tensor as in 4 and trace to get that
˜ r Fj̄ Fq̄ )
∆ϕ (|∂F |2ωϕ ) = 2Re(g̃pq̄ ∆ϕ F )p Fq̄ ) + 2Re(g̃ij̄ g̃ pq̄ g̃ rk̄ T̃pik̄ ∇
(11) ˜ i (T̃qj t̄ )Fr̄ Fp + g̃pq̄ g̃kℓ̄ R̃kq̄ F Fp
+ g̃ ij̄ g̃pq̄ g̃tr̄ ∇ ℓ̄

− g̃ pq̄ g̃ kℓ̄ g̃rs̄ ∇ ˜ ∇F

˜ q̄ (T̃rks̄ )F Fp + |∇ ¯˜ |2 .
˜ ∇F
˜ |2ω + |∇
ℓ̄ ϕ ωϕ

For a general real-valued function A(F ),

e−A(F ) ∆ϕ (eA(F ) |∂F |2ωϕ ) = ∆ϕ (|∂F |2ωϕ ) + 2A′ Re(g̃ ij̄ g̃kℓ̄ (Fi Fk Fℓ̄j̄ + Fi Fℓ̄ Fkj̄ ))
+ (A′2 + A′′ )|∂F |4ωϕ + A′ ∆ϕ F |∂F |2ωϕ ,
˜ j̄ ∇
where Fℓ̄j̄ denotes ∇ ˜ F . Substituting (10) for the first term in the above equation and
ℓ̄
using the completed square:
˜ ∇F
A′2 |∂F |4ωϕ + 2A′ Re(g̃ ij̄ g̃kℓ̄ Fi Fk Fℓ̄j̄ ) + |∇ ˜ |2 ≥ 0,
ωϕ

we obtain that
˜ r Fj̄ Fq̄ )
e−A(F ) ∆ϕ (eA(F ) |∂F |2ωϕ ) ≥ 2Re(g̃pq̄ (∆ϕ F )p Fq̄ ) + 2Re(g̃ij̄ g̃pq̄ g̃rk̄ T̃pik̄ ∇
˜ i (T̃qj t̄ )Fr̄ Fp + g̃pq̄ g̃kℓ̄ R̃kq̄ F Fp
+ g̃ ij̄ g̃pq̄ g̃tr̄ ∇ ℓ̄
pq̄ k ℓ̄ rs̄ ˜ ˜ ¯
˜
− g̃ g̃ g̃ ∇q̄ (T̃rks̄ )Fℓ̄ Fp + |∇∇F |ωϕ + 2A′ g̃ij̄ g̃ kℓ̄ Fi Fℓ̄ Fkj̄
2

+ A′′ |∂F |4ωϕ + A′ ∆ϕ F |∂F |2ωϕ .

Applying the relation R̃kq̄ = Rkq̄ − Fkq̄ to switch the Ricci curvature of ωϕ to that of ω,
e−A(F ) ∆ϕ (eA(F ) |∂F |2ωϕ )
˜ r Fj̄ Fq̄ ) + g̃ ij̄ g̃ pq̄ g̃tr̄ ∇
≥ 2Re(g̃pq̄ (∆ϕ F )p Fq̄ ) + 2Re(g̃ ij̄ g̃ pq̄ g̃ rk̄ T̃pik̄ ∇ ˜ i (T̃qj t̄ )Fr̄ Fp
˜ (T̃ )F F + |∇
+ g̃pq̄ g̃ kℓ̄ R F F − g̃pq̄ g̃ kℓ̄ g̃ rs̄ ∇
k q̄ ℓ̄ p q̄
˜ ∇F˜¯ |2
rks̄ ℓ̄ p ωϕ
′ ij̄ k ℓ̄ ′′
(12) + (2A − 1)g̃ g̃ Fi Fℓ̄ Fkj̄ + A |∂F |4ωϕ + A ∆ϕ F |∂F |2ωϕ
′

˜ r Fj̄ Fq̄ ) + g̃ ij̄ g̃ pq̄ g̃tr̄ ∇

≥ 2Re(g̃pq̄ (∆ϕ F )p Fq̄ ) + 2Re(g̃ ij̄ g̃ pq̄ g̃ rk̄ T̃pik̄ ∇ ˜ i (T̃qj t̄ )Fr̄ Fp
˜ (T̃ )F F + (1 − (A′ − 1 ))|∇
+ g̃pq̄ g̃ kℓ̄ R F F − g̃pq̄ g̃ kℓ̄ g̃ rs̄ ∇ ˜ ∇F¯˜ |2
k q̄ ℓ̄ p q̄ rks̄ℓ̄ p 2 ωϕ
′′
+ (A − (A′ − 12 ))|∂F |4ωϕ +A ′
∆ϕ F |∂F |2ωϕ ,
where we used the following Cauchy-Schwarz inequality:
¯˜ |2
˜ ∇F
(2A′ − 1)g̃ ij̄ g̃kℓ̄ Fi Fℓ̄ Fkj̄ ≥ −(A′ − 12 )|∂F |4ωϕ − (A′ − 12 )|∇ ωϕ

for A′ > 21 .
In order to control the bad torsion terms (the second, third and fifth terms in the last line of
(12)), we will need to strategically choose our function A(F ) to ensure that 1−(A′ − 21 ) > 0
and A′′ − (A′ − 12 ) > 0. We can accomplish this by choosing
A(F ) = κeF + F ( 12 − ε),
1
so that A′ (F ) = κeF + 2 − ε and A′′ (F ) = κeF . We can then pick ε, κ > 0 such that

(
κeminX F − ε ≥ 0
0 ≤ A′′ − ε = A′ − 1
2 ≤ 1
2 ⇔
κemaxX F − ε ≤ 12 .

9
We can first choose κ small enough such that κemaxX F ≤ 21 . Then choose ε small enough
such that κemin F ≥ ε. This ensures that A′ ∈ ( 21 , 1).
It follows that
e−A(F ) ∆ϕ (eA(F ) |∂F |2ωϕ )
≥ 2Re g̃ pq̄ ∆ϕ F )p Fq̄ + 2Re(g̃ij̄ g̃pq̄ g̃rk̄ T̃pik̄ ∇ ˜ r Fj̄ Fq̄ ) + g̃ij̄ g̃ pq̄ g̃ tr̄ ∇
˜ i (T̃qj t̄ )Fr̄ Fp

˜ q̄ (T̃rks̄ )F Fp + 1 |∇
+ g̃pq̄ g̃kℓ̄ Rkq̄ Fp Fℓ̄ − g̃pq̄ g̃ kℓ̄ g̃ rs̄ ∇ ¯˜ |2 + ε|∂F |4 + A′ ∆ F |∂F |2
˜ ∇F
ℓ̄ 2 ωϕ ωϕ ϕ ωϕ
pq̄ ij̄ pq̄ r k̄ ˜ r Fj̄ Fq̄ ) + g̃ g̃ g̃ ∂i Tqj t̄ Fr̄ Fp
i j̄ pq̄ tr̄


≥ 2Re g̃ ∆ϕ F )p Fq̄ + 2Re(g̃ g̃ g̃ Tpik̄ ∇
− g̃ij̄ g̃ pq̄ g̃ tr̄ g̃ sk̄ ∂i g̃tk̄ Tqjs̄ Fr̄ Fp + g̃pq̄ g̃ kℓ̄ Rkq̄ Fp Fℓ̄ − g̃ pq̄ g̃ kℓ̄ g̃rs̄ ∂q̄ (Trks̄ )Fℓ̄ Fp
+ g̃pq̄ g̃kℓ̄ g̃ rs̄ g̃ij̄ ∂ g̃ T F F + 1 |∇ ˜ ∇F ¯˜ |2 + ε|∂F |4 − |∆ F ||∂F |2
q̄ is̄ rk j̄ ℓ̄ p 2 ωϕ ωϕ ϕ ωϕ

where we converted the covariant derivatives to partial derivatives as in (2) and passed
the torsion terms of g̃ to those of g as in (3).
Applying Young’s inequality and choosing B to be at least 3(n − 1), where the factor of
n − 1 comes from the fact that trω ωϕ ≤ C(trωϕ ω)n−1 , we have
e−A(F ) ∆ϕ (eA(F ) |∂F |2ωϕ )
˜ r Fj̄ Fq̄ )
≥ 2Re(g̃pq̄ ∆ϕ F )p Fq̄ ) + 2Re(g̃ij̄ g̃pq̄ g̃rk̄ Tpik̄ ∇
+ g̃ ij̄ g̃ pq̄ g̃tr̄ ∂i Tqj t̄ Fr̄ Fp − g̃ij̄ g̃ pq̄ g̃ tr̄ g̃ sk̄ ∂i g̃tk̄ Tqjs̄ Fr̄ Fp + g̃pq̄ g̃kℓ̄ Rkq̄ Fp Fℓ̄
− g̃ pq̄ g̃ kℓ̄ g̃rs̄ ∂ (T )F F + g̃ pq̄ g̃ kℓ̄ g̃rs̄ g̃ ij̄ ∂ g̃ T F F + 1 |∇ ˜ ∇F¯˜ |2
q̄ rks̄ℓ̄ p q̄ is̄ rk j̄ ℓ̄ p 2 ωϕ

+ ε|∂F |4ωϕ − |∆ϕ F ||∂F |2ωϕ

¯˜ |2
˜ ∇F
≥ 2Re(g̃pq̄ ∆ϕ F )p Fq̄ ) − C(trω ωϕ )B g ij̄ g̃ kℓ̄ g̃ pq̄ ∂i g̃kq̄ ∂j̄ g̃pℓ̄ + 14 |∇ ωϕ

− C(trω ωϕ )B |∂F |2ωϕ − C(trω ωϕ )B ,

where the constant in front of the fourth term depends on ||Rϕ ||0 . Now, we use the
following computation in the proof of Equation (9.5) of [29] for ∆ϕ trω ωϕ :
p p q
∆ϕ trω ωϕ = g̃pj̄ g̃ iq̄ gkℓ̄ ∇k g̃ij̄ ∇ℓ̄ g̃pq̄ + 2Re(g̃ij̄ gkℓ̄ Tki ∇ℓ̄ g̃pj̄ ) + g̃ ij̄ gkℓ̄ Tik Tjℓ g̃pq̄
ℓ + g̃ ij̄ g k ℓ̄ ∇ T p − g̃ ij̄ g k ℓ̄ g̃ (∇ T q − R
+ gij̄ Fij̄ − R + g̃ ij̄ ∇i Tjℓ pq̄
ℓ̄ ik k q̄ i jℓ iℓ̄pj̄ g )
p q
− g̃ij̄ g kℓ̄ Tik Tjℓ gpq̄ .
Converting the first term into covariant derivatives and applying Young’s inequality, we
have
g̃ pj̄ g̃iq̄ gkℓ̄ ∇k g̃ij̄ ∇ℓ̄ g̃pq̄ ≥ g̃ pj̄ g̃iq̄ gkℓ̄ ∂k g̃ij̄ ∂ℓ̄ g̃pq̄ − 2ε g̃pj̄ g̃iq̄ gkℓ̄ ∂k g̃ij̄ ∂ℓ̄ g̃pq̄ − C(trω ωϕ )n .
Likewise, the second term can be bounded below by
p
2Re(g̃ij̄ g kℓ̄ Tki ∇ℓ̄ g̃pj̄ ) ≥ − 2ε g̃ pj̄ g̃iq̄ gkℓ̄ ∂k g̃ij̄ ∂ℓ̄ g̃pq̄ − C(trω ωϕ )n ,
and the fourth term by
|∇ ¯
˜ |2
˜ ∇F
gij̄ Fij̄ ≥ − − Cδ(trω ωϕ )2 .
ωϕ
δ
It is straightforward to see that the remaining terms can be bounded below by −C(trω ωϕ )n .
Choosing B ≥ n and δ = 4e−A(F ) N (B + 1)(trω ωϕ )B , we arrive at the following:
∆ tr ω ≥ (1 − ε)g̃ pj̄ g̃ iq̄ gkℓ̄ ∂ g̃ ∂ g̃ − eA(F ) ¯˜ |2 − C(tr ω )B+2 .
˜ ∇F
|∇
ϕ ω ϕ k ij̄ ℓ̄ pq̄ 4N (B+1)(tr ω ωϕ )B ωϕ ω ϕ

Observe that

10
 ∆ϕ (trω ωϕ )B+1 = (B + 1)B(trω ωϕ )B−1 |∂ trω ωϕ |2ωϕ + (B + 1)(trω ωϕ )B ∆ϕ trω ωϕ
(13)
≥ (B + 1)(trω ωϕ )B ∆ϕ trω ωϕ .

Choosing N sufficiently large and letting Q := eA(F ) |∂F |2ωϕ + N (trω ωϕ )B+1 , we arrive at

∆ϕ Q = ∆ϕ (eA(F ) |∂F |2ωϕ + N (trω ωϕ )B+1 )

≥ 2eA(F ) Re(g̃pq̄ ∆ϕ F )p Fq̄ ) − C(trω ωϕ )B |∂F |2ωϕ − C(trω ωϕ )2B+2
≥ 2eA(F ) Re(g̃pq̄ ∆ϕ F )p Fq̄ ) − C(trω ωϕ )B+1 Q.

Now we demonstrate how to handle the non-vanishing first derivative of ∆ϕ F in the first
term on the right-hand side of the above inequality. We will use integration by parts on
this term in order to bound it, this follows from the same argument used by Chen-Cheng
in Section 3 of [10]. Let us compute

∆ϕ Q2p+1 = (2p + 1)2p|∇Q|2ωϕ Q2p−1 + (2p + 1)Q2p ∆ϕ Q.

Integrating both sides and applying our bound from Equation 13, we have that

Z Z
2p|∇Q|2ωϕ Q2p−1 ωϕn = Q2p (−∆ϕ Q)ωϕn
X X
Z
≤ −2 eA(F ) Re(g̃ pq̄ ∆ϕ F )p Fq̄ )Q2p ωϕn
X
Z
+C (trω ωϕ )B+1 Q2p+1 ωϕn .
X

The first integral on the right-hand side can be integrated by parts using Lemma 1 in the
following way

Z Z
A(F ) pq̄
−2 e Re(g̃ ∆ϕ F )p Fq̄ )Q2p ωϕn =2 Re(g̃ pq̄ ∇p (eA(F ) Fq̄ Q2p ))∆ϕ F ωϕn
X X
Z
−2 Re(g̃pq̄ eA(F ) Fq̄ Q2p (tr T̃ )p )∆ϕ F ωϕn
X

Now, the first term on the right-hand side of the above equation can be handled exactly
as in Equations (3.14)-(3.18) of [10]. Let us demonstrate how we can handle the second
integral with the torsion term.

s = g̃ sℓ̄ T p g and that we can bound the ∆ F term by

Firstly, note that (tr T̃ )p = T̃sp sp pℓ̄ ϕ

|∆ϕ F | ≤ ||Rϕ ||0 + || trωϕ Ric ||0 ≤ C(1 + (trω ωϕ )n−1 ).

11
Computing, the integral involving torsion can be controlled as follows:
Z Z
pq̄ A(F ) 2p n
−2 Re(g̃ e Fq̄ Q (tr T̃ )p )∆ϕ F ωϕ = −2 Re(g̃pq̄ eA(F ) Fq̄ Q2p g̃ sℓ̄ Tsp
r
grℓ̄ )∆ϕ F ωϕn
X X
Z
≤2 ˜ |ωϕ |T r |ωϕ |∆ϕ F |ω n
eA(F ) Q2p (trωϕ ω)|∇F sp ϕ
X
Z Z
≤ ˜ |2ω (∆ϕ F )2 ωϕn +
e2A(F ) Q2p |∇F Q2p (trωϕ ω)2 |Tsp |ωϕ ωϕn
r 2
ϕ
X X
Z Z
≤ eA(F ) Q2p+1 (∆ϕ F )2 ωϕn + Q2p (trωϕ ω)3 |Tsp r 2 n
|ω ωϕ
X X
Z Z
≤C Q2p+1 (trω ωϕ )2n−2 ωϕn + C Q2p+1 ωϕn
ZX X

≤C Q2p+1 (trω ωϕ )2n−2 ωϕn ,

X
where we used Young’s inequality in the third line. Now, our bound on the torsion integral
is a constant multiple of the bound of the remaining terms as seen in Equation (3.20) of
[10]. The remainder of the proof for the L∞ bound on Q follows from Moser iteration and
several applications of the Hölder and Sobolev inequalities with respect to the reference
metric ω, see Section 3 of [10] and Section 4 of [9]. The constants and powers of the
trace differ slightly from those in the Kähler case, but will not affect the iteration method.
Lastly, to show an L1 bound on the quantity Q follows immediately from the fact that
|∂F |2ωϕ holds the same way as in (4.35) of [9] and the L1 bound on (trω ωϕ )B+1 follows
using the LB+1 norm we obtain from the Lp bound.
Finally, using that we have an upper and lower bound on trω ω̃ gives us the quasi-isometry
of ω and ω̃. The higher order estimates of ϕ can then be obtained using a bootstrapping
argument as in the proof of Proposition 1.2 in [9] where we use a result for the non-Kähler
setting by [25] to obtain the C 2,α estimate. 

We now prove the main result of the paper:

Theorem 3. Starting from an initial metric ω0 satisfying ∂ ∂ω ¯ k = 0 for k = 1, 2, we have

0
that a solution to the flow defined in Equation (1) exists as long as the Chern scalar cur-
vature of the evolving metric remains bounded. In addition, if the Chern scalar curvature
remains uniformly bounded for all time, then the flow converges smoothly to the unique
¯ class of ω0 .
Chern-Ricci-flat metric in the ∂ ∂-

Proof. Firstly, we know that along the flow, the Mabuchi energy of ωϕ with respect to ω0
is decreasing, in other words,
ωϕn 2
Z 
∂
Mabω0 (ωϕ ) = − Rϕ + 2 Re(gϕj k̄ (tr Tϕ )j ∂k̄ log F n ) ωϕn ≤ 0.
∂t X e ω0
In addition, the Mabuchi energy is bounded from below since
ωϕn n
Z Z
Mabω0 (ωϕ ) = log n ωϕ − F (ωϕn − ω0n )
X ω 0
Z X
≥ −C − sup |F | ωϕn − C
X X
≥ −C

12
where the first integral is bounded using the fact that the map x 7→ x log x has a lower
bound for x > 0. This gives us a uniform bound on the Mabuchi energy along the
flow. Combining the fact that we have short-time existence of the flow and that we have
uniform bounds on the potential ϕ along the flow as long as Rϕ stays bounded, as shown
in Theorem 2, we obtain long-time existence conditional on this Rϕ bound.
Let us now assume that the Rϕ remains uniformly bounded for all time t ∈ [0, ∞) so that
the flow exists for all time. Using that we have C ∞ estimates on ωϕ , we have that
∂
Z Z Z Z 1/2
ϕ̇2 ωϕn = 2ϕ̇ϕ̈ωϕn + ϕ̇2 ∆ϕ ϕ̇ωϕn ≤ C ϕ̇2 ωϕn .
∂t X X X X
R 
Let us denote f (t) = ϕ̇2ω (t). The above differential inequality is equivalent to
X ϕ

f′
f ′ ≤ Cf 1/2 ≤ C(f + δ)1/2 ⇒ = ((f + δ)1/2 )′ ≤ C,
(f + δ)1/2
for a constant δ > 0. For t′ > t, integrating the above equation from t to t′ gives us that
(f (t′ ) + δ)1/2 − (f (t) + δ)1/2 ≤ C(t′ − t).
Since this holds for every δ > 0, we may take δ → 0 and rearrange to obtain
f (t′ ) ≤ (f (t)1/2 + C(t′ − t))2 ,
where this tells us that f (t) can grow at most quadratically. Since we know that the
∂
Mabuchi energy is uniformly bounded and ∂t Mabω0 (ωϕ ) = −f (t), integration gives us
that
Z ∞
f (t)dt < ∞.
0

Let us assume for contradiction that f (t) does not converge to 0 for all t. Then, there
exists an ε ∈ (0, 1) and a sequence of times (ti )∞
i=1 → ∞ such that f (ti ) ≥ ε. For each
interval [ti − 1, ti ),
Z ti Z ti
f (t)dt ≥ (ε1/2 − C(ti − t))2 dt
ti −1 max(ti −1,{t : ε1/2 −C(ti −t)>0})
Zti
1/2
= 1/2
(ε − C(ti − t))2 dt
max(ti −1,ti − ε C )
1 3/2
ε − max(0, (ε1/2 − C)3 )


=
3C
1
= min(ε3/2 , 3Cε + 3Cε1/2 + C 3 )
3C
≥ Cε3/2 .
However, since there are infinitely many such ti where ti → ∞, the integral of f (t) cannot
be bounded, giving us the desired contradiction.
This implies that f (t) = X ϕ̇2 ωϕn → 0 for at t → ∞ which implies that ϕ̇(t) → 0 as
R
t → ∞. Since we have C ∞ estimates on ϕ, by the Arzela-Ascoli theorem, there exists a
sequence of times (tj )∞ ∞
j=1 such that ϕ(tj ) → ϕ∞ in C , where ϕ∞ is smooth. In fact,
√
2 ¯ ∞ is
R
since X ϕ̇∞ ωϕ∞ = 0, ϕ̇∞ = 0 which by Lemma 4 implies that ω∞ = ω0 + −1∂ ∂ϕ
¯
Chern-Ricci-flat. By the uniqueness of Chern-Ricci-flat metrics in the ∂ ∂ class of a metric
[27], ω∞ is the unique Chern-Ricci-flat metric along the flow. We can show that we indeed

13
have C ∞ convergence without passing to a sequence. To see this, suppose not. Then there
exists a k ∈ N, ε > 0 and a sequence of times (tn )∞
n=1 such that for all n,
||ϕ(tn ) − ϕ∞ ||C k (X) ≥ ε.
Since we have C k+1 bounds on ϕ(tn ), the Arzela-Ascoli theorem gives us that there exists
a subsequence (tnj )∞ k ′
j=1 such that ϕ(tnj ) converges in C to a limit ϕ∞ , with

||ϕ′∞ − ϕ∞ ||C k (X) ≥ ε.

This implies that ωϕ′ 6= ωϕ , but by the above argument ωϕ′ is also Chern-Ricci-flat, and so
¯
this contradicts the uniqueness of the Chern-Ricci-flat metric in the ∂ ∂−class of ω0 . This
completes the proof of the theorem.


Acknowledgements
The author is very grateful to thank her thesis advisor Ben Weinkove for his helpful
suggestions and his continued support and encouragement. She would also like to thank
Gregory Edwards, Antoine Song and Jonathan Zhu for some useful discussions.

References
1. D. Angella, S. Calamai, and C. Spotti, On the Chern-Yamabe problem, Math. Res. Lett. 24 (2017),
no. 3, 645–677.
2. L. Bedulli and L. Vezzoni, A parabolic flow of balanced metrics, J. Reine Angew. Math. 723 (2017),
79–99.
3. R. J. Berman, T. Darvas, and C. H. Lu, Convexity of the extended K-energy and the large time behavior
of the weak Calabi flow, Geometry & Topology 21 (2017), 2945–2988.
4. R. Bryant and F. Xu, Laplacian flow for closed G2-structures: short time behavior, arXiv:1101.2004.
5. E. Calabi, Extremal Kähler metrics. In Seminar on Differential Geometry, Ann. of Math. Stud. Prince-
ton Univ. Press, Princeton, NJ. 102 (1982), 259–290.
6. , Extremal Kähler Metrics II. In Differential Geometry and Complex Analysis, pp. 95–114,
Springer-Verlag, New York, 1985.
7. S.-C. Chang, The 2-dimensional Calabi flow, Nagoya Math. J. 181 (2006), 63–73.
8. X. Chen and W. Y He, On the Calabi flow, Amer. J. Math. 130 (2008), 539–570.
9. X. X. Chen and J. Cheng, On the constant scalar curvature Kähler metrics, apriori estimates,
arXiv:1712.06697 (2017).
10. , On the constant scalar curvature Kähler metrics, existence results, arXiv:1801.00656 (2018).
11. X.X. Chen and S. Sun, Calabi flow, Geodesic rays, and uniqueness of constant scalar curvature Kähler
metrics, Annals of Mathematics 180 (2014), 407–454.
12. P. T. Chruściel, Semi-global existence and convergence of solutions of the Robinson-Trautman (2-
dimensional Calabi) equation, Comm. Math. Phys. 137 (1991), 289–313.
13. R. Feng and H. Huang, The global existence and convergence of the Calabi flow on Cn /Zn + iXn , J,
Funct. Anal. 263 (2012), 1129–1146.
14. J. Fine, Calabi flow and projective embeddings, J. Differential Geom. 84 (2010), no. 3, 489–523.
15. M. Gill, Convergence of the parabolic complex Monge-Ampère equation on compact Hermitian mani-
folds, Comm. Anal. Geom. 19 (2011), no. 1, 65–78.
16. W. He, On the convergence of the Calabi flow, Proc. Amer. Math. Soc. 143 (2015), 1273–1281.
17. H. Huang and K. Zheng, Stability of the Calabi flow near an extremal metric, Ann. Sc. Norm. Super.
Pisa Cl. Sci. 11 (2012), 167–175.
18. H. Li, B. Wang, and K. Zheng, Regularity scales and convergence of the Calabi flow, J. Geom. Anal.
28 (2018), 2050–2101.
19. D. H. Phong, S. Picard, and X. Zhang, New curvature flows in complex geometry, arXiv:1806.11235
(2018).
20. S. Picard, Calabi-Yau Manifolds with Torsion and Geometric Flows, CIME Summer School,
http://people.math.harvard.edu/ spicard/cetraro.pdf (2018).
21. X. S. Shen, Estimates for metrics of constant Chern scalar curvature, arXiv preprint arXiv: 1909.13445
(2019).

14
22. J. Streets and G. Tian, Hermitian curvature flow, J. Eur. Math. Soc. 13 (2011), 601–634.
23. G. Székelyhidi, Remark on the Calabi flow with bounded curvature, Univ. Iagel. Acta Math. 50 (2013),
107–115.
24. G. Tian, Canonical metrics in Kähler geometry, Notes taken by Meike Akveld. Lectures in Mathematics
ETH Zürich, Birkhäuser Verlag, Basel, 2000.
25. V. Tosatti, Y. Wang, B. Weinkove, and X. Yang, C 2,α estimates for nonlinear elliptic equations in
complex and almost complex geometry, Calc. Var. Partial Differential Equations 54 (2015), no. 1,
431–453.
26. V. Tosatti and B. Weinkove, The Calabi flow with small initial energy, Math. Res. Lett. 14 (2007),
1033–1039.
27. , The complex Monge-Ampère equation on compact Hermitian manifolds, J. Amer. Math. Soc.
23 (2010), 1187–1195.
28. , The Chern-Ricci flow on complex surfaces, Comp. Math. 149 (2013), 2101–2138.
29. , On the evolution of a Hermitian metric by its Chern-Ricci form, J. Differential Geom. 99
(2015), no. 1, 125–163.
30. Y. Ustinovskiy, Hermitian curvature flow on manifolds with non-negative Griffiths curature, Amer. J.
Math. 141 (2019), no. 6, 1751–1775.

Department of Mathematics, Northwestern University, Evanston, IL USA 60201

Email address, X. S. Shen: xss@math.northwestern.edu

15