THE EVOLUTION OF THE WEYL TENSOR UNDER THE RICCI FLOW

GIOVANNI CATINO AND CARLO MANTEGAZZA

A BSTRACT. We compute the evolution equation of the Weyl tensor under the Ricci flow of a Rie-
mannian manifold and we discuss some consequences for the classification of locally conformally
flat Ricci solitons.

C ONTENTS
1. The Evolution Equation of the Weyl Tensor 1
2. Locally Conformally Flat Ricci Solitons 12
2.1. Compact LCF Ricci Solitons 14
2.2. LCF Ricci Solitons with Constant Scalar Curvature 14
2.3. Gradient LCF Ricci Solitons with Nonnegative Ricci Tensor 15
2.4. The Classification of Steady and Shrinking Gradient LCF Ricci Solitons 17
3. Singularities of Ricci Flow with Bounded Weyl Tensor 19
References 20

1. T HE E VOLUTION E QUATION OF THE W EYL T ENSOR

The Riemann curvature operator of a Riemannian manifold (M n , g) is defined as in [14] by
Riem(X, Y )Z = ∇Y ∇X Z − ∇X ∇Y Z + ∇[X,Y ] Z .
In a local coordinate system
∂ the components of the (3, 1)–Riemann curvature tensor are given by
∂ ∂ ∂
Rlijk ∂x l = Riem ∂xi , ∂xj ∂x k and we denote by R ijkl = glm R m
ijk its (4, 0)–version.

In all the paper the Einstein convention of summing over the repeated indices will be adopted.
With this choice, for the sphere Sn we have Riem(v, w, v, w) = Rijkl v i wj v k wl > 0.
The Ricci tensor is obtained by the contraction Rik = g jl Rijkl and R = g ik Rik will denote the
scalar curvature.
The so called Weyl tensor is then defined by the following decomposition formula (see [14,
Chapter 3, Section K]) in dimension n ≥ 3,
R 1
Wijkl = Rijkl + (gik gjl − gil gjk ) − (Rik gjl − Ril gjk + Rjl gik − Rjk gil )
(n − 1)(n − 2) n−2
= Rijkl + Aijkl + Bijkl ,
where we introduced the tensors
R
Aijkl = (gik gjl − gil gjk )
(n − 1)(n − 2)
and
1
Bijkl = − (Rik gjl − Ril gjk + Rjl gik − Rjk gil ) .
n−2
The Weyl tensor satisfies all the symmetries of the curvature tensor and all its traces with the
metric are zero, as it can be easily seen by the above formula.
In dimension three W is identically zero for every Riemannian manifold (M 3 , g), it becomes

Date: October 1, 2013.

1
2 GIOVANNI CATINO AND CARLO MANTEGAZZA

relevant instead when n ≥ 4 since its nullity is a condition equivalent for (M n , g) to be locally
conformally flat, that is, around every point p ∈ M n there is a conformal deformation geij = ef gij
of the original metric g, such that the new metric is flat, namely, the Riemann tensor associated
to ge is zero in Up (here f : Up → R is a smooth function defined in a open neighborhood Up of p).
We suppose now that (M n , g(t)) is a Ricci flow in some time interval, that is, the time–dependent
metric g(t) satisfies
∂
gij = −2Rij .
∂t
We have then the following evolution equations for the curvature (see for instance [15]),
∂
R = ∆R + 2|Ric|2
∂t
∂
Rij = ∆Rij + 2Rkl Rkilj − 2g pq Rip Rjq ,
∂t
∂
(1.1) Rijkl = ∆Rijkl + 2(Cijkl − Cijlk + Cikjl − Ciljk )
∂t
− g pq (Rip Rqjkl + Rjp Riqkl + Rkp Rijql + Rlp Rijkq ) ,

where Cijkl = g pq g rs Rpijr Rslkq .

All the computations which follow will be done in a fixed local frame, not in a moving frame.

The goal of this section is to work out the evolution equation under the Ricci flow of the Weyl
tensor Wijkl . In the next sections we will see the geometric consequences of the assumption that
a manifold evolving by the Ricci flow is locally conformally flat at every time. In particular, we
will be able to classify the so called Ricci solitons under the hypothesis of locally conformally
flatness.

Since Wijkl = Rijkl + Aijkl + Bijkl and we already have the evolution equation (1.1) for Rijkl ,
we start differentiating in time the tensors Aijkl and Bijkl

∂ ∆R + 2|Ric|2
Aijkl = (gik gjl − gil gjk )
∂t (n − 1)(n − 2)
R
+ (−2Rik gjl − 2Rjl gik + 2Ril gjk + 2Rjk gil )
(n − 1)(n − 2)
2|Ric|2 2R
= ∆Aijkl + (gik gjl − gil gjk ) + Bijkl
(n − 1)(n − 2) n−1
and
∂ 1 
Bijkl = − (∆Rik + 2Rpq Rpiqk − 2g pq Rip Rkq )gjl
∂t n−2
− (∆Ril + 2Rpq Rpiql − 2g pq Rip Rlq )gjk
+ (∆Rjl + 2Rpq Rpjql − 2g pq Rjp Rlq )gik
− (∆Rjk + 2Rpq Rpjqk − 2g pq Rjp Rkq )gil

+ 4Rjk Ril − 4Rik Rjl
2  pq
= ∆Bijkl − (R Rpiqk − g pq Rip Rkq )gjl − (Rpq Rpiql − g pq Rip Rlq )gjk
n−2

+ (Rpq Rpjql − g pq Rjp Rlq )gik − (Rpq Rpjqk − g pq Rjp Rkq )gil
4
+ (Rik Rjl − Rjk Ril ) .
n−2
 THE EVOLUTION OF THE WEYL TENSOR UNDER THE RICCI FLOW 3

Now we deal with the terms like Rpq Rpiqk .

We have by definition Rpq Rpiqk = Rpq Wpiqk − Rpq Apiqk − Rpq Bpiqk and

R
Rpq Apiqk = (Rpq gpq gik − Rpq gpk giq )
(n − 1)(n − 2)
R
= (Rgik − Rik ) ,
(n − 1)(n − 2)
1
Rpq Bpiqk =− (Rpq Rpq gik − Rpq Rpk giq + Rpq Rik gpq − Rpq Riq gpk )
n−2
1
=− (|Ric|2 gik + RRik − 2g pq Rip Rkq ) ,
n−2

hence, we get

R
Rpq Rpiqk = Rpq Wpiqk − (Rgik − Rik )
(n − 1)(n − 2)
1
+ (|Ric|2 gik + RRik − 2g pq Rip Rkq )
n−2
1 R
= Rpq Wpiqk + (|Ric|2 gik − 2g pq Rip Rkq ) + (nRik − Rgik ) .
n−2 (n − 1)(n − 2)
4 GIOVANNI CATINO AND CARLO MANTEGAZZA

∂
Substituting these terms in the formula for ∂t Bijkl we obtain

∂ 2
Bijkl = ∆Bijkl − (Rpq Wpiqk gjl − Rpq Wpiql gjk + Rpq Wpjql gik − Rpq Wpjqk gil )
∂t n−2
2|Ric|2
− (gik gjl − gil gjk + gjl gik − gjk gil )
(n − 2)2
4
+ (g pq Rip Rkq gjl − g pq Rip Rlq gjk + g pq Rjp Rlq gik − g pq Rjp Rkq gil )
(n − 2)2
2nR
− (Rik gjl − Ril gjk + Rjl gik − Rjk gil )
(n − 1)(n − 2)2
2R2
+ (gik gjl − gil gjk + gjl gik − gjk gil )
(n − 1)(n − 2)2
2
+ (g pq Rip Rkq gjl − g pq Rip Rlq gjk + g pq Rjp Rlq gik − g pq Rjp Rkq gil )
n−2
4
+ (Rik Rjl − Rjk Ril )
n−2
2
= ∆Bijkl − (Rpq Wpiqk gjl − Rpq Wpiql gjk + Rpq Wpjql gik − Rpq Wpjqk gil )
n−2
2n
+ (g pq Rip Rkq gjl − g pq Rip Rlq gjk + g pq Rjp Rlq gik − g pq Rjp Rkq gil )
(n − 2)2
2nR
− (Rik gjl − Ril gjk + Rjl gik − Rjk gil )
(n − 1)(n − 2)2
2R2 − 2(n − 1)|Ric|2
+ (gik gjl − gil gjk + gjl gik − gjk gil )
(n − 1)(n − 2)2
4
+ (Rik Rjl − Rjk Ril )
n−2
2
= ∆Bijkl − (Rpq Wpiqk gjl − Rpq Wpiql gjk + Rpq Wpjql gik − Rpq Wpjqk gil )
n−2
2n
+ (g pq Rip Rkq gjl − g pq Rip Rlq gjk + g pq Rjp Rlq gik − g pq Rjp Rkq gil )
(n − 2)2
2nR 4R 4|Ric|2
+ Bijkl + Aijkl − (gik gjl − gil gjk )
(n − 1)(n − 2) n−2 (n − 2)2
4
+ (Rik Rjl − Rjk Ril ) .
n−2
 THE EVOLUTION OF THE WEYL TENSOR UNDER THE RICCI FLOW 5

Hence,
(1.2)
∂  ∂ 
− ∆ Wijkl = − ∆ (Rijkl + Aijkl + Bijkl )
∂t ∂t
= 2(Cijkl − Cijlk + Cikjl − Ciljk )
− g pq (Rip Rqjkl + Rjp Riqkl + Rkp Rijql + Rlp Rijkq )
2|Ric|2 2R
+ (gik gjl − gil gjk ) + Bijkl
(n − 1)(n − 2) n−1
2
− (Rpq Wpiqk gjl − Rpq Wpiql gjk + Rpq Wpjql gik − Rpq Wpjqk gil )
n−2
2n
+ (g pq Rip Rkq gjl − g pq Rip Rlq gjk + g pq Rjp Rlq gik − g pq Rjp Rkq gil )
(n − 2)2
2nR 4R 4|Ric|2
+ Bijkl + Aijkl − (gik gjl − gil gjk )
(n − 1)(n − 2) n−2 (n − 2)2
4
+ (Rik Rjl − Rjk Ril )
n−2
= 2(Cijkl − Cijlk + Cikjl − Ciljk )
− g pq (Rip Rqjkl + Rjp Riqkl + Rkp Rijql + Rlp Rijkq )
2
− (Rpq Wpiqk gjl − Rpq Wpiql gjk + Rpq Wpjql gik − Rpq Wpjqk gil )
n−2
2n
+ (g pq Rip Rkq gjl − g pq Rip Rlq gjk + g pq Rjp Rlq gik − g pq Rjp Rkq gil )
(n − 2)2
4R
− (Rik gjl − Ril gjk + Rjl gik − Rjk gil )
(n − 2)2
4R2 − 2n|Ric|2 4
+ (gik gjl − gil gjk ) + (Rik Rjl − Rjk Ril ) .
(n − 1)(n − 2)2 n−2
Now, in order to simplify the formulas, we assume to be in an orthonormal basis, then Cijkl =
Rpijq Rqlkp and we have
Cijkl = Rpijq Rqlkp
= Wpijq Wqlkp + Apijq Aqlkp + Bpijq Bqlkp + Apijq Bqlkp + Bpijq Aqlkp
− Wpijq Aqlkp − Wpijq Bqlkp − Apijq Wqlkp − Bpijq Wqlkp .
Substituting the expressions for the tensors A and B in the above terms and simplifying, we
obtain the following identities.
R2
Apijq Aqlkp = (gik gjl + (n − 2)gij glk ) ,
(n − 1)2 (n − 2)2
1
Bpijq Bqlkp = (Rpj giq + Riq gpj − Rpq gij − Rij gpq )(Rqk glp + Rlp gqk − Rpq glk − Rlk gpq )
(n − 2)2
1
= 2Rik Rlj + (n − 4)Rij Rlk + Rpj Rpl gik + Rpk Rpi glj − 2Rpj Rpi glk − 2Rpl Rpk gij
(n − 2)2
+ RRij glk + RRlk gij + |Ric|2 gij glk ,

R

Apijq Bqlkp = − 2
Rik glj + Rlj gik − Rij glk + (n − 3)Rlk gij + Rgij glk ,
(n − 1)(n − 2)
R

Bpijq Aqlkp =− Rlj gik + Rik glj − Rlk gij + (n − 3)Rij glk + Rgij glk ,
(n − 1)(n − 2)2
R
Wpijq Aqlkp = Wlijk ,
(n − 1)(n − 2)
6 GIOVANNI CATINO AND CARLO MANTEGAZZA

R
Apijq Wqlkp = Wilkj ,
(n − 1)(n − 2)

1
Wpijq Bqlkp = − (Wlijp Rpk + Wpijk Rlp − Wpijq Rpq glk ) ,
n−2

1
Bpijq Wqlkp = − (Wilkp Rpj + Wplkj Rpi − Wqlkp Rpq gij )
n−2

where in these last four computations we used the fact that every trace of the Weyl tensor is null.
Interchanging the indexes and summing we get

Apijq Aqlkp − Apijq Aqklp + Apikq Aqljp − Apilq Aqkjp

R2 
= gik gjl + (n − 2)gij glk − gil gjk − (n − 2)gij glk
(n − 1)2 (n − 2)2

+ gij gkl + (n − 2)gik glj − gij gkl − (n − 2)gil gjk
R2
= (gik gjl − gil gjk ) ,
(n − 1)(n − 2)2

Bpijq Bqlkp − Bpijq Bqklp + Bpikq Bqljp − Bpilq Bqkjp

1 
= 2Rik Rlj + (n − 4)Rij Rlk + Rpj Rpl gik + Rpk Rpi glj
(n − 2)2
− 2Rpj Rpi glk − 2Rpl Rpk gij + RRij glk + RRlk gij + |Ric|2 gij glk

− 2Ril Rkj − (n − 4)Rij Rlk − Rpj Rpk gil − Rpl Rpi gkj

+ 2Rpj Rpi glk + 2Rpk Rpl gij − RRij glk − RRlk gij − |Ric|2 gij glk

+ 2Rij Rlk + (n − 4)Rik Rlj + Rpk Rpl gij + Rpj Rpi glk

− 2Rpk Rpi glj − 2Rpl Rpj gik + RRik glj + RRlj gik + |Ric|2 gik glj

− 2Rij Rkl − (n − 4)Ril Rjk − Rpl Rpk gij − Rpj Rpi gkl

+ 2Rpl Rpi gjk + 2Rpk Rpj gil − RRil gjk − RRjk gil − |Ric|2 gil gjk
1 
= (n − 2)(Rik Rlj − Ril Rjk )
(n − 2)2
− Rpj Rpl gik − Rpk Rpi glj + Rpl Rpi gjk + Rpk Rpj gil

+ R(Rik glj + Rlj gik − Ril gjk − Rjk gil )


+ |Ric|2 (gik glj − gil gjk ) ,
 THE EVOLUTION OF THE WEYL TENSOR UNDER THE RICCI FLOW 7

Apijq Bqlkp + Bpijq Aqlkp −Apijq Bqklp − Bpijq Aqklp

+ Apikq Bqljp + Bpikq Aqljp − Apilq Bqkjp − Bpilq Aqkjp
R 
=− Rik glj + Rlj gik − Rij glk + (n − 3)Rlk gij + Rgij glk
(n − 1)(n − 2)2
+ Rlj gik + Rik glj − Rlk gij + (n − 3)Rij glk + Rglk gij
− Ril gkj − Rjk gil + Rij gkl − (n − 3)Rkl gij − Rgij gkl
− Rkj gil − Ril gkj + Rkl gij − (n − 3)Rij gkl − Rgkl gij
+ Rij glk + Rlk gij − Rik glj + (n − 3)Rlj gik + Rgik glj
+ Rlk gij + Rij glk − Rlj gik + (n − 3)Rik glj + Rglj gik
− Rij gkl − Rlk gij + Ril gkj − (n − 3)Rkj gil − Rgil gkj

− Rkl gij − Rij gkl + Rkj gil − (n − 3)Ril gkj − Rgkj gil
R  
=− Rik gjl + Rjl gik − Rjk gil − Ril gjk
(n − 1)(n − 2)
2R2
− (gik gjl − gil gjk )
(n − 1)(n − 2)2

and

Wpijq Aqlkp − Wpijq Aqklp + Wpikq Aqljp − Wpilq Aqkjp

R
= (Wlijk − Wkijl + Wlikj − Wkilj ) = 0 ,
(n − 1)(n − 2)

since the Weyl tensor, sharing the same symmetries of the Riemann tensor, is skew–symmetric in
the third–fourth indexes.
The same result holds for the other sum as
R R
Apijq Wqlkp = Wilkj = Wlijk = Wpijq Aqlkp
(n − 1)(n − 2) (n − 1)(n − 2)

hence,
Apijq Wqlkp − Apijq Wqklp + Apikq Wqljp − Apilq Wqkjp = 0 .

Finally, for the remaining two terms we have

−Wpijq Bqlkp − Bpijq Wqlkp + Wpijq Bqklp + Bpijq Wqklp

− Wpikq Bqljp − Bpikq Wqljp + Wpilq Bqkjp + Bpilq Wqkjp
1 
= Wlijp Rpk + Wpijk Rlp − Wpijq Rpq glk
n−2
+ Wilkp Rpj + Wplkj Rpi − Wqlkp Rpq gij
− Wkijp Rpl − Wpijl Rkp + Wpijq Rpq gkl
− Wiklp Rpj − Wpklj Rpi + Wqklp Rpq gij
+ Wlikp Rpj + Wpikj Rlp − Wpikq Rpq gjl
+ Wiljp Rpk + Wpljk Rpi − Wqljp Rpq gik
− Wkilp Rpj − Wpilj Rkp + Wpilq Rpq gkj

− Wikjp Rpl − Wpkjl Rpi + Wqkjp Rpq gil
1  
= Wpilq Rpq gkj + Wqkjp Rpq gil − Wpikq Rpq gjl − Wqljp Rpq gik
n−2
8 GIOVANNI CATINO AND CARLO MANTEGAZZA

where we used repeatedly the symmetries of the Weyl and the Ricci tensors.
Hence, summing all these terms we conclude

(1.3) 2(Cijkl −Cijlk + Cikjl − Ciljk ) = 2(Dijkl − Dijlk + Dikjl − Diljk )

2R2
+ (gik gjl − gil gjk )
(n − 1)(n − 2)2
2
+ (Rik Rlj − Ril Rjk )
n−2
2
+ (−Rpj Rpl gik − Rpk Rpi glj + Rpl Rpi gjk + Rpk Rpj gil )
(n − 2)2
2R 2|Ric|2
+ (R ik glj + R lj gik − R il gjk − R jk gil ) + (gik glj − gil gjk )
(n − 2)2 (n − 2)2
2R
− (Rik gjl + Rjl gik − Rjk gil − Ril gjk )
(n − 1)(n − 2)
4R2
− (gik gjl − gil gjk )
(n − 1)(n − 2)2
2
+ (Wpilq Rpq gkj + Wqkjp Rpq gil − Wpikq Rpq gjl − Wqljp Rpq gik )
n−2
= 2(Dijkl − Dijlk + Dikjl − Diljk )
2(n − 1)|Ric|2 − 2R2
+ (gik gjl − gil gjk )
(n − 1)(n − 2)2
2
+ (Rik Rlj − Ril Rjk )
n−2
2
− (Rpj Rpl gik + Rpk Rpi glj − Rpl Rpi gjk − Rpk Rpj gil )
(n − 2)2
2R
+ (Rik gjl + Rjl gik − Rjk gil − Ril gjk )
(n − 1)(n − 2)2
2
+ (Wpilq Rpq gkj + Wqkjp Rpq gil − Wpikq Rpq gjl − Wqljp Rpq gik ) ,
n−2

where Dijkl = Wpijq Wqlkp .

 THE EVOLUTION OF THE WEYL TENSOR UNDER THE RICCI FLOW 9

Then we deal with the following term appearing in equation (1.2),

Rip Rpjkl + Rjp Ripkl + Rkp Rijpl + Rlp Rijkp

= Rip Wpjkl + Rjp Wipkl + Rkp Wijpl + Rlp Wijkp
R  
− Rip (gpk gjl − gpl gjk ) + Rjp (gik gpl − gil gpk )
(n − 1)(n − 2)
R  
− Rkp (gip gjl − gil gjp ) + Rlp (gik gjp − gip gjk )
(n − 1)(n − 2)
1
+ (Rip (Rpk gjl − Rpl gjk + Rjl gpk − Rjk gpl ))
n−2
1
+ (Rjp (Rik gpl − Ril gpk + Rpl gik − Rpk gil ))
n−2
1
+ (Rkp (Rip gjl − Ril gjp + Rjl gip − Rjp gil ))
n−2
1
+ (Rlp (Rik gjp − Rip gjk + Rjp gik − Rjk gip ))
n−2
= Rip Wpjkl + Rjp Wipkl + Rkp Wijpl + Rlp Wijkp
1
+ (Rip Rpk gjl − Rip Rpl gjk + Rjl Rik − Ril Rjk )
n−2
1
+ (Rjl Rik − Rjk Ril + Rjp Rpl gik − Rjp Rpk gil )
n−2
1
+ (Rkp Rip gjl − Rjk Ril + Rik Rjl − Rkp Rjp gil )
n−2
1
+ (Rjl Rik − Rlp Rip gjk + Rlp Rjp gik − Ril Rjk )
n−2
2R
− (Rik gjl − Ril gjk + Rjl gik − Rjk gil )
(n − 1)(n − 2)
= Rip Wpjkl + Rjp Wipkl + Rkp Wijpl + Rlp Wijkp
2
+ (Rip Rkp gjl − Rip Rlp gjk + Rjp Rlp gik − Rjp Rkp gil )
n−2
4
+ (Rik Rjl − Rjk Ril )
n−2
2R
− (Rik gjl − Ril gjk + Rjl gik − Rjk gil ) .
(n − 1)(n − 2)
10 GIOVANNI CATINO AND CARLO MANTEGAZZA

Inserting expression (1.3) and this last quantity in equation (1.2) we obtain

∂ 
− ∆ Wijkl = 2(Dijkl − Dijlk + Dikjl − Diljk )
∂t
2(n − 1)|Ric|2 − 2R2
+ (gik gjl − gil gjk )
(n − 1)(n − 2)2
2
+ (Rik Rlj − Ril Rjk )
n−2
2
− (Rpj Rpl gik + Rpk Rpi glj − Rpl Rpi gjk − Rpk Rpj gil )
(n − 2)2
2R
+ (Rik gjl + Rjl gik − Rjk gil − Ril gjk )
(n − 1)(n − 2)2
2
+ (Wpilq Rpq gkj + Wqkjp Rpq gil − Wpikq Rpq gjl − Wqljp Rpq gik )
n−2
− Rip Wpjkl − Rjp Wipkl − Rkp Wijpl − Rlp Wijkp
2
− (Rip Rkp gjl − Rip Rlp gjk + Rjp Rlp gik − Rjp Rkp gil )
n−2
4
− (Rik Rjl − Rjk Ril )
n−2
2R
+ (Rik gjl − Ril gjk + Rjl gik − Rjk gil )
(n − 1)(n − 2)
2
− (Rpq Wpiqk gjl − Rpq Wpiql gjk + Rpq Wpjql gik − Rpq Wpjqk gil )
n−2
2n
+ (Rip Rkp gjl − Rip Rlp gjk + Rjp Rlp gik − Rjp Rkp gil )
(n − 2)2
4R
− (Rik gjl − Ril gjk + Rjl gik − Rjk gil )
(n − 2)2
4R2 − 2n|Ric|2 4
+ 2
(gik gjl − gil gjk ) + (Rik Rjl − Rjk Ril )
(n − 1)(n − 2) n−2
= 2(Dijkl − Dijlk + Dikjl − Diljk )
− (Rip Wpjkl + Rjp Wipkl + Rkp Wijpl + Rlp Wijkp )
2(R2 − |Ric|2 )
+ (gik gjl − gil gjk )
(n − 1)(n − 2)2
2
+ (Rik Rlj − Ril Rjk )
n−2
2
+ (Rpj Rpl gik + Rpk Rpi glj − Rpl Rpi gjk − Rpk Rpj gil )
(n − 2)2
2R
− (Rik gjl + Rjl gik − Rjk gil − Ril gjk ) .
(n − 2)2

Hence, we resume this long computation in the following proposition, getting back to a standard
coordinate basis.
 THE EVOLUTION OF THE WEYL TENSOR UNDER THE RICCI FLOW 11

Proposition 1.1. During the Ricci flow of an n–dimensional Riemannian manifold (M n , g), the Weyl
tensor satisfies the following evolution equation
∂ 
− ∆ Wijkl = 2 (Dijkl − Dijlk + Dikjl − Diljk )
∂t
− g pq (Rip Wqjkl + Rjp Wiqkl + Rkp Wijql + Rlp Wijkq )
2
+ g pq (Rip Rqk gjl − Rip Rql gjk + Rjp Rql gik − Rjp Rqk gil )
(n − 2)2
2R
− (Rik gjl − Ril gjk + Rjl gik − Rjk gil )
(n − 2)2
2 2(R2 − |Ric|2 )
+ (Rik Rjl − Rjk Ril ) + (gik gjl − gil gjk ) ,
n−2 (n − 1)(n − 2)2
where Dijkl = g pq g rs Wpijr Wslkq .
From this formula we immediately get the following rigidity result on the eigenvalues of the
Ricci tensor.
Corollary 1.2. Suppose that under the Ricci flow of (M n , g) of dimension n ≥ 4, the Weyl tensor remains
identically zero. Then, at every point, either the Ricci tensor is proportional to the metric or it has an
eigenvalue of multiplicity (n − 1) and another of multiplicity 1.
Proof. By the above proposition, as every term containing the Weyl tensor is zero, the following
relation holds at every point in space and time
2
0= g pq (Rip Rqk gjl − Rip Rql gjk + Rjp Rql gik − Rjp Rqk gil )
(n − 2)2
2R2 2|Ric|2
+ (g ik gjl − gil gjk ) − (gik gjl − gil gjk )
(n − 1)(n − 2)2 (n − 1)(n − 2)2
2R 2
− (Rik gjl − Ril gjk + Rjl gik − Rjk gil ) + (Rik Rjl − Rjk Ril ) .
(n − 2)2 n−2
In normal coordinates such that the Ricci tensor is diagonal we get, for every couple of different
eigenvectors vi with relative eigenvalues λi ,
(1.4) (n − 1)[λ2i + λ2j ] − (n − 1)R(λi + λj ) + (n − 1)(n − 2)λi λj + R2 − |Ric|2 = 0 .
As n ≥ 4, fixing i, then the equation above is a second order polynomial in λj , hence it can only
have at most 2 solutions, hence, we can conclude that there are at most three possible values for
the eigenvalues of the Ricci tensor.
Since the dimension is at least four, at least one eigenvalues must have multiplicity two, let us
say λi , hence the equation (1.4) holds also for i = j, and it remains at most only one possible value
for the other eigenvalues λl with l 6= i. In conclusion, either the eigenvalues are all equal or they
divide in only two possible values, λ with multiplicity larger than one, say k and µ 6= λ. Suppose
that µ also has multiplicity larger than one, that is, k < n − 1, then we have
|Ric|2 − R2
(1.5) nλ2 − 2Rλ =
n−1
2 |Ric|2 − R2
nµ − 2Rµ =
n−1
taking the difference and dividing by (λ − µ) we get
n(λ + µ) = 2R = 2[kλ + (n − k)µ]
then,
(n − 2k)λ = (n − 2k)µ
hence, n = 2k, but then getting back to equation (1.5), R = n(µ + λ)/2 and
n(λ2 + µ2 )/2 − n2 (µ2 + λ2 + 2λµ)/4
nλ2 − n(µ + λ)λ =
n−1
12 GIOVANNI CATINO AND CARLO MANTEGAZZA

which implies

n(n − 2) 2 2n2
−4nλµ = − (λ + µ2 ) − µλ
n−1 n−1

that is, after some computation,

2n(n − 2) n(n − 2) 2
µλ = (λ + µ2 ) ,
n−1 n−1

which implies λ = µ.
At the end we conclude that at every point of M n , either Ric = λg or there is an eigenvalue λ
of multiplicity (n − 1) and another µ of multiplicity 1. 

Remark 1.3. Notice that in dimension three equation (1.4) becomes

2[λ2i + λ2j ]− 2R(λi + λj ) + 2λi λj + R2 − |Ric|2

= 2(λi + λj )2 − 2R(λi + λj ) − 2λi λj + R2 − |Ric|2
= − 2λl (λi + λj ) − 2λi λj + R2 − |Ric|2
=0,

where λi , λj and λl are the three eigenvalues of the Ricci tensor.

Hence, the condition is void and our argument does not work. This is clearly not unexpected as
the Weyl tensor is identically zero for every three–dimensional Riemannian manifold.

2. L OCALLY C ONFORMALLY F LAT R ICCI S OLITONS

n
Let (M , g), for n ≥ 4, be a connected, complete, Ricci soliton, that is, there exists a smooth
1–form ω and a constant α ∈ R such that

1 α
Rij + (∇i ωj + ∇j ωi ) = gij .
2 n

If α > 0 we say that the soliton is shrinking, if α = 0 steady, if α < 0 expanding.

If there exists a smooth function f : M n → R such that df = ω we say that the soliton is a gradient
Ricci soliton and f its potential function, then we have

α
Rij + ∇2ij f = gij .
n

If the metric dual field to the form ω is complete, then a Ricci soliton generates a self–similar solution to
the Ricci flow (if the soliton is a gradient soliton this condition is automatically satisfied [33]).
In all this section we will assume to be in this case.
In this section we discuss the classification of Ricci solitons (M n , g), for n ≥ 4, which are locally
conformally flat (LCF). As a consequence of Corollary 1.2 we have the following fact.

Proposition 2.1. Let (M n , g) be a complete, LCF Ricci soliton of dimension n ≥ 4. Then, at every point,
either the Ricci tensor is proportional to the metric or it has an eigenvalue of multiplicity (n − 1) and
another of multiplicity 1.
 THE EVOLUTION OF THE WEYL TENSOR UNDER THE RICCI FLOW 13

If a manifold (M n , g) is LCF, it follows that

0 = ∇l Wijkl
 R 1 
= ∇l Rijkl + (gik gjl − gil gjk ) − (Rik gjl − Ril gjk + Rjl gik − Rjk gil )
(n − 1)(n − 2) n−2
∇j R ∇i R
= − ∇i Rjk + ∇j Rik + gik − gjk
(n − 1)(n − 2) (n − 1)(n − 2)
1
− (∇j Rik − ∇l Ril gjk + ∇l Rjl gik − ∇i Rjk gil )
n−2
n−3 ∇j R ∇i R
=− (∇i Rjk − ∇j Rik ) + gik − gjk
n−2 (n − 1)(n − 2) (n − 1)(n − 2)
1
+ (∇i Rgjk /2 − ∇j Rgik /2)
2(n − 2)
n − 3h (∇i Rgjk − ∇j Rgik ) i
=− ∇i Rjk + ∇j Rik −
n−2 2(n − 1)
n − 3h  1   1 i
= ∇j Rik − Rgik − ∇i Rjk − Rgjk ,
n−2 2(n − 1) 2(n − 1)
where we used the second Bianchi identity and Schur’s Lemma ∇R = 2 div Ric.
Hence, since we assumed that the dimension n is at least four, the Schouten tensor defined by
1
S = Ric − 2(n−1) Rg satisfies the equation

(∇X S) Y = (∇Y S) X, X, Y ∈ T M .
Any symmetric two tensor satisfying this condition is called a Codazzi tensor (see [2, Chapter 16]
for a general overview of Codazzi tensors).
Suppose that we have a local orthonormal frame {E1 , . . . , En } in an open subset Ω of M n such
that Ric(E1 ) = λE1 and Ric(Ei ) = µEi for i = 2, . . . , n and λ 6= µ. For every point in Ω also
the Schouten tensor S has two distinct eigenvalues σ1 of multiplicity one and σ2 of multiplicity
(n − 1), with the same eigenspaces of λ and µ respectively, and
2n − 3 1 1 1
σ1 = λ− µ and σ2 = µ− λ.
2(n − 1) 2 2 2(n − 1)
Splitting results for Riemannian manifolds admitting a Codazzi tensor with only two distinct
eigenvalues were obtained by Derdzinski [11] and Hiepko–Reckziegel [20, 21] (see again [2,
Chapter 16] for further discussion). In particular, it can be proved that, if the two distinct eigen-
values σ1 and σ2 are both “constant along the eigenspace span{E2 , . . . , En }” then the manifold
is locally a warped product on an interval of R of a (n − 1)–dimensional Riemannian manifold
(see [2, Chapter 16] and [31]).
Since σ2 has multiplicity (n − 1), larger than 2, we have for any two distinct indexes i, j ≥ 2,
∂i σ2 = ∂i S(Ej , Ej )
= ∇i Sjj + 2S(∇Ei Ej , Ej )
= ∇j Sij + 2σ2 g(∇Ei Ej , Ej )
= ∂j S(Ei , Ej ) − S(∇Ej Ei , Ej ) − S(Ei , ∇Ej Ej )
= − σ2 g(∇Ej Ei , Ej ) − σ2 g(Ei , ∇Ej Ej )
=0,

hence, σ2 is always constant along the eigenspace span{E2 , . . . , En }. The eigenvalue σ1 instead,
for a general LCF manifold, can vary, for example Rn endowed with the metric
dx2
g=
[1 + (x21 + x22+ · · · + x2n−1 )]2
14 GIOVANNI CATINO AND CARLO MANTEGAZZA

is LCF and
Rgij = −(n − 2)(∇2ij log A − ∇i log A∇j log A) + (∆ log A − (n − 2)|∇ log A|2 )δij
where the derivatives are the standard ones of Rn and A(x) = 1 + (x21 + x22 + · · · + x2n−1 ) (see [2,
Theorem 1.159]). Hence, this Ricci tensor “factorizes” on the eigenspaces he1 , . . . , en−1 i and hen i
but the eigenvalue σ1 of the Schouten tensor, which is given by
σ1 = g nn Rgnn = (∆ log A − (n − 2)|∇ log A|2 )A2
= A∆A − (n − 1)|∇A|2
= 2(n − 1)A − 4(n − 1)(A − 1)
= − 2(n − 1)(A − 2) ,
is clearly not constant along the directions e1 , . . . , en−1 .
The best we can say in general is that the metric of (M n , g) locally around every point can be
written as I × N and
dt2 + σ K (p)
g(t, p) =
[α(t) + β(p)]2
where σ K is a metric on N of constant curvature K, α : I → R and β : N → R are smooth
functions such that HessK β = f σ K , for some function f : N → R and where HessK is the
Hessian of (N, σ K ).

2.1. Compact LCF Ricci Solitons. A compact Ricci soliton is actually a gradient soliton (by the
work of Perelman [27]).
In general (even if they are not LCF), steady and expanding compact Ricci solitons are Einstein,
hence, when also LCF, they are of constant curvature (respectively zero and negative).
In [7, 12] it is proved that also shrinking, compact, LCF Ricci solitons are of constant positive
curvature, hence quotients of spheres.
Any compact, n–dimensional, LCF Ricci soliton is a quotient of Rn , Sn and Hn with their canonical
metrics, for every n ∈ N.

2.2. LCF Ricci Solitons with Constant Scalar Curvature. Getting back to the Schouten tensor,
if the scalar curvature R of an LCF Ricci soliton (M n , g) is constant, we have that also the other
eigenvalue σ1 of the Schouten tensor is constant along the eigenspace span{E2 , . . . , En }, that is,
∂i σ1 = 0, by simply differentiating the equality R = 2(n−1)
n−2 (σ1 + (n − 1)σ2 ).
Hence, by the above discussion, we can conclude that around every point of M n in the open
set Ω ⊂ M n where the two eigenvalues of the Ricci tensor are distinct the manifold (M n , g) is
locally a warped product I × N with g(t, p) = dt2 + h2 (t)σ(p) (this argument is due to Derdzin-
ski [11]).
Then the LCF hypothesis implies that the warp factor (N, σ) is actually a space of constant cur-
vature K (see for instance [4]).
As the scalar curvature R is constant, by the evolution equation ∂t R = ∆R + 2|Ric|2 we see
that also |Ric|2 is constant, that is, locally R = λ + (n − 1)µ = C1 and |Ric|2 = λ2 + (n − 1)µ2 = C2 .
Putting together these two equations it is easy to see that then both the eigenvalues µ and λ are
locally constant in Ω. Hence, by connectedness, either (M n , g) is Einstein, so a constant curvature
space, or the Ricci tensor has two distinct constant eigenvalues everywhere. Using now the local
warped product representation, the Ricci tensor is expressed by (see [2, Proposition 9.106] or [10,
p. 65] or [5, p. 168])
h00 2
dt + (n − 2)K − h h00 − (n − 2)(h0 )2 σ K .


(2.1) Ric = −(n − 1)
h
hence, h /h and ((n−2)K−h h −(n−2)(h0 )2 )/h2 are constant in t. This implies that (K−(h0 )2 )/h2
00 00

is also constant and h00 = Ch, then locally either the manifold (M n , g) is of constant curvature or
it is the Riemannian product of a constant curvature space with an interval of R.
 THE EVOLUTION OF THE WEYL TENSOR UNDER THE RICCI FLOW 15

By a maximality argument, passing to the universal covering of the manifold, we get the follow-
ing conclusion.
If n ≥ 4, any n–dimensional, LCF Ricci soliton with constant scalar curvature is either a quotient
of Rn , Sn and Hn with their canonical metrics or a quotient of the Riemannian products R × Sn−1 and
R × Hn−1 (see also [29]).

2.3. Gradient LCF Ricci Solitons with Nonnegative Ricci Tensor. Getting back again to the
Codazzi property of the Schouten tensor S, for every index i > 1, we have locally
∂1 R ∂i R ∂i R
0 = ∇1 Ri1 − ∇i R11 − gi1 + g11 = ∇1 Ri1 − ∇i R11 + .
2(n − 1) 2(n − 1) 2(n − 1)
α
If the soliton is a gradient LCF Ricci soliton, that is, Ric = −∇2 f + n g, we have R = −∆f + α
and taking the divergence of both sides
∂i R/2 = div Rici
= g jk ∇k Rij
= − g jk ∇k ∇i ∇j f
= − g jk ∇i ∇k ∇j f − g jk Rkijl ∇l f
= − ∇i ∆f − Ril ∇l f
= ∂i R − Ril ∇l f ,
where we used Schur’s Lemma ∂i R = 2 div Rici and the formula for the interchange of covariant
derivatives.
Hence, the relation ∂i R = 2Ril ∇l f holds and
Rij ∇j f
∇1 ∇2i1 f − ∇i ∇211 f = .
n−1
By means of the fact that W = 0, we compute now for i > 1 (this is a special case of the compu-
tation in Lemma 3.1 of [6]),
µ Rij ∇j f
∇i f =
n−1 n−1
= ∇1 ∇2i1 f − ∇i ∇211 f
= R1i1j ∇j f
 
1 R
= (R11 gij − R1j gi1 + Rij g11 − Ri1 g1j ) − (g11 gij − g1j gi1 ) ∇j f
n−2 (n − 1)(n − 2)
 
1 R
= (λgij + µgij ) − gij ∇j f
n−2 (n − 1)(n − 2)
 
λ+µ R
= − ∇i f
n−2 (n − 1)(n − 2)
(n − 1)λ + (n − 1)µ − λ − (n − 1)µ
= ∇i f
(n − 1)(n − 2)
λ
= ∇i f .
n−1
Then, in the open set Ω ⊂ M n where the two eigenvalues of the Ricci tensor are distinct, the vector
field ∇f is parallel to E1 , hence it is an eigenvector of the Ricci tensor and ∂i R = 2Ril ∇l f = 0,
for every index i > 1.
n−2
As σ1 = 2(n−1) R − (n − 1)σ2 we get that also ∂i σ1 = 0 for every index i > 1.
The set Ω is dense, otherwise its complement where Ric − Rg/n = 0 has interior points and,
by Schur’s Lemma, the scalar curvature would be constant in some open set of M n . Then, strong
16 GIOVANNI CATINO AND CARLO MANTEGAZZA

maximum principle applied to the equation ∂t R = ∆R + 2|Ric|2 implies that R is constant every-
where on M n , and we are in the previous case.
So we can conclude also in this case by the previous argument that the manifold, locally
around every point in Ω, is a warped product on an interval of R of a constant curvature space
LK . Moreover, Ω is obviously invariant by “translation” in the LK –direction.
We consider a point p ∈ Ω and the maximal geodesic curve γ(t) passing from p orthogonal to
K
L , contained in Ω. It is easy to see that for every compact, connected segment of such geodesic
we have a neighborhood U and a representation of the metric in g as
g = dt2 + h2 (t)σ K ,
covering the segment with the local charts and possibly shrinking them in the orthogonal direc-
tions.
Assuming from now on that the the Ricci tensor is nonnegative, by the local warped representa-
tion formula (2.1) we see that h00 ≤ 0 along such geodesic, as Rtt ≥ 0.
If such geodesic has no “endpoints”, being concave the function h must be constant and we have
either a flat quotient of Rn or the Riemannian product of R with a quotient of Sn−1 . The same
holds if the function h is constant in some interval, indeed, the manifold would be locally a Rie-
mannian product and the scalar curvature would be locally constant (hence we are in the case
above).
If there is at least one endpoint, one of the following two situations happens:
• the function h goes to zero at such endpoint,
• the geodesic hits the boundary of Ω.
If h goes to zero at an endpoint, by concavity (h0 )2 must converge to some positive limit and by
the smoothness of the manifold, considering again formula (2.1), the quantity K − (h0 )2 must go
to zero as h goes to zero, hence K > 0 and the constant curvature space LK must be a quotient of
the sphere Sn−1 (if the same happens also at the other endpoint, the manifold is compact). Then,
by topological reasons we conclude that actually the only possibility for LK is the sphere Sn−1
itself.
Assuming instead that h does not go to zero at any endpoint, where the geodesic hits the bound-
ary of Ω the Ricci tensor is proportional to the metric, hence, again by the representation for-
mula (2.1), the quantity K − (h0 )2 is going to zero and either K = 0 or K > 0.
The case K = 0 is impossible, indeed h0 would tend to zero at such endpoint, then by the concav-
ity of h the function h0 has a sign, otherwise h is constant in an interval, implying that in some
open set (M n , g) is flat, which cannot happens since we are in Ω. Thus, being h0 6= 0, h concave
and we assumed that h does not go to zero, there must be another endpoint where the geodesic
hits the boundary of Ω, which is in contradiction with K = 0 since also in this point K − (h0 )2
must go to zero but instead h0 tends to some nonzero value. Hence, K must be positive and also
in this case we are dealing with a warped product of a quotient of Sn−1 on an interval of R.
Resuming, in the non–product situation, every connected piece of Ω is a warped product of
a quotient of the sphere Sn−1 on some intervals of R. Then, we can conclude that the univer-
sal cover (M f, ge) can be recovered “gluing together”, along constant curvature spheres, warped
product pieces that can be topological “caps” (when h goes to zero at an endpoint) and “cylin-
ders”. Nontrivial quotients (M, g) of (M f, ge) are actually possible only when there are no “caps”
in this gluing procedure. In such case, by its concavity, the function h must be constant along
every piece of geodesic and the manifold (M f, ge) is a Riemannian product. If there is at least one
“cap”, the whole manifold is a warped product of Sn−1 on an interval of R.
Remark 2.2. We do not know if the condition on (M n , g) to be a gradient LCF Ricci soliton is
actually necessary to have locally a warped product. We conjecture that such conclusion should
hold also for nongradient LCF Ricci solitons.

If n ≥ 4, any n–dimensional, LCF gradient Ricci soliton with nonnegative Ricci tensor is either a
quotient of Rn and Sn with their canonical metrics, or a quotient of R × Sn−1 or it is a warped product of
Sn−1 on a proper interval of R.
 THE EVOLUTION OF THE WEYL TENSOR UNDER THE RICCI FLOW 17

2.4. The Classification of Steady and Shrinking Gradient LCF Ricci Solitons. The class of soli-
tons with nonnegative Ricci tensor is particularly interesting as it includes all the shrinking and
steady Ricci solitons.
Indeed, by the same arguments of [32] (keeping in mind, in following the proof of the main
Proposition 3.2 in such paper, that the nonnegativity of the scalar curvature for every complete,
ancient Ricci was proved in [8, Corollary 2.5]), where the author generalizes the well–known
Hamilton–Ivey curvature estimate to locally conformally flat, gradient, shrinking Ricci solitons
(Corollary 3.3 in the same paper [32]), it follows that actually every complete ancient solution g(t)
to the Ricci flow whose Weyl tensor is identically zero for all times, is forced to have nonnegative
curvature operator for every time t.
In particular, this holds for any complete, steady or shrinking Ricci soliton (even if not gradient)
as they generate self–similar ancient solutions of Ricci flow.
By the previous discussion and the analysis of Bryant in the steady case [5] (see also [9, Chap-
ter 1, Section 4]) showing that there exists a unique (up to dilation of the metric) nonflat, steady,
gradient Ricci soliton which is a warped product of Sn−1 on a halfline of R, called Bryant soliton,
we get the following classification.
Proposition 2.3. The steady, gradient, LCF Ricci solitons of dimension n ≥ 4 are given by the quotients
of Rn and the Bryant soliton.
This classification result, including also the three–dimensional LCF case, was first obtained
recently by H.-D. Cao and Q. Chen [6].
In the shrinking case, the analysis of Kotschwar [22] of rotationally invariant shrinking, gra-
dient Ricci solitons gives the following classification where the Gaussian soliton is defined as the
flat Rn with a potential function f = α|x|2 /2n, for a constant α ∈ R.
Proposition 2.4. The shrinking, gradient, LCF Ricci solitons of dimension n ≥ 4 are given by the quo-
tients of Sn , the Gaussian solitons with α > 0 and quotients of R × Sn−1 .
This classification of shrinking, gradient, LCF Ricci solitons follows by the works of L. Ni and
N. Wallach [26], P. Petersen and W. Wylie [29] and Z.-H. Zhang [32].
Several other authors contributed to the subject, including X. Cao, B. Wang and Z. Zhang [7], B.-
L. Chen [8], M. Fernández–López and E. Garcı́a–Rı́o [13], M. Eminenti, G. La Nave and C. Man-
tegazza [12], O. Munteanu and N. Sesum [24] and again P. Petersen and W. Wilye [28].
We show now that every complete, warped, LCF Ricci soliton with nonnegative Ricci tensor
is actually a gradient soliton.
Proving our conjecture in Remark 2.2 that every Ricci soliton is locally a warped product would
then lead to have a general classification of also nongradient Ricci solitons, in the steady and
shrinking cases.
Remark 2.5. In the compact case, the fact that every Ricci soliton is actually a gradient is a con-
sequence of the work of Perelman [27]. Naber [25] showed that it is true also for shrinking Ricci
solitons with bounded curvature.
For examples of nongradient Ricci solitons see Baird and Danielo [1].
Proposition 2.6. Let (M n , g) be a complete, warped, LCF Ricci soliton with nonnegative Ricci tensor,
then it is a gradient Ricci soliton with a potential function f : M n → R depending only on the t variable
of the warping interval.
Proof. We assume that (M n , g) is globally described by M n = I × LK and
g = dt2 + h2 (t)σ K ,
where I is an interval of R or S1 and (LK , σ K ) is a complete space of constant curvature K.
In the case h is constant, which clearly follows if I = S1 , as h00 ≤ 0 the conclusion is trivial.
We deal then with the case where h : I → R is zero at some point, let us say h(0) = 0 and
I = [0, +∞), (if the interval I is bounded the manifold M n is compact and we are done). Then,
18 GIOVANNI CATINO AND CARLO MANTEGAZZA

LK = Sn−1 with its constant curvature metric σ K . As a consequence, we have M n = Rn , simply

connected. We consider the form ω satisfying the structural equation
1 α
Rγβ + (∇γ ωβ + ∇β ωγ ) = gγβ ,
2 n
If ϕ : Sn−1 → Sn−1 is an isometry of the standard sphere, the associated map φ : M n → M n
given by φ(t, p) = (t, ϕ(p)) is also an isometry, moreover, by the warped structure of M n we have
that the 1–form φ∗ ω also satisfies
1  α
Rγβ + (∇φ∗ ω)γβ + (∇φ∗ ω)βγ = gγβ ,
2 n
Calling I the Lie group of isometries of Sn−1 and ξ the Haar unit measure associated to it, we
define the following 1–form Z
θ= φ∗ ω dξ(ϕ) .
I
By the linearity of the structural equation, we have
1 α
Rγβ + (∇γ θβ + ∇β θγ ) = gγβ ,
2 n
moreover, by construction, we have LX θ = 0 for every vector field X on M n which is a generator
of an isometry φ of M n as above (in other words, θ depends only on t). Computing in normal
coordinates on Sn−1 , we get
∇i θj = − θ(∇j ∂i ) = −Γtij θt = hh0 σij
K
θt ,
h0
∇i θt = − θ(∇t ∂i ) = −Γjti θj = − θi .
h

Hence,
α h00
= Rtt + ∇t θt = −(n − 1) + ∂t θt ,
n h
h0
0 = ∇i θt + ∇t θi = ∂t θi − 2 θi ,
h
α K 1
gij = Rij + (∇i θj + ∇j θi ) = (n − 2)(K − (h0 )2 ) − hh00 + hh0 θt gij

K
.
n 2
It is possible to see that, by construction, actually θi = 0 for every i at every point, but it is easier
to consider directly the 1–form σ = θt dt on M n and checking that it also satisfies these three
equations as θ, hence the structural equation
1 α
Rγβ + (∇γ σβ + ∇β σγ ) = gγβ .
2 n
It is now immediate to see that, dσit = ∇i σt − ∇t σi = 0 and dσij = ∇i σj − ∇j σi = 0, so the form
σ is closed and being M n simply connected, there exists a smooth function f : M → R such that
df = σ, thus
α
Rγβ + ∇2γβ f = gγβ ,
n
that is, the soliton is a gradient soliton.
It is also immediate to see that the function f depends only on t ∈ I. 
In the expanding, noncompact case (in the compact case the soliton can be only a quotient of
the hyperbolic space Hn ), if the Ricci tensor is nonnegative and (M n , g) is a gradient soliton, then
either it is a warped product of Sn−1 (and M n = Rn ) or it is the product of R with a constant
curvature space, but this last case is possible only if the soliton is the Gaussian expanding Ricci
soliton, α < 0, on the flat Rn .
For a discussion of the expanding Ricci solitons which are warped products of Sn−1 see [9,
Chapter 1, Section 5], where the authors compute, for instance, an example with positive Ricci
tensor (analogous to the Bryant soliton).
 THE EVOLUTION OF THE WEYL TENSOR UNDER THE RICCI FLOW 19

To our knowledge, the complete classification of complete, expanding, gradient, LCF Ricci soli-
tons is an open problem, even if they are rotationally symmetric.

3. S INGULARITIES OF R ICCI F LOW WITH B OUNDED W EYL T ENSOR

Let (M , g(t)) be a Ricci flow with M n compact on the maximal interval [0, T ), with T < +∞.
n

Hamilton proved that

max |Rm|(·, t) → ∞
M
as t → T .
We say that the solution has a Type I singularity if
max (T − t)|Rm|(p, t) < +∞ ,
M ×[0,T )

otherwise we say that the solution develops a Type IIa singularity.

By Hamilton’s procedure in [19], one can choose a sequence of points pi ∈ M n and times ti ↑ T
such that, dilating the flow around these points in space and time, such sequence of rescaled Ricci
flows (using Hamilton–Cheeger–Gromov compactness theorem in [18] and Perelman’s injectivity
radius estimate in [27]) converges to a complete maximal Ricci flow (M∞ , g∞ (t)) in an interval
t ∈ (−∞, b) where 0 < b ≤ +∞.
Moreover, in the case of a Type I singularity, we have 0 < b < +∞, |Rm∞ |(p∞ , 0) = 1 for some
point p∞ ∈ M∞ and |Rm∞ |(p, t) ≤ 1 for every t ≤ 0 and p ∈ M∞ .
In the case of a Type IIa singularity, b = +∞, |Rm∞ |(p∞ , 0) = 1 for some point p∞ ∈ M∞ and
|Rm∞ |(p, t) ≤ 1 for every t ∈ R and p ∈ M∞ .
These ancient limit flows were called by Hamilton singularity models. We want now to discuss
them in the special case of a Ricci flow with uniformly bounded Weyl tensor (or with a blow up
rate of the Weyl tensor which is of lower order than the one of the Ricci tensor). The Ricci flow
under this condition is investigated also in [23].
Clearly, any limit flow consists of LCF manifolds, hence, by Corollary 1.2 and the cited results
of Chen [8] and Zhang [32] at every time and every point the manifold has nonnegative curva-
ture operator and either the Ricci tensor is proportional to the metric or it has an eigenvalue of
multiplicity (n − 1) and another of multiplicity 1.
We follow now the argument in the proof of Theorem 1.1 in [6].
We recall the following splitting result (see [10, Chapter 7, Section 3]) which is a consequence of
Hamilton’s strong maximum principle for systems in [16].
Theorem 3.1. Let (M n , g(t)), t ∈ (0, T ) be a simply connected complete Ricci flow with nonnegative
curvature operator. Then, for every t ∈ (0, T ) we have that (M n , g(t)) is isometric to the product of the
following factors,
(1) the Euclidean space,
(2) an irreducible nonflat compact Einstein symmetric space with nonnegative curvature operator and
positive scalar curvature,
(3) a complete Riemannian manifold with positive curvature operator,
(4) a complete Kähler manifold with positive curvature operator on real (1, 1)–forms.
Since we are in the LCF case, every Einstein factor above must be a sphere (the scalar curva-
ture is positive). The Kähler factors can be excluded as the following relation holds for Kähler
manifolds of complex dimension m > 1 at every point (see [2, Proposition 2.68])
3(m − 1)
|W|2 ≥ R2 .
m(m + 1)(2m − 1)
Thus, any Kähler factor would have zero scalar curvature, hence would be flat. Finally, by the
structure of the Ricci tensor and the fact that these limit flows are nonflat, it is easy to see that
only a single Euclidean factor of dimension one is admissible, moreover, in this case there is only
another factor Sn−1 .
20 GIOVANNI CATINO AND CARLO MANTEGAZZA

In conclusion, passing to the universal cover, the possible limit flows are quotients of R × Sn−1
or have a positive curvature operator.
Proposition 3.2 (LCF Type I singularity models). Let (M n , g(t)), for t ∈ [0, T ), be a compact smooth
solution to the Ricci flow with uniformly bounded Weyl tensor.
If g(t) develops a Type I singularity, then there are two possibilities:
(1) M n is diffeomorphic to a quotient of Sn and the solution to the normalized Ricci flow converges to
a constant positive curvature metric.
In this case the singularity model must be a shrinking compact Ricci soliton by a result of Sesum [30],
hence by the analysis in the previous section, a quotient of Sn (this also follows by the work of Böhm
and Wilking [3]).
(2) There exists a sequence of rescalings which converges to the flow of a quotient of R × Sn−1 .
Proof. By the previous discussion, either the curvature operator is positive at every time or the
limit flow is a quotient of R × Sn−1 .
Hence, we assume that every manifold in the limit flow has positive curvature operator. The
family of metrics g∞ (t) is a complete, nonflat, LCF, ancient solution with uniformly bounded
positive curvature operator which is k–non collapsed at all scales (hence a k–solution in the sense
of [27]). By a result of Perelman in [27], we can find a sequence of times ti & −∞ such that a
sequence of suitable dilations of g∞ (ti ) converges to a nonflat, gradient, shrinking, LCF Ricci
soliton. Hence, we can find an analogous sequence for the original flow. By the classification in
the previous section, the thesis of the proposition follows. 
Remark 3.3. Notice that in case (2) we are not claiming that every Type I singularity model is a
gradient shrinking Ricci soliton.
This problem is open also in the LCF situation.
Proposition 3.4 (LCF Type IIa singularity models). Let (M n , g(t)), for t ∈ [0, T ), be a compact
smooth solution to the Ricci flow with uniformly bounded Weyl tensor. If the flow develops a Type IIa
singularity, then there exists a sequence of dilations which converges to the Bryant soliton.
Proof. As we said, if the curvature operator gets some zero eigenvalue, the limit flow is a quotient
of R × Sn−1 which cannot be a steady soliton as it is not eternal. Hence, the curvature operator is
positive.
By Hamilton’s work [17], any Type IIa singularity model with nonnegative curvature operator
and positive Ricci tensor is a steady, nonflat, gradient Ricci soliton. Since in our case such soliton
is also LCF, by the analysis of the previous section, it must be the Bryant soliton. 

Acknowledgments. We thank Peter Petersen and Gérard Besson for several valuable suggestions.
We also wish to thank Fabrizio Bracci, Alessandro Cameli and Paolo Dell’Anna for several interesting
comments on earlier versions of the paper.
The authors are partially supported by the Italian project FIRB–IDEAS “Analysis and Beyond”.
The second author is partially supported by the Italian GNAMPA (INdAM) Group.

R EFERENCES
1. P. Baird and L. Danielo, Three–dimensional Ricci solitons which project to surfaces, J. Reine Angew. Math. 608 (2007),
65–91.
2. A. L. Besse, Einstein manifolds, Springer–Verlag, Berlin, 2008.
3. C. Böhm and B. Wilking, Manifolds with positive curvature operators are space forms, Ann. of Math. (2) 167 (2008), no. 3,
1079–1097.
4. M. Brozos-Vázquez, E. Garcı́a-Rı́o, and R. Vázquez-Lorenzo, Some remarks on locally conformally flat static space-times,
J. Math. Phys. 46 (2005), no. 2, 022501, 11.
5. R. L. Bryant, Local existence of gradient Ricci solitons, Unpublished work, 1987.
6. H.-D. Cao and Q. Chen, On locally conformally flat gradient steady Ricci solitons, Trans. Amer. Math. Soc. 364 (2012),
2377–2391.
7. X. Cao, B. Wang, and Z. Zhang, On locally conformally flat gradient shrinking Ricci solitons, Commun. Contemp. Math.
13 (2011), no. 2, 269–282.
 THE EVOLUTION OF THE WEYL TENSOR UNDER THE RICCI FLOW 21

8. B.-L. Chen, Strong uniqueness of the Ricci flow, J. Diff. Geom. 82 (2009), 363–382.
9. B. Chow, S.-C. Chu, D. Glickenstein, C. Guenther, J. Isenberg, T. Ivey, D. Knopf, P. Lu, F. Luo, and L. Ni, The Ricci
flow: techniques and applications. Part I. Geometric aspects, Mathematical Surveys and Monographs, vol. 135, American
Mathematical Society, Providence, RI, 2007.
10. B. Chow, P. Lu, and L. Ni, Hamilton’s Ricci flow, Graduate Studies in Mathematics, vol. 77, American Mathematical
Society, Providence, RI, 2006.
11. A. Derdzinski, Some remarks on the local structure of Codazzi tensors, Global differential geometry and global analysis
(Berlin, 1979), Lect. Notes in Math., vol. 838, Springer–Verlag, Berlin, 1981, pp. 243–299.
12. M. Eminenti, G. La Nave, and C. Mantegazza, Ricci solitons: the equation point of view, Manuscripta Math. 127 (2008),
no. 3, 345–367.
13. M. Fernández-López and E. Garcı́a-Rı́o, Rigidity of shrinking Ricci solitons, Math. Z. 269 (2011), no. 1-2, 461–466.
14. S. Gallot, D. Hulin, and J. Lafontaine, Riemannian geometry, Springer–Verlag, 1990.
15. R. S. Hamilton, Three–manifolds with positive Ricci curvature, J. Diff. Geom. 17 (1982), no. 2, 255–306.
16. , Four–manifolds with positive curvature operator, J. Diff. Geom. 24 (1986), no. 2, 153–179.
17. , Eternal solutions to the Ricci flow, J. Diff. Geom. 38 (1993), no. 1, 1–11.
18. , A compactness property for solutions of the Ricci flow, Amer. J. Math. 117 (1995), no. 3, 545–572.
19. , The formation of singularities in the Ricci flow, Surveys in differential geometry, Vol. II (Cambridge, MA, 1993),
Int. Press, Cambridge, MA, 1995, pp. 7–136.
20. S. Hiepko, Eine innere Kennzeichnung der verzerrten Produkte, Math. Ann. 241 (1979), no. 3, 209–215.
21. S. Hiepko and H. Reckziegel, Über sphärische Blätterungen und die Vollständigkeit ihrer Blätter, Manuscripta Math. 31
(1980), no. 1–3, 269–283.
22. B. Kotschwar, On rotationally invariant shrinking Ricci solitons, Pacific J. Math. 236 (2008), no. 1, 73–88.
23. L. Ma and L. Cheng, On the conditions to control curvature tensors or Ricci flow, Ann. Global Anal. Geom. 37 (2010), no. 4,
403–411.
24. O. Munteanu and N. Sesum, On gradient Ricci solitons, J. Geom. Anal. 23 (2013), no. 2, 539–561.
25. A. Naber, Noncompact shrinking four solitons with nonnegative curvature, J. Reine Angew. Math. 645 (2010), 125–153.
26. L. Ni and N. Wallach, On a classification of gradient shrinking solitons, Math. Res. Lett. 15 (2008), no. 5, 941–955.
27. G. Perelman, The entropy formula for the Ricci flow and its geometric applications, ArXiv Preprint Server – http://arxiv.org,
2002.
28. P. Petersen and W. Wylie, Rigidity of gradient Ricci solitons, Pacific J. Math. 241 (2009), no. 2, 329–345.
29. , On the classification of gradient Ricci solitons, Geom. Topol. 14 (2010), no. 4, 2277–2300.
30. N. Sesum, Convergence of the Ricci flow toward a soliton, Comm. Anal. Geom. 14 (2006), no. 2, 283–343.
31. R. Tojeiro, Conformal de Rham decomposition of Riemannian manifolds, Houston J. Math. 32 (2006), no. 3, 725–743 (elec-
tronic).
32. Z.-H. Zhang, Gradient shrinking solitons with vanishing Weyl tensor, Pacific J. Math. 242 (2009), no. 1, 189–200.
33. , On the completeness of gradient Ricci solitons, Proc. Amer. Math. Soc. 137 (2009), no. 8, 2755–2759.

(Giovanni Catino) SISSA – I NTERNATIONAL S CHOOL FOR A DVANCED S TUDIES , V IA B ONOMEA 265, T RIESTE , I TALY,
34136
E-mail address, G. Catino: [email protected]

(Carlo Mantegazza) S CUOLA N ORMALE S UPERIORE DI P ISA , P. ZA C AVALIERI 7, P ISA , I TALY, 56126
E-mail address, C. Mantegazza: [email protected]