Positive intermediate Ricci curvature on connected sums

Philipp Reiser∗ and David J. Wraith

arXiv:2310.02746v1 [math.DG] 4 Oct 2023

October 5, 2023

Abstract
We consider the problem of performing connected sums in the context of positive
kth intermediate Ricci curvature. We show that such connected sums are possible
if the manifolds involved possess ‘k-core metrics’ for some k. Here, a k-core metric
is a generalization of the notion of core metric introduced by Burdick for positive
Ricci curvature. Further, we show that connected sums of linear sphere bundles over
bases admitting such metrics admit positive kth intermediate Ricci curvature for k in
a particular range. This follows from a plumbing result we establish, which generalizes
other recent plumbing results in the literature and is possibly of independent interest.
As an example of a manifold admitting a k-core metric, we prove that HP n admits a
(4n − 3)-core metric and that OP 2 admits a 9-core metric, and we show that in both
cases these are optimal.

1 Introduction
We consider the question of whether a given curvature condition can be preserved under con-
nected sums. While this is always possible for positive scalar curvature as shown by Gromov–
Lawson [GL], Gromov’s Betti number bound [Gr] implies that the same cannot be true for
positive sectional curvature. For positive Ricci curvature, the situation is more delicate:
By the theorem of Bonnet–Myers the connected sum of two closed, non-simply-connected
Ricci-positive manifolds cannot admit a metric of positive Ricci curvature. However, if at
least one of the manifolds is simply-connected, the question is open. This problem was sys-
tematically studied by Burdick [Bu1, Bu2, Bu3], who, based on earlier work by Perelman
[Pe], introduced the notion of core metrics and showed that the connected sum of manifolds
with core metrics admits a metric of positive Ricci curvature.
In this article, we extend Burdick’s techniques to a family of stricter curvature conditions:
Definition 1.1. A Riemannian manifold (M n , g) has positive k-th intermediate Ricci curva-
ture for some k ∈ {1, . . . , n − 1}, denoted Rick > P
0, if for every unit tangent vector v ∈ T M
and any orthonormal k-frame (e ) in v the sum ki=1 K(v, ei ) is positive, where K denotes
i ⊥

the sectional curvature.

2010 Mathematics subject classification. 53C20
The author acknowledges funding by the SNSF-Project 200020E 193062 and the DFG-Priority pro-
∗

gramme SPP 2026.

1
 Note that for k = 1 and k = n − 1, we recover the conditions of positive sectional
curvature and positive Ricci curvature, respectively.
Definition 1.2. Let M be an n-dimensional manifold and let k ∈ {1, . . . , n − 1}. A Rie-
mannian metric g on M is called a k-core metric if g has Rick > 0 and if there exists an
embedding ϕ : D n ֒→ M such that
(i) The induced metric g|ϕ(S n−1 ) is the round metric of radius one, and
(ii) IIϕ(S n−1 ) is positive semi-definite with respect to the outward normal of S n−1 ⊆ D n .
Note that for k = n − 1 we recover the original definition given in [Bu2] except for the
fact that the second fundamental form is required to be strictly positive in [Bu2]. However,
a core metric in the sense of Definition 1.2 can always be deformed into a core metric in the
sense of [Bu2], see e.g. [Bu1, Proposition 1.2.11].
In [Bu2, Theorem B], it is shown that connected sums of manifolds with (n − 1)-core
metrics support positive Ricci curvature. We can now generalize this as follows.
Theorem A. Let M1 , . . . , Mℓ be n-dimensional manifolds that admit k-core metrics, where
k ≥ 2. Then M1 # . . . #Mℓ admits a metric with Rick > 0.
The main ingredients in the proof of Theorem A are the gluing theorem for positive
intermediate Ricci curvature established in [RW3], together with the construction of a metric
with Ric2 > 0 on S n \ ⊔ℓ (D n )◦ , (which is called the docking station in [Bu2]), whose second
fundamental form on the boundary can be made arbitrarily small, see Theorem 3.1.
Remark 1.3. Since the metric on the docking station is invariant under the action of O(n −
1)O(2) ⊆ O(n + 1), we can take quotients by finite subgroups of O(n − 1)O(2) that act
freely as in [Bu2, Corollary 4.7]. In this way we obtain in the situation of Theorem A that
RP n #M1 # . . . #Mℓ and L#M1 # . . . #Mℓ admits a metric of Rick > 0, where L is any n-
dimensional lens space (and n is assumed to be odd in this case). Note that by [Bu1, Lemma
1.2.9], lens spaces and real projective spaces are the only additional summands we can obtain
in this way.
Concerning the existence of k-core metrics, by a result of Wu [Wu], the boundary condi-
tion (ii) in Definition 1.2 imposes the following topological obstruction.
Proposition 1.4. Let M be a closed n-dimensional manifold that admits a k-core metric.
Then M is (n − k)-connected. In particular, if k ≤ ⌊ n+1
2
⌋, then M is a homotopy sphere.
We immediately obtain the following restrictions in low dimensions: Every closed 3-
manifold with a k-core metric is diffeomorphic to the standard sphere and the same holds
in dimension 5 when k ≤ 3. In dimension 4 every closed manifold with a k-core metric is
homeomorphic to the standard sphere when k ≤ 2.
On the other hand, it is easy to see that the round metric on S n is a 1-core metric. Further,
by [Bu2], complex and quaternionic projective spaces and the Cayley plane of dimension n
admit (n−1)-core metrics, where n denotes the real dimension of the corresponding manifold.
By Proposition 1.4, this is optimal for complex projective spaces. For quaternionic projective
spaces and the Cayley plane we obtain the following improvement, which again is optimal
by Proposition 1.4.

2
Theorem B. HP n admits a (4n − 3)-core metric and OP 2 admits a 9-core metric.
In [RW1] it was shown that a Betti number bound as in the case of non-negative sectional
curvature [Gr] cannot hold for Rick > 0 for all k ≥ ⌊ n2 ⌋ + 2, where n denotes the dimension.
By considering connected sums of copies of HP 2 and OP 2 using Theorems A and B, we can
slightly improve this result as follows.
Corollary C. For any ℓ ∈ N the manifold #ℓ HP 2 admits a metric of Ric5 > 0 and the
manifold #ℓ OP 2 admits a metric of Ric9 > 0. In particular, Gromov’s Betti number bound
does not hold in dimension 8 for Ric5 > 0 and in dimension 16 for Ric9 > 0.
By using manifolds with k-core metrics as base manifolds of fibre bundles, we can also
consider plumbings as in the following theorem, which generalizes results for positive Ricci
curvature in [Re1] and [Wr1], and for positive intermediate Ricci curvature in [RW2].
Theorem D. Let W be the manifold obtained by plumbing linear disc bundles with compact
base manifolds according to a simply-connected graph. Suppose the following:
(i) For a fixed bundle in this graph the base admits a metric with Rick1 > 0 for some k1 .
Denote the base dimension by q + 1 and the fibre dimension by p + 1,
(ii) Every other bundle in this graph with base dimension q + 1 admits a k1 -core metric,
(iii) Every bundle with base dimension p + 1 admits a k2 -core metric for some k2 .
Then, if p, q ≥ 2, the manifold ∂W admits a metric of Rick > 0 for all k ≥ max{p + 2, p +
k1 , q + 2, q + k2 }.
We can use plumbings as in Theorem D to construct connected sums of sphere bundles
as follows.
Corollary E. Let Ei → Biq , 1 ≤ i ≤ ℓ, be linear S p -bundles with compact base manifolds
such that B1 admits a metric of Rick > 0 and each Bi , 2 ≤ i ≤ ℓ, admits a k-core metric.
Then the connected sum E1 # . . . #Eℓ admits a metric of Rick > 0 for all k ≥ max{p + 2, p +
k, q + 1}.
This paper is laid out as follows. In Section 2 we prove a generalization of the main
technical result in [RW1]. The aim is to establish criteria which identify when metrics (of
the type under consideration in this paper) have Rick > 0. In Section 3 we prove that the
neck construction from [Pe] actually gives a metric with Ric2 > 0, and we use this to prove
Theorem A. The remaining results (Theorems B, D, and Corollary E) are then established
in Section 4.

2 Preliminaries
Let (M n , g) be a Riemannian manifold. To characterize the condition Rick > 0 we consider
the curvature operator R : Λ2 T M → Λ2 T M defined by

g(R(v1 ∧ v2 ), v3 ∧ v4 ) = g(R(v1, v2 )v4 , v3 ),

3
where Λ2 T M is equipped with the Riemannian metric which is the natural extension of g to
Λ2 T M, i.e.
g(v1 ∧ v2 , v3 ∧ v4 ) = g(v1 , v3 )g(v2 , v4 ) − g(v1 , v4 )g(v2 , v3 ).
We recall the following definitions of [RW1]: For an inner product space V the set
{v0 ∧v1 , . . . , v0 ∧vk } ⊆ Λ2 V , where (v0 , . . . , vk ) is an orthonormal (k +1)-frame in V , is called
a k-chain with base v0 . For a linear map A : Λ2 V → Λ2 V and a k-chain {v0 ∧ v1 , . . . , v0 ∧ vk }
the sum
X k
hA(v0 ∧ vi ), v0 ∧ vi i
i=1

is the value of A on this k-chain. Note that (M, g) has Rick > 0 if and only if at every point
in M the value of R on every k-chain is positive.
In [RW1] we considered the condition Rick > 0 for doubly warped product metrics. In
this case each tangent space splits orthogonally into a direct sum V1 ⊕ V2 ⊕ V3 such that each
subspace Vi ∧ Vj is an eigenspace for R. Below we will be interested in the following more
general situation.

Proposition 2.1. Let (V, h·, ·i) be a finite-dimensional inner product space of dimension n
and let A : Λ2 V → Λ2 V be a linear self-adjoint map. Suppose that V splits orthogonally as

V = V1 ⊕ V2 ⊕ V3

so that V1 and V2 are one-dimensional and A is given by

A(v1 ∧ v2 ) =λ12 v1 ∧ v2 ,
A(v1 ∧ w1 ) =λ13 v1 ∧ w1 + λ̃v2 ∧ w1 ,
A(v2 ∧ w1 ) =λ23 v2 ∧ w1 + λ̃v1 ∧ w1 ,
A(w1 ∧ w2 ) =λ3 w1 ∧ w2 ,

for some λ12 , λ13 , λ23 , λ̃, λ3 ∈ R, where v1 and v2 are unit vectors in V1 and V2 , respectively,
and w1 , w2 ∈ V3 . Then for 2 ≤ k ≤ n − 3 the value of A on every k-chain is positive if and
only if the following inequalities are satisfied:

(i) λ12 + 21 (k − 1)(λ13 + λ23 ) > 0,

(ii) (λ12 + (k − 1)λ13 )(λ12 + (k − 1)λ23 ) > (k − 1)2 λ̃2 ,

(iii) λ13 λ23 > λ̃2 ,

(iv) λ13 , λ23 , λ3 > 0.

For k = n − 2, n − 1 these inequalities are still sufficient, but not necessary.

4
Proof. First note that if λ̃ = 0, then the spaces Vi ∧ Vj are eigenspaces for A, so we are in
the situation of [RW1, Proposition 2.3]. Observe that (i)–(iv) in this case now become

λ12 + (k − 1)λ13 > 0,

λ12 + (k − 1)λ23 > 0,
λ13 , λ23 , λ3 > 0,

and these are precisely the inequalities appearing in [RW1, Proposition 2.3] for k ≤ n − 3,
and for k = n − 2, n − 1 these inequalities are easily seen to be implied by those appearing
in [RW1, Proposition 2.3].
From now on we can therefore assume that λ̃ 6= 0. We modify the vectors v1 and v2 as
follows. First, Let q
λ13 − λ23 + (λ13 − λ23 )2 + 4λ̃2
µ= .
2λ̃
and define
v1′ = µv1 + v2 , v2′ = −v1 + µv2 .
Let V1′ and V2′ be the subspaces generated by v1′ and v2′ , respectively, and set V3′ = V3 . Then
V1′ and V2′ are orthogonal and V1′ ⊕ V2′ = V1 ⊕ V2 . A calculation shows that the spaces Vi′ ∧ Vj′
are eigenspaces for A with eigenvectors λ′ij given by

λ′12 = λ12 ,
 
1
q
′
λ13 = λ13 + λ23 + (λ13 − λ23 )2 + 4λ̃2
2
 
1
q
′
λ23 = λ13 + λ23 − (λ13 − λ23 )2 + 4λ̃2
2
λ′33 = λ3 .

By [RW1, Proposition 2.3], the value of A on every k-chain is positive if and only if the sum
of any k non-diagonal elements in each row of the following (n × n) matrix is positive.

0 λ′12 λ′13 · · · · · · λ′13

 
λ′12 0 λ′23 · · · · · · λ′23 
 ′
λ13 λ′23 0 λ′33 · · · λ′33 

 . .. .. 
 . .. ..
 . . λ33′ . . . 

 . .. .. .. ..
 ..

. . . . λ′33 
λ′13 λ′23 λ′33 · · · λ′33 0

For 2 ≤ k ≤ n − 3 this is equivalent to the following system of inequalities.

λ′12 + (k − 1)λ′13 > 0, λ′12 + (k − 1)λ′23 > 0,

λ′13 + λ′23 + (k − 2)λ′33 > 0, λ′13 + (k − 1)λ′33 > 0,
λ′23 + (k − 1)λ′33 > 0, λ′13 , λ′23 , λ′33 > 0.

5
For k = n − 2, n − 1 these inequalities are still sufficient, but not all are necessary.
Note that the inequalities λ′13 + λ′23 + (k − 2)λ′33 > 0, λ′13 + (k − 1)λ′33 > 0 and λ′23 + (k −
1)λ′33 > 0 are superfluous. Hence, we arrive at the following system of inequalities.
 
1
q
2
λ12 + (k − 1) λ13 + λ23 + (λ13 − λ23 ) + 4λ̃ > 0, 2 (1)
2
 
1
q
2
λ12 + (k − 1) λ13 + λ23 − (λ13 − λ23 ) + 4λ̃ > 0, 2 (2)
2
q
λ13 + λ23 + (λ13 − λ23 )2 + 4λ̃2 > 0, (3)
q
λ13 + λ23 − (λ13 − λ23 )2 + 4λ̃2 > 0, (4)
λ3 > 0. (5)

First note that (2) implies (1), and (4) implies (3). Hence, we are left with (2), (4) and
(5). Next, observe that (4) implies λ13 + λ23 > 0 and is therefore equivalent to

λ13 λ23 > λ̃2 .

In particular, λ13 λ23 > 0. This observation, together with λ13 + λ23 > 0, is equivalent to

λ13 , λ23 > 0.

Hence, (4) is equivalent to

λ13 , λ23 > 0, λ13 λ23 > λ̃2 .
Finally, (2) is equivalent to (i) and (ii), since it is equivalent to
1
λ12 + (k − 1)(λ13 + λ23 ) > 0
2
and 2
(k − 1)2 

1 
λ12 + (k − 1)(λ13 + λ23 ) > (λ13 − λ23 )2 + 4λ̃2 .
2 4
A calculation now shows that the second inequality is equivalent to (ii).

Remark 2.2. By adapting the arguments in the proof of Proposition 2.1, one can also obtain
equivalent characterizations in the cases k = 1, n − 2, n − 1. We omit this as it is not needed
in this article.

3 Perelman’s neck construction

In this section we prove the following result, which is the main ingredient in the proof of
Theorem A.

6
Theorem 3.1. For any ν > 0 sufficiently small, ℓ ∈ N, n ≥ 3 and all k ≥ 2 there exists a
metric of Rick > 0 on S n \⊔ℓ (D n )◦ such that the induced metric on each boundary component
is the round metric of radius one and the principal curvatures are all given by −ν.

The construction of the metric in Theorem 3.1 follows that of [Pe] and consists of two
parts: First, the ambient space, which is a metric of positive sectional curvature on S n \
⊔ℓ (D n )◦ , where the metric on each boundary component is a warped product metric whose
’waist’ can be chosen arbitrarily small and with principal curvatures all at least -1. It is
already established in [Pe] that the metric has positive sectional curvature. Second, the
neck, which is a metric on S n−1 × [0, 1] connecting the metrics on the boundary components
of the ambient space to round metrics with constant and arbitrarily small second fundamental
form. This metric on the neck is shown to have positive Ricci curvature in [Pe] and we show
below that it has in fact Ric2 > 0:

Proposition 3.2. Let g be a metric on S n , n ≥ 2, of the form

g = dt2 + B 2 (t)dsn−1 ,

where t ∈ [0, πR], and we set r = maxt B(t). Assume that g has sectional curvatures greater
1
than 1 and suppose that r < R2 . Let ρ ∈ (r 2 , R). Then there exists a metric of Ric2 > 0 on
S n × [0, 1] such that
ρ
(i) The induced metric on S n × {0} is the round metric of radius λ
and satisfies II ≡ −λ
for some λ > 0,

(ii) The induced metric on S n × {1} is isometric to g and satisfies II > 1.

The metric we will construct in the proof of Proposition 3.2 is of the form dt2 +A(t, x)2 dx2 +
B(t, x)2 ds2m , where dx2 denotes the standard metric on S 1 . We first compute the curvatures
of such a metric.

Lemma 3.3. Let t0 < t1 and denote by dt2 the standard metric on [t0 , t1 ], and by dx2 the
standard metric on S 1 . Let A, B : [t0 , t1 ] × S 1 → R>0 be smooth positive functions and define
the metric g on [t0 , t1 ] × S 1 × S m by

g = dt2 + A(t, x)2 dx2 + B(t, x)2 ds2m .

Let v1 , v2 denote vectors tangent to S m . Then the curvature tensor of g is given by

Att
R(∂t ∧ ∂x ) = − ∂t ∧ ∂x ,
A  
Btt Bxt At Bx
R(∂t ∧ v1 ) = − ∂t ∧ v1 + − 2 + 3 ∂x ∧ v1 ,
B A B A B
   
Bxt At Bx At Bt Bxx Ax Bx
R(∂x ∧ v1 ) = − + ∂t ∧ v1 + − − 2 + 3 ∂x ∧ v1 ,
B AB AB AB A B
1 − Bt2 Bx2
 
R(v1 ∧ v2 ) = − 2 2 v1 ∧ v2 .
B2 A B

7
Proof. By using the Koszul formula one easily verifies that the Levi-Civita connection of g
is given by

∇∂t ∂t =0,
At
∇∂t ∂x =∇∂x ∂t = ∂x ,
A
Bt
∇∂t v1 =∇v1 ∂t = v1 ,
B
Ax
∇∂x ∂x = − AAt ∂t + ∂x ,
A
Bx
∇∂x v1 =∇v1 ∂x = v1 ,
B  
BBx m
∇v1 v2 = − ds2m (v1 , v2 ) BBt ∂t + 2 ∂x + ∇Sv1 v2 .
A

From this one can now calculate the full curvature tensor.
Proof of Proposition 3.2. We use the same metric as constructed in [Pe, Section 2]. This
metric is constructed as follows.
We rewrite the metric g as

g = r 2 cos2 (x)ds2n−1 + A2 (x)dx2 ,

x ∈ [− π2 , π2 ], where A satisfies A(± π2 ) = r, A′ (± π2 ) = 0. Then, since

Z π
2
A(x)dx = πR,
− π2

there exists x ∈ [− π2 , π2 ] with A(x) ≥ R (> r), (note that R < 1 by the theorem of Bonnet–
Myers). Hence, we can rewrite A as

A(x) = r(1 − η(x) + η(x)a∞ ),

where η is a function satisfying maxx η(x) = 1, η(± π2 ) = 0 and η ′ (± π2 ) = 0, and a∞ ∈ R

with a∞ ≥ Rr .
For t0 < t∞ we define the metric gt0 ,t∞ on S n × [t0 , t∞ ] by

gt0 ,t∞ = dt2 + A2 (t, x)dx2 + B 2 (t, x)ds2n−1 ,

where
B(t, x) = tb(t) cos(x), A(t, x) = tb(t)(1 − η(x) + η(x)a(t))
and a, b are functions satisfying a(t0 ) = 1, a′ (t0 ) = 0, b(t0 ) = ρ, b′ (t0 ) = 0, a(t∞ ) = a∞ > 1
r
and b(t∞ ) > r. This metric will later be rescaled by t∞ b(t ∞)
to satisfy the required properties.

8
 Using Lemma 3.3 we see that the curvature operator Rgt0 ,t∞ of this metric has the form
of the map A in Proposition 2.1 with
Att
λ12 = − = K(∂t ∧ ∂x ),
A
Btt
λ13 = − = K(∂t ∧ v),
B
Bxt At Bx 1
λ̃ = − 2 + 3 = Ric(∂t , ∂x ),
A B A B n−1
At Bt Bxx Ax Bx
λ23 = − − 2 + 3 = K(∂x ∧ v),
AB A B AB
1 − Bt2 Bx2
λ3 = − 2 2 = K(v1 ∧ v2 ).
B2 A B
Here v, v1 , v2 are tangent to S n .
The functions a and b are now explicitly defined by

b′ β(t − t0 )
=− 2 , t0 ≤ t ≤ 2t0 ,
b 2t0 ln(2t0 )
b′ β ln(2t0 )
=− , t ≥ 2t0 ,
b t ln(t)2
a′ b′
=−α , t ≥ t0 .
a b
The constants α and β are defined by

ln(ρ) − ln(r)
β =(1 − ε) 1 ,
1 + 4 ln(2t0)

(1 + δ) ln(a∞ ) (1 + δ) ln(a∞ )
α= 1 =
β 1 + 4 ln(2t0 ) (1 − ε) ln(ρ/r)
R∞ R∞
for some ǫ, δ > 0 small. These values imply that t0 b′ /b = (1 − ǫ)(ln r − ln ρ) and t0 a′ /a =
(1 + δ) ln a∞ .
Similarly as in [Pe] we estimate α as follows: At a maximum point of η we have η(x) = 1
and η ′ (x) tan(x) = 0. Hence, the sectional curvatures of g at this point satisfy (e.g. by
applying Lemma 3.3)

sin(x)A′ (x)
 
1 1
Kg (∂x ∧ v) = 2
1− = 2 2 .
A(x) cos(x)A(x) r a∞

Since Kg > 1, it follows that a∞ < 1/r. Thus,

   2
1 1ρ ρ
ln(a∞ ) < ln < ln = 2 ln .
r r r r

Hence, ln(a∞ )/ ln(ρ/r) < 2.

9
 We also have a∞ ≥ R/r > ρ/r, so that ln(a∞ ) > ln(ρ/r). Hence, for ǫ and δ sufficiently
small, we have α ∈ (1, 2).
By choosing ǫ smaller if necessary, we can assume that (ρ/r)ǫ g still has sectional cur-
vatures at least 1. The following estimates are now established in [Pe, Section 2] for t0
sufficiently large (see also [Bu1, Lemma 2.6, Corollary C.2.9 and Corollary C.3.3]).
c1
λ23 , λ3 ≥ ,
t2
c2 ln(t0 )
|λ12 |, |λ13 |, |λ̃| ≤ 2
t ln(t)2

for some c1 , c2 > 0. To estimate λ13 a calculation now shows that

 ′′
2b′

b c3 ln(t0 )
λ13 = − + ≥ 2
b tb t ln(t)2

for some c3 > 0.

By Proposition 2.1 we need to satisfy the following inequalities:
1
λ12 + (λ13 + λ23 ) > 0, (6)
2
(λ12 + λ13 )(λ12 + λ23 ) > λ̃2 , (7)
λ13 λ23 > λ̃2 , (8)
λ13 , λ23 , λ3 > 0. (9)

From the above estimates it follows directly that (6), (8) and (9) are satisfied for t0 sufficiently
large. For (7) we show that
c4 ln(t0 )
λ12 + λ13 > 2
t ln(t)2
for some c4 > 0, from which (7) follows. We calculate
  ′ ′  ′ 2  ′ 2 !
2b′ a′ b′
 
αηa b b ηa a
λ12 + λ13 = −2 + −2 − 2 + .
1 − η + ηa b tb b 1 − η + ηa ab a

Similarly as in [Pe, end of p. 161] we see that, since α < 2 and η ≤ 1, the first factor in the
first summand is negative and uniformly bounded from above. Hence, the first summand
is bounded from below by ct25 ln(t)
ln(t0 )
2 for some c5 > 0 and the absolute value of the remaining

terms is bounded from above by ct26 ln(t)

ln(t0 )
4 for some c6 > 0. It follows that the required estimate

holds for t0 sufficiently large. Thus, the metric has Ric2 > 0 for t0 sufficiently large.
Note that δ can still be chosen freely (which then determines t∞ via a(t∞ ) = a∞ ). This
is now done as in [Pe] to ensure that the required conditions on the principal curvatures are
satisfied.
We can now give the proof of Theorems 3.1 and A. For this, we recall the following gluing
theorem which was established in [RW3].

10
Theorem 3.4 ([RW3, Theorem A]). Let (M1n , h1 ) and (M2n , h2 ) be Riemannian manifolds
of Rick > 0 for some 1 ≤ k ≤ n − 1 with compact boundaries, and let φ : (∂M1 , h1 |∂M1 ) →
(∂M2 , h2 |∂M2 ) be an isometry. If the sum of second fundamental forms II∂M1 + φ∗ II∂M2 is
positive semi-definite, then M1 ∪φ M2 admits a smooth metric of Rick > 0 which coincides
with the C 0 -metric h = h1 ∪φ h2 outside an arbitrarily small neighbourhood of the gluing area.

We will also need the following result of Perelman:

Proposition 3.5 ([Pe, Section 3]). For every n ≥ 3, ℓ ≥ 0, R0 ∈ (0, 1) and r > 0 sufficiently
small there exists a metric g on S n \ ⊔ℓ (D n )◦ such that

(i) g has positive sectional curvature,

(ii) The induced metric on each boundary component is of the form dt2 + B(t)2 ds2n−2 with
4 /4)
t ∈ [0, π cos(r)] and maxt B(t) = cos(r)R0 sin(r+r
sin(r)
, and has sectional curvature at
least 1, and

(iii) The principal curvatures at each boundary are all at least -1.

Proof of Theorem 3.1. We equip S n \ ⊔ℓ (D n )◦ with the metric provided by Proposition 3.5,
4)
where R0 is so small that R0 < ν 2 , and r is so small so that cos(r) > ν and cos(r)R0 sin(r+r
sin(r)
<
2
ν . Hence, using Theorem 3.4, we can glue a copy of the neck obtained in Proposition 3.2
to each of the ℓ boundary components of S n \ ⊔ℓ (D n )◦ to obtain a metric of Ric2 > 0 on the
resulting manifold. Note that cos(r) in Proposition 3.5 corresponds to R in Proposition 3.2
4)
and cos(r)R0 sin(r+r
sin(r)
in Proposition 3.5 corresponds to r in Proposition 3.2, and we choose
ρ = ν. Finally, we rescale the metric by λρ so that the induced metric on each boundary
component is the round metric of radius 1 and the principal curvatures are all given by
−ρ = −ν.
Proof of Theorem A. The proof is essentially similar to the proof of [Bu2, Theorem B]. We
denote by ϕi : D n ֒→ Mi the embedding provided by Definition 1.2. We now slightly perturb
the k-core metric on each Mi \ ϕi (D n )◦ , e.g. as in [Bu1, Proposition 1.2.11], such that the
second fundamental form is strictly positive. Let ν0 > 0 be the smallest principal curvature
of all these metrics. Thus, by Theorem 3.4, we can glue each Mi \ ϕi (D n )◦ to S n \ ⊔ℓ (D n )◦
with the metric provided by Theorem 3.1 by choosing ν < ν0 . Hence, we obtain a metric of
Rick > 0 on the connected sum M1 # . . . #Mℓ .

4 k-core metrics
In this section we consider k-core metrics. We begin by restating Proposition 1.4.

Proposition 4.1. Let M be a closed n-dimensional manifold that admits a k-core metric.
Then M is (n − k)-connected. In particular, if k ≤ ⌊ n+1
2
⌋, then M is a homotopy sphere.

11
Proof. Since the boundary of M \ϕ(D n )◦ has positive semi-definite second fundamental form,
it follows from [Wu, Theorem 1] that M \ ϕ(D n )◦ is obtained from ϕ(S n−1) by attaching cells
of dimension at least n − k + 1. By viewing ϕ(D n ) as a 0-cell, we obtain a CW structure for
M with no cells in dimensions between 1 and n − k. It follows that M is (n − k)-connected.
Now if k ≤ ⌊ n+1
2
⌋, we obtain by Poincaré duality that M is a closed simply-connected
manifold with non-trivial homology groups only in degrees 0 and n. Hence, M is a homotopy
sphere.
We will now consider examples of manifolds with k-core metrics and applications to
plumbing.

4.1 Projective spaces

To prove Theorem B, we will adapt the construction in [Ch], where a metric of non-negative
sectional curvature and round totally geodesic boundary is constructed on CP n , HP n and
OP 2 with a disc removed. We will follow [BM, Sections 3 and 4] and also include the
arguments for CP n as they are entirely similar.
The key observation is that, by considering cohomogeneity-one actions on these spaces,
they can all be written as a disc bundle G ×K D → G/H, where H ⊆ K ⊆ G are compact
Lie groups. Here K acts by isometries on a Euclidean vector space V with principal isotropy
group H via a representation ρ : K → O(V ) , and D ⊆ V is the unit disc. The corresponding
groups are given as follows, see [AB, Section 6.3], [BM, Section 4.1] and [Iw, Example 1].

G K H
◦
CP n \ D 2n U(n) U(n − 1)U(1) U(n − 1)
4n ◦
HP n \ D Sp(n) Sp(n − 1)Sp(1) Sp(n − 1)
◦
OP 2 \ D 16 Spin(9) Spin(8) Spin(7)

Table 1: Cohomogeneity one structure of projective spaces with a disc removed.

The representation ρ is given by projection onto U(1) (resp. Sp(1)) followed by inclusion
into O(2) (resp. O(4)) for CP n (resp. HP n ). For OP 2 it is given by the covering map
Spin(8) → SO(8).
We will construct a k-core metric on G ×K D by defining a K-invariant metric on G × D,
which then descends to G ×K D such that the projection G × D → G ×K D is a Riemannian
submersion. On G × D we consider the metric

g = L + (dt2 + f (t)2 ds2m ),

where m = dim(V ) − 1, L is a left-invariant metric on G which is AdK -invariant and

f : [0, t0 ] → R≥0 is a smooth function for some t0 > 0 which is odd at t = 0 with f ′ (0) = 1
and f (t) > 0 for t ∈ (0, t0 ].
Let g = k ⊕ m and k = h ⊕ p be L-orthogonal decompositions of the Lie algebras. For
X ∈ k, t ∈ [0, t0 ] and v ∈ S m we denote by Xtv
∗
the action field at tv ∈ D defined by X, i.e.

12
 d
∗


Xtv = ds ρ expK (sX) (tv)|s=0 . Then the vertical and horizontal subspaces V(e,tv) and H(e,tv)
of T(e,tv) (G × D) with respect to g are given for t > 0 by
∗
V(e,tv) = (h ⊕ {0}) ⊕ {(−X, Xtv ) | X ∈ p},
2
H(e,tv) = (m ⊕ {0}) ⊕ {(f (t) BY, Ytv∗ ) | Y ∈ p} ⊕ h∂t i, (10)

where B : p → p is the L-symmetric and AdH -linear automorphism defined by L(X, BY ) =

ds2m (Xtv
∗
, Ytv∗ ), cf. [BM, Equation (3.1)]. For t = 0 we have

V(e,0) = k ⊕ {0},
H(e,0) = m ⊕ T0 D. (11)

With this description we can now give the proof of Theorem B.

Proof of Theorem B. We equip G with the left-G-invariant and right-K-invariant which in-
duces the round metric on G/H (note that this metric does not need to be normal homoge-
neous). For CP n (resp. HP n ) the restriction to U(1) (resp. Sp(1)) is then biinvariant, hence
it is the round metric of some radius. In particular, the map B is a multiple of the identity
map. For OP 2 , the action of H on p is irreducible, so B is also a multiple of the identity
map by Schur’s Lemma.
Hence, there exists b ∈ R so that B = b · Idp . For ǫ > 0 we now define the metric Lǫ on
G via
Lǫ = (1 + ǫ)L|k + L|m ,
1
so Lǫ is again left-G-invariant and right-K-invariant and the map Bǫ is given by 1+ǫ
b · Idp .
Then the metric
gǫ = Lǫ + (dt2 + f (t)2 ds2m )
on G × D induces a metric ǧǫ on G ×K D such that the projection G × D → G ×K D is a
Riemannian submersion. The metric induced on a slice G ×K S m = G ×K (K/H) ∼ = G/H
for t > 0 is then given by
b b
f (t)2 1+ǫ f (t)2 1+ǫ
b
Lǫ |p + Lǫ |m = (1 + ǫ) b
L|p + L|m ,
1 + f (t)2 1+ǫ 1 + f (t)2 1+ǫ
q
1+ǫ
see e.g. [Ch], [GZ], [BM, Lemma 3.1]. In particular, if f (t) = bǫ
, then this metric
coincides with the metric induced from L on G/H, i.e. it is the round metric. Thus, we will
assume from
q now on that for given ǫ, the function f (and the value of t0 ) is chosen such that
f (t0 ) = 1+ǫ
bǫ
, so that the induced metric on the boundary of G ×K D is round. Moreover,
we assume that f ′ (t0 ) ≥ 0, so the second fundamental form on the boundary is positive
semi-definite.
We will now analyse the curvatures of the metric ǧǫ on G×K D. We assume that f ′′ < 0, so
the metric hf = dt2 +f (t)2 ds2m on D has positive sectional curvature. We choose ǫ sufficiently
small such that the metric induced on G/H by Lǫ also has positive sectional curvature. It
then follows that the metric ǧǫ has non-negative sectional curvature, see [Ch], [BM, Lemma

13
4.1]. Thus, to determine the smallest value k for which this metric has Rick > 0, we only
need to identify the 2-planes of vanishing curvature, i.e. for given u ∈ T (G ×K D) we need
to determine the set
Zu = {v ∈ u⊥ \ {0} | secǧǫ (u ∧ v) = 0}.
Let A denote the A-tensor of the Riemannian submersion (G × D, gǫ) → (G ×K D, ǧǫ )
and decompose A into A = A1 + A2 according to the splitting (10), i.e. A1 has image in
h ⊕ {0} and A2 has image in {(−X, Xtv ∗
) | X ∈ p}. As in [BM, Proof of Lemma 4.1] we
conclude that for horizontal vectors u = (u1 , u2 ), v = (v1 , v2 ) in T(e,tv) (G × D) with t > 0, we
have
A1 u v = AG/H u1 v1 ,
where AG/H is the A-tensor of the Riemannian submersion G → G/H (where we consider
G equipped with the metric Lǫ ). It follows from the O’Neill formulas that
ǧǫ (Rǧǫ (u, v)v, u) = gǫ (Rgǫ (u, v)v, u) + 3|Au v|2
= Lǫ (RLǫ (u1 , v1 )v1 , u1 ) + hf (Rhf (u2 , v2 )v2 , u2 ) + 3|AG/H 2 2 2
u1 v1 | + 3|Au v|

= Lǫ (RG/H (u1 , v1 )v1 , u1 ) + hf (Rhf (u2 , v2 )v2 , u2) + 3|A2u v|2 .

Since both the metric on G/H and the metric hf have strictly positive sectional curvature,
this expression can only vanish if the pairs (u1 , v1 ) and (u2, v2 ) are both linearly dependant.
If we write, according to (10),
u = (u1 , u2 ) = (X + f (t)2 Bǫ Y, Ytv∗ + λ∂t ),
∗
v = (v1 , v2 ) = (X ′ + f (t)2 Bǫ Y ′ , Y ′ tv + λ′ ∂t ),
this is satisfied if and only if there exist a1 , a2 ∈ R such that
(X ′ , Y ′ ) = a1 (X, Y ) or (X, Y ) = (0, 0), and
(Y ′ , λ′ ) = a2 (Y, λ) or (Y, λ) = (0, 0).
If Y 6= 0, then a1 = a2 , hence v = a1 u and Zu is empty. Hence, we can assume that Y = 0.
Then, if X, λ 6= 0, we have X ′ = a1 X, λ′ = a2 λ and Y ′ = 0, hence Zu is contained in a
1-dimensional subspace. Thus, we are left with the cases X = 0, λ 6= 0 and X 6= 0, λ = 0. In
the first case, we have Y ′ = 0 and λ′ = a2 λ, so Zu is contained in a dim(G/K)-dimensional
subspace. In the second case we have Y ′ = 0 and X ′ = a1 X, so Zu is contained in a
1-dimensional subspace.
Hence, we have shown that Zu is contained in a dim(G/K)-dimensional subspace for all
u ∈ Te,tv (G ×K D) and t > 0. By G-invariance of the metric ǧǫ this holds for all points (g, tv)
with t > 0. Similar arguments using (11) show that this result extends to the case t = 0.
Thus, the metric ǧǫ has Ricdim(G/K)+1 > 0. For CP n this gives a metric of Ric2n−1 > 0, for
HP n a metric of Ric4n−3 > 0 and for OP 2 a metric of Ric9 > 0.

4.2 Generalized surgery and plumbing

To prove Theorem D we need two additional results: A surgery result extending [RW2,
Theorem 3.2] and [Re1, Theorem A] and a deformation result that ensures that we can

14
satisfy the assumptions of the surgery theorem in our setting. For ρ > 0 we denote by S p (ρ)
q+1
the round sphere of radius ρ and for R, N > 0 we denote by DR (N) a geodesic ball of
q+1
radius R in S (N).

Theorem 4.2. Suppose we have the following:

(i) A Riemannian manifold (M p+q+1 , gM ) of Rick1 > 0,

q+1
(ii) An isometric embedding ι : S p (ρ) × DR (N) ֒→ (M, gM ), (which implies k1 ≥ max(p +
1, q + 2)),
π
(iii) A linear S q -bundle E −
→ B p+1 , where B is compact and admits a k2 -core metric gB .

Then, if p, q ≥ 2, for any r > 0 sufficiently small, there exists a constant κ = κ(p, q, R/N, gB , r),
such that if Nρ < κ, then the manifold

Mι,π = M \ im(ι)◦ ∪∂ π −1 (B \ ϕ(D p+1)◦ )

admits a metric of Rick > 0 for all k ≥ max(p + 2, q + 2, q + k2 ). This metric coincides
outside the gluing area with a submersion metric on E with totally geodesic round fibres of
radius r and a scalar multiple of the metric gM on M.

Proof. We equip E with a submersion metric with totally geodesic and round fibres of radius
r according to a horizontal distribution which is integrable over ϕ(D p+1 ) ⊆ B. Then, for
r sufficiently small, this metric has Rick > 0 for all k ≥ max(p + 2, q + k2 ) by [RW2,
Corollary 3.1]. Further, over ϕ(D p+1 ), the metric is a product, in particular it is given over
ϕ(S p ) by ds2p + r 2 ds2q . As noted below Definition 1.2, we can slightly deform the metric on
π −1 (B \ ϕ(D p+1)◦ ) so that the induced metric on the boundary remains unchanged and the
second fundamental form on the boundary is strictly positive.
Now, by [RW1, Theorem C and Remark 4.2] there exists a metric of Rick > 0 on the
manifold
Mι = M \ im(ι)◦ ∪∂ (D p+1 × S q )
for all k ≥ max(p, q) + 2 such that the metric near the centre of D p+1 × S q is given by
p+1 R′
DR ′ q ′ ′ ′ ′
′ (N ) × S (ρ ), where the values of R , N , ρ can be chosen freely (provided
N′
< π2 ). We
p+1 2
choose ρ′ = r and R′ , N ′ so that the induced metric on ∂DR ′
′ (N ) is dsp and the principal

curvatures at the boundary are at least −ǫ for given ǫ > 0. (Note that they converge to 0
R′ π
as N ′ → 2 ).
p+1 q ′
It follows that DR ′
′ (N ) × S (ρ ) and π
−1
(B \ ϕ(D p+1)◦ ) have isometric boundaries, and
for ǫ sufficiently small the principal curvatures of π −1 (B \ ϕ(D p+1 )◦ ) at the boundary are
p+1 q ′ p+1
greater than those of DR ′
′ (N ) × S (ρ ). Hence, by Theorem 3.4, we can replace D × Sq
−1 p+1 ◦
in Mι by π (B \ ϕ(D ) ) to construct Mι,π while preserving Rick > 0.
To satisfy condition (ii) of Theorem 4.2, we need the following deformation result, which
generalizes [Wr2, Theorem 1.10].

15
Lemma 4.3. Let (M n , g0 ) be a Riemannian manifold of Rick > 0 and let N p ⊆ M be
a compact embedded submanifold. Let g1 be a metric of Rick > 0 defined in a tubular
neighbourhood U of N. If the 1-jets of g0 and g1 on N coincide, then there exists a metric g̃
of Rick > 0 on M that equals g0 outside U and equals g1 on a (smaller) tubular neighbourhood
of N.

Proof. We consider for t ∈ [0, 1] the metric gt = (1 − t)g0 + tg1 on U. Since the 1-jets of
g0 and g1 coincide on N and since the sectional curvatures depend linearly on the second
derivatives of the metric, we have Kgt = (1 − t)Kg0 + tKg1 on N. In particular, gt has
Rick > 0 on N and by compactness this holds in a neighbourhood of N. By [BH, Theorem
1.2] the local deformation gt can now be extended to a global deformation of g0 , which leaves
g0 unchanged outside a neighbourhood of N and coincides with the deformation gt on a
(smaller) tubular neighbourhood of N.

Corollary 4.4. Let (M n , g) be a Riemannian manifold of Rick > 0 and let p1 , . . . , pℓ ∈ M.

Then the metric g can be deformed into a metric of Rick > 0 that has constant sectional
curvature 1 in a neighbourhood of each pi .

Proof. We consider normal coordinates around each pi , i.e. coordinates (x1 , . . . , xn ) in which
the metric is given by gab = δab + O(r 2), where r denotes the distance to pi . In particular, the
first derivatives ∂c gab all vanish at pi . By considering normal coordinates at a point in the
round sphere of radius 1, we obtain a second metric around each pi with the same property.
Applying Lemma 4.3 now yields the required deformation.
Proof of Theorem D. The proof of Theorem D follows the same lines as the proof of [Re1,
Theorem B] by observing that ∂W is obtained by iterated generalized surgeries as in Theorem
4.2. We simply replace [Re1, Theorem A] by Theorem 4.2, [Re1, Proposition 2.2] by [RW2,
Corollary 3.1] and the deformation result used in the proof of [Re1, Theorem B] by Corollary
4.4.
Proof of Corollary E. Let E i → Bi denote the disc bundle corresponding to Ei → Bi . We
define W as the manifold obtained by plumbing according to the following graph, where we
denote by D m m
M the trivial disc bundle M × D → M over a manifold M.

D p+1
Sq D p+1
Sq

E1 D qS p+1 E2 D qS p+1 ... Eℓ

By [Re2, Propositions 3.2 and 3.3], see also [CW, Proposition 2.6] and [Bu2, Section
5], the manifold ∂W is diffeomorphic to the connected sum E1 # . . . #Eℓ . By Theorem D,
the manifold ∂W admits a metric of Rick > 0 for all k ≥ max{p + 2, p + k, q + 1, q} =
max{p + 2, p + k, q + 1}.

16
Remark 4.5. In [Bu3] it is shown that the manifolds constructed in Theorem D and Corollary
E admit a core metric, provided each base manifold of the bundles involved admits a core
metric. We conjecture that these manifolds in fact admit k-core metrics with k as given in
these results. However, this conjecture is open even in the simplest case of a linear sphere
bundle over a manifold with a core metric (which can be viewed as a plumbing according to
a graph with a single vertex).

References
[AB] M. Alexandrino, R. Bettiol, Lie groups and geometric aspects of isometric actions,
Springer, Cham (2015), x+213 pp.

[Bu1] B. L. Burdick, Metrics of Positive Ricci Curvature on Connected Sums: Projec-

tive Spaces, Products, and Plumbings, ProQuest LLC, Ann Arbor, MI (2019). Thesis
(Ph.D.)–University of Oregon.

[Bu2] B. L. Burdick, Ricci-positive metrics on connected sums of projective spaces, Differen-

tial Geom. Appl. 62 (2019), 212–233.

[Bu3] B. L. Burdick, Metrics of positive Ricci curvature on the connected sums of products
with arbitrarily many spheres, Ann. Global Anal. Geom. 58(4) (2020), 433–476.

[BH] C. Bär, B. Hanke, Local flexibility for open partial differential relations, Comm. Pure
Appl. Math. 75 (2022), no.6, 1377–1415.

[BM] R. Bettiol, R. Mendes, Strongly nonnegative curvature, Math. Ann. 368 (2017), no.
3–4, 971–986.

[Ch] J. Cheeger, Some examples of manifolds of nonnegative curvature, J. Differential Ge-

ometry 8 (1973), 623–628.

[CW] D. Crowley, D. J. Wraith, Positive Ricci curvature on highly connected manifolds, J.

Differential Geom. 106 (2017), no.2, 187–243.

[Gr] M. Gromov, Curvature, diameter and Betti numbers, Comm. Math. Helv. 56 (1981),
179–195.

[GL] M. Gromov, H. B. Lawson, The classification of simply connected manifolds of positive

scalar curvature, Ann. of Math. (2) 111 (1980), 423–434.

[GZ] K. Grove, W. Ziller, Curvature and symmetry of Milnor spheres, Ann. of Math. (2)
152 (2000), no.1, 331–367.

[Iw] K. Iwata, Compact transformation groups on rational cohomology Cayley projective

planes, Tohoku Math. J. (2) 33 (1981), no.4, 429–442.

[Re1] P. Reiser, Generalized surgery on Riemannian manifolds of positive Ricci curvature,

Trans. Amer. Math. Soc. 376 (2023), no.5, 3397–3418.

17
[Re2] P. Reiser, Metrics of Positive Ricci Curvature on Simply-Connected Manifolds of Di-
mension 6k, arXiv:2210.15610 (2022).

[RW1] P. Reiser, D. J. Wraith, Intermediate Ricci curvatures and Gromov’s Betti number
bound, J. Geom. Anal. 33 (2023), 364.

[RW2] P. Reiser, D. J. Wraith, Positive intermediate Ricci curvature on fibre bundles,

arXiv:2211.14610 (2022).

[RW3] P. Reiser, D. J. Wraith, A generalization of the Perelman gluing theorem and appli-
cations, arXiv:2308.06996 (2023).

[Pe] G. Perelman, Construction of manifolds of positive Ricci curvature with big volume and
large Betti numbers, Comparison Geometry, MSRI publications 30 (1997), 157–163.

[Wr1] D. J. Wraith, Exotic spheres with positive Ricci curvature, J. Differential Geom. 45
(1997), no.3, 638–649.

[Wr2] D. J. Wraith, Deforming Ricci positive metrics, Tokyo J. Math. 25 (2002), no.1, 181–
189.

[Wu] H. Wu, Manifolds of partially positive curvature, Indiana Univ. Math. J. 36(3) (1987),
525–548.

Philipp Reiser, Department of Mathematics, University of Fribourg, Switzerland, Email:

philipp.reiser@unifr.ch

David Wraith, Department of Mathematics and Statistics, National University of Ireland

Maynooth, Maynooth, County Kildare, Ireland. Email: david.wraith@mu.ie.

18