Show Summary Details
More options …

Editor-in-Chief: Radulescu, Vicentiu / Squassina, Marco

IMPACT FACTOR 2017: 4.674

CiteScore 2018: 5.03

SCImago Journal Rank (SJR) 2018: 3.215
Source Normalized Impact per Paper (SNIP) 2018: 3.225

Mathematical Citation Quotient (MCQ) 2017: 1.89

Open Access
Online
ISSN
2191-950X
See all formats and pricing
More options …
Volume 7, Issue 1

# On the bifurcation of solutions of the Yamabe problem in product manifolds with minimal boundary

Elkin Dario Cárdenas Diaz
/ Ana Cláudia da Silva Moreira
• Corresponding author
• Instituto de Matemática e Estatística, Universidade de São Paulo, 05508-090 São Paulo, Brazil
• Email
• Other articles by this author:
Published Online: 2016-01-19 | DOI: https://doi.org/10.1515/anona-2015-0170

## Abstract

In this paper, we study the multiplicity of solutions of the Yamabe problem on product manifolds with minimal boundary via bifurcation theory.

Keywords: Yamabe problem; bifurcation theory

MSC 2010: 53C20; 58E11; 58J32

## 1 Introduction

Geometers have been interested in finding canonical metrics on a Riemannian manifold for years. Probably, the best known problem on this topic is the uniformization problem for compact surfaces, which assures that there exists a metric of constant Gaussian curvature in each conformal class. For higher dimensions, a Japanese mathematician named Hidehiko Yamabe proposed the following question, that later came to be known as the Yamabe problem:

Let $\left(M,{g}_{0}\right)$ be a compact Riemannian manifold (without boundary) of dimension $m\ge 3$. Is there a metric g, conformal to ${g}_{0}$, having constant scalar curvature?

In 1960, Yamabe [17] published an article with a proof of the statement. But in 1968, Trudinger [16] found an error in Yamabe’s proof and gave a complete argument for the existence of a solution to the problem for non-positive scalar curvature. Only in 1989, the combined work of Aubin [2], Trudinger [16], and Schoen [13, 14] lead to a complete proof of the existence of a solution for the Yamabe problem in its full generality. It is known that the critical points of the Hilbert–Einstein functional F, restricted to the set of metrics on M having unit volume, are the Einstein metrics of unit volume on M, and the critical points of F, restricted to the metrics conformal to g with unit volume, are the constant scalar curvature metrics of unit volume.

From then on, many questions have been raised on the uniqueness and the multiplicity of solutions, on the compactness and the non-compactness of the set of solutions, and so on. Moreover, variations of the problem have been proposed by several authors. Nowadays, a great number of results on the Yamabe problem and its generalizations can be found in the literature. Our starting point originates from an article of de Lima, Piccione, and Zedda [12], where the authors studied the local rigidity and the multiplicity of constant scalar curvature metrics on compact product manifolds using a bifurcation result. Given a family ${\left\{{g}_{s}\right\}}_{s\in \left[a,b\right]}$ of solutions of the Yamabe problem on the product manifold, the existence of a bifurcation instant ${s}_{*}$ gives an entirely new sequence ${\left({g}_{n}\right)}_{n>0}$ of solutions. They proved that if ${g}_{{s}_{*}}$ is a degenerate critical point of F, restricted to the conformal class (with unit volume), then ${s}_{*}$ is a candidate for a bifurcation instant. In fact, except for a finite number of instants s for which ${g}_{s}$ is degenerate, all the rest are bifurcation instants.

Bifurcation techniques for the Yamabe problem have been used in the case of collapsing Riemannian manifolds and in the case of homogeneous metrics on spheres, see [4, 3]. There are several possible formulations of the Yamabe problem on manifolds $\left(M,\overline{g}\right)$ with boundary. Here, we consider the following:

Let $\left(M,\overline{g}\right)$ be a compact Riemannian manifold with boundary of dimension $m\ge 3$. The Yamabe problem in $\left(M,\overline{g}\right)$ consists in finding a metric $\stackrel{~}{g}$, conformal to $\overline{g}$, for which M has constant scalar curvature and $\partial M$ has vanishing mean curvature.

The question of existence and some other aspects of solutions for the above problem have been studied, for instance, in [1, 5, 8, 7, 9]. Writing a metric conformal to $\overline{g}$ as $\stackrel{~}{g}={u}^{\frac{4}{n-2}}\overline{g}$, then $\stackrel{~}{g}$ is a solution with constant scalar curvature and vanishing mean curvature of the Yamabe problem on manifolds with boundary if and only if the function u is a solution of the Neumann problem

where K is constant, ${\mathrm{\Delta }}_{\overline{g}}$ is the Laplacian operator of $\overline{g}$, ${R}_{\overline{g}}$ is the scalar curvature of $\overline{g}$, ${H}_{\overline{g}}$ is the mean curvature of $\partial M$ relative to $\overline{g}$, and ${\eta }_{\overline{g}}$ is the $\overline{g}$-unit normal field along $\partial M$ pointing inside of M. Solving the above problem is equivalent to finding critical points for the Hilbert–Einstein functional $F:{\mathcal{ℳ}}^{k,\alpha }\left(M\right)\to ℝ$, defined by

$F\left(g\right)={\int }_{M}{R}_{g}{\omega }_{g},$

restricted to the set of metrics in the conformal class of $\overline{g}$ having unit volume.

The aim of this paper is to determine the multiplicity of solutions of the Yamabe problem in manifolds obtained as a product of a compact manifold (without boundary) and a compact manifold with boundary, using bifurcation theory. We consider a setup similar to the problem studied in [12]. We consider here the case of manifolds with boundary, which introduces new elements with respect to the theory developed in [12]. Given a compact Riemannian manifold $\left({M}_{1},{g}^{\left(1\right)}\right)$ with $\partial {M}_{1}=\mathrm{\varnothing }$ and a compact Riemannian manifold $\left({M}_{2},{\overline{g}}^{\left(2\right)}\right)$ with minimal boundary, both having constant scalar curvature, consider the product manifold $M={M}_{1}×{M}_{2}$, whose boundary is given by $\partial M={M}_{1}×\partial {M}_{2}$. Let ${m}_{1}$ and ${m}_{2}$ be the dimensions of ${M}_{1}$ and ${M}_{2}$, respectively, and assume that $\mathrm{dim}\left(M\right)=m={m}_{1}+{m}_{2}\ge 3$. For each $s\in \left(0,+\mathrm{\infty }\right)$, define a family ${\overline{g}}_{s}={g}^{\left(1\right)}\oplus s{\overline{g}}^{\left(2\right)}$ of metrics on M. Then, $\left(M,{\overline{g}}_{s}\right)$ has constant scalar curvature and the mean curvature of $\partial M$ relative to ${\overline{g}}_{s}$ vanishes for all $s>0$.

The main difficulty faced in the transition from the case without boundary to the case with boundary was to find a result on uniqueness (to ensure rigidity) that can be well adapted to the case with boundary. Several results on uniqueness are known. Escobar [6], for instance, proved that if there is a metric $\stackrel{~}{g}$ with constant scalar curvature and vanishing mean curvature in the ${\mathcal{𝒞}}^{k,\alpha }$-conformal class of an Einstein metric $\overline{g}$, then $\stackrel{~}{g}$ is Einstein. Moreover, if $\overline{g}$ is not conformal to the round metric, then $\stackrel{~}{g}$ is unique, except for homothety. Here, we adapt the result proved by de Lima, Piccione, and Zedda [11], which states that every ${\mathcal{𝒞}}^{k,\alpha }$-conformal class of metrics of unit volume has at most one metric of unit volume and constant scalar curvature in a neighborhood of a non-degenerate metric with constant scalar curvature.

The main results in this paper, Theorem 4.4 and Theorem 4.5, state that when the scalar curvature of each factor $\left({M}_{i},{g}^{\left(i\right)}\right)$, $i=1,2$, is positive, then there are two sequences (one tending to zero and the other to $+\mathrm{\infty }$) of instants ${s}_{*}\in \left(0,+\mathrm{\infty }\right)$ such that the corresponding metric ${\overline{g}}_{{s}_{*}}$ is the limit of a sequence of metrics of distinct solutions of the Yamabe problem. Precise definitions of these bifurcating branches of solutions will be given below. In particular, we have a multiplicity of solutions of the Yamabe problem on the normalized conformal classes of these new metrics. For all other values of s, the family is locally rigid, which means that, locally, the metrics of the family are the unique solutions of the Yamabe problem, up to homotheties.

The paper is organized as follows. In Section 2, we study the variational framework used for the bifurcation result. This framework is given by considering the Hilbert–Einstein functional restricted to the normalized conformal classes. In Section 3, we discuss some results about rigidity and bifurcation that are used to obtain the conclusions of this work. Finally, in Section 4, we verify that essentially the same results obtained by de Lima, Piccione, and Zedda [12] also remain valid in the case of manifolds with boundary.

## 2.1 Manifolds and conformal classes

Let $\left(M,\overline{g}\right)$ be an m-dimensional oriented compact Riemannian manifold with boundary $\partial M\ne \mathrm{\varnothing }$ for $m\ge 3$. As the metric $\overline{g}$ induces inner products and norms in all spaces of tensors on M and the Levi-Civita connection $\overline{\nabla }$ of $\overline{g}$ induces a connection in all vector spaces of tensor fields on M, the space ${\mathrm{\Gamma }}^{k}\left({T}^{*}M\otimes {T}^{*}M\right)$ of ${\mathcal{𝒞}}^{k}$-sections of the vector bundle ${T}^{*}M\otimes {T}^{*}M$ of symmetric $\left(0,2\right)$-tensors of class ${\mathcal{𝒞}}^{k}$ on M is a Banach space with the norm

$\parallel \tau {\parallel }_{{\mathcal{𝒞}}^{k}}=\underset{j=0,\mathrm{\dots },k}{\mathrm{max}}\left[\underset{p\in M}{\mathrm{max}}{\parallel {\overline{\nabla }}^{\left(j\right)}\tau \left(p\right)\parallel }_{\overline{g}}\right]$

and, therefore, it is a Banach manifold.

Given $k\ge 3$ and $\alpha \in \left(0,1\right]$, denote by ${\mathcal{ℳ}}^{k,\alpha }\left(M\right)$ the set of all Riemannian metrics of class ${\mathcal{𝒞}}^{k,\alpha }$ on M, in the sense that the coefficients of the metrics are ${\mathcal{𝒞}}^{k,\alpha }$-functions on M. The set ${\mathcal{ℳ}}^{k,\alpha }\left(M\right)$ is an open cone of ${\mathrm{\Gamma }}^{k,\alpha }\left({T}^{*}M\otimes {T}^{*}M\right)$, so it is a Banach manifold itself, and ${T}_{\overline{g}}{\mathcal{ℳ}}^{k,\alpha }\left(M\right)={\mathrm{\Gamma }}^{k,\alpha }\left({T}^{*}M\otimes {T}^{*}M\right)$ for a metric $\overline{g}\in {\mathcal{ℳ}}^{k,\alpha }\left(M\right)$.

Consider the open subset

${\mathcal{𝒞}}_{+}^{k,\alpha }\left(M\right)=\left\{\varphi \in {\mathcal{𝒞}}^{k,\alpha }\left(M\right):\varphi >0\right\}$

of the Banach space ${\mathcal{𝒞}}^{k,\alpha }\left(M\right)$. Now, for each $\overline{g}\in {\mathcal{ℳ}}^{k,\alpha }\left(M\right)$, define the ${\mathcal{𝒞}}^{k,\alpha }$-conformal class of $\overline{g}$ by

$\left[\overline{g}\right]:=\left\{\varphi \overline{g}:\varphi \in {\mathcal{𝒞}}_{+}^{k,\alpha }\left(M\right)\right\}.$

#### Proposition 2.1.

The ${\mathcal{C}}^{k\mathrm{,}\alpha }$-conformal class of a Riemannian metric $\overline{g}\mathrm{\in }{\mathcal{M}}^{k\mathrm{,}\alpha }\mathit{}\mathrm{\left(}M\mathrm{\right)}$ is an embedded submanifold of ${\mathcal{M}}^{k\mathrm{,}\alpha }\mathit{}\mathrm{\left(}M\mathrm{\right)}$.

#### Proof.

Given $\overline{g}\in {\mathcal{ℳ}}^{k,\alpha }\left(M\right)$, consider the injective map

${\mathcal{ℐ}}_{\overline{g}}:{\mathcal{𝒞}}_{+}^{k,\alpha }\left(M\right)\to {\mathcal{ℳ}}^{k,\alpha }\left(M\right),$$\varphi ↦\varphi \overline{g},$

whose differential ${\left(\mathrm{d}{\mathcal{ℐ}}_{\overline{g}}\right)}_{\varphi }:{\mathcal{𝒞}}^{k,\alpha }\left(M\right)\to {\mathrm{\Gamma }}^{k,\alpha }\left({T}^{*}M\otimes {T}^{*}M\right)$ is injective and has a left-inverse given by the linear bounded operator

${\mathcal{𝒥}}_{\overline{g}}:{\mathrm{\Gamma }}^{k,\alpha }\left({T}^{*}M\otimes {T}^{*}M\right)\to {\mathcal{𝒞}}^{k,\alpha }\left(M\right),$$h↦\frac{1}{m}{\mathrm{tr}}_{\overline{g}}h.$

Consequently,1 the image $\mathrm{Im}\mathrm{d}{\mathcal{ℐ}}_{\overline{g}}$ has a closed complement in ${\mathrm{\Gamma }}^{k,\alpha }\left({T}^{*}M\otimes {T}^{*}M\right)$ and $\mathrm{Im}{\mathcal{ℐ}}_{\overline{g}}=\left[\overline{g}\right]$ is an embedded submanifold of ${\mathcal{ℳ}}^{k,\alpha }\left(M\right)$. ∎

In particular, $\left[\overline{g}\right]$ is a Banach manifold with differential structure induced by ${\mathcal{𝒞}}^{k,\alpha }\left(M\right)$ and its tangent space is

${T}_{\overline{g}}\left[\overline{g}\right]=\left\{\psi \overline{g}:\psi \in {\mathcal{𝒞}}^{k,\alpha }\left(M\right)\right\},$

which can be identified with ${\mathcal{𝒞}}^{k,\alpha }\left(M\right)$.

For each $\overline{g}\in {\mathcal{ℳ}}^{k,\alpha }\left(M\right)$, denote by ${\mathrm{Ric}}_{\overline{g}}$ the Ricci curvature and by ${R}_{\overline{g}}$ the scalar curvature with respect to $\overline{g}$. Let ${\eta }_{\overline{g}}$ be the unit (inward) vector field normal to $\partial M$ and denote by ${H}_{\overline{g}}$ the mean curvature of the boundary, induced by $\overline{g}$. These are ${\mathcal{𝒞}}^{k-2,\alpha }$-functions, ${\mathrm{Ric}}_{\overline{g}}$ and ${R}_{\overline{g}}$ defined on M and ${H}_{\overline{g}}$ defined on $\partial M$. Let ${\omega }_{\overline{g}}$ be the volume form on M with respect to $\overline{g}$ and let ${\sigma }_{\overline{g}}$ be the volume form induced on $\partial M$.

The volume function on ${\mathcal{ℳ}}^{k,\alpha }\left(M\right)$ is defined as

$\mathcal{𝒱}\left(\overline{g}\right):={\int }_{M}{\omega }_{\overline{g}}.$

Observe that $\mathcal{𝒱}$ is smooth and that its differential is given by

${\left(\mathrm{d}\mathcal{𝒱}\right)}_{\overline{g}}\left(h\right)=\frac{1}{2}{\int }_{M}{\mathrm{tr}}_{\overline{g}}\left(h\right){\omega }_{\overline{g}}$(2.1)

for each $\overline{g}\in {\mathcal{ℳ}}^{k,\alpha }\left(M\right)$ and $h\in {T}_{\overline{g}}{\mathcal{ℳ}}^{k,\alpha }\left(M\right)$.

We define

${\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1}:=\left\{\overline{g}\in {\mathcal{ℳ}}^{k,\alpha }\left(M\right):\mathcal{𝒱}\left(\overline{g}\right)=1\right\}$

to be the subset of unit volume metrics in ${\mathcal{ℳ}}^{k,\alpha }\left(M\right)$.

#### Proposition 2.2.

${\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1}$ is a smooth embedded submanifold of ${\mathcal{M}}^{k\mathrm{,}\alpha }\mathit{}\mathrm{\left(}M\mathrm{\right)}$.

#### Proof.

Consider the smooth volume function $\mathcal{𝒱}$ defined on ${\mathcal{ℳ}}^{k,\alpha }\left(M\right)$. For $\overline{g}\in {\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1}$, set $h=\overline{g}$. Then, from (2.1) we get

${\left(\mathrm{d}\mathcal{𝒱}\right)}_{\overline{g}}\left(\overline{g}\right)=\frac{1}{2}{\int }_{M}{\mathrm{tr}}_{\overline{g}}\left(\overline{g}\right){\omega }_{\overline{g}}=\frac{m}{2}\mathcal{𝒱}\left(\overline{g}\right)=\frac{m}{2}\ne 0,$

that is, the differential is surjective. So, ${\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1}={\mathcal{𝒱}}^{-1}\left(1\right)$ is the inverse image of a regular value. Moreover, the kernel (the tangent space ${T}_{\overline{g}}{\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1}$)

$\mathrm{ker}\left\{{\left(\mathrm{d}\mathcal{𝒱}\right)}_{\overline{g}}\right\}=\left\{h\in {T}_{\overline{g}}{\mathcal{ℳ}}^{k,\alpha }\left(M\right):{\int }_{M}{\mathrm{tr}}_{\overline{g}}\left(h\right){\omega }_{\overline{g}}=0\right\}$

has a closed complementary space, so the result follows. ∎

Observe that if $\overline{g}\in {\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1}$, the conformal metric $\varphi \overline{g}$, for some $\varphi \in {\mathcal{𝒞}}_{+}^{k,\alpha }\left(M\right)$, is not in ${\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1}$, in general. Indeed,

$\mathcal{𝒱}\left(\varphi \overline{g}\right)={\int }_{M}{\omega }_{\varphi \overline{g}}={\int }_{M}{\varphi }^{\frac{m}{2}}{\omega }_{\overline{g}}.$

So, for each $\overline{g}\in {\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1}$, we define

${\left[\overline{g}\right]}_{1}=\left\{\varphi \overline{g}:\varphi \in {\mathcal{𝒞}}_{+}^{k,\alpha }\left(M\right),{\int }_{M}{\varphi }^{\frac{m}{2}}{\omega }_{\overline{g}}=1\right\},$

which is an embedded submanifold of $\left[\overline{g}\right]$. The proof is similar to that of Proposition 2.2.

It is proved in [11] that ${\left[\overline{g}\right]}_{1}$ is an embedded submanifold of ${\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1}$.

#### Proposition 2.3.

The ${\mathcal{C}}^{k\mathrm{,}\alpha }$-conformal class of metrics ${\mathrm{\left[}\overline{g}\mathrm{\right]}}_{\mathrm{1}}$ of unit volume is an embedded submanifold of ${\mathcal{M}}^{k\mathrm{,}\alpha }\mathit{}{\mathrm{\left(}M\mathrm{\right)}}_{\mathrm{1}}$. Moreover,

${T}_{\overline{g}}{\left[\overline{g}\right]}_{1}=\left\{\psi \overline{g}:\psi \in {\mathcal{𝒞}}^{k,\alpha }\left(M\right),{\int }_{M}\psi {\omega }_{\overline{g}}=0\right\}.$

#### Proof.

Since we know that ${\mathcal{𝒞}}_{+}^{k,\alpha }\left(M\right)$ is a Banach manifold, we can define a smooth function ${\mathcal{𝒱}}_{\overline{g}}:{\mathcal{𝒞}}_{+}^{k,\alpha }\left(M\right)\to ℝ$ for each $\overline{g}\in {\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1}$ by

${\mathcal{𝒱}}_{\overline{g}}\left(\varphi \right):=\mathcal{𝒱}\left(\varphi \overline{g}\right)$

with differential given by

${\left(\mathrm{d}{\mathcal{𝒱}}_{\overline{g}}\right)}_{\varphi }\left(\psi \right)=\frac{m}{2}{\int }_{M}{\varphi }^{\frac{m}{2}-1}\psi {\omega }_{\overline{g}}$

for each $\psi \in {T}_{\varphi }{\mathcal{𝒞}}_{+}^{k,\alpha }\left(M\right)={\mathcal{𝒞}}^{k,\alpha }\left(M\right)$.

Let $\varphi \in {\mathcal{𝒞}}_{+}^{k,\alpha }\left(M\right)$ be such that ${\mathcal{𝒱}}_{\overline{g}}\left(\varphi \right)=1$. Take $\psi =\varphi$ to see that ${\left(\mathrm{d}{\mathcal{𝒱}}_{\overline{g}}\right)}_{\varphi }\left(\varphi \right)=\frac{m}{2}\ne 0$, that is, ${\left(\mathrm{d}{\mathcal{𝒱}}_{\overline{g}}\right)}_{\varphi }$ is surjective. Moreover, its kernel has complement in ${\mathcal{𝒞}}^{k,\alpha }\left(M\right)$, which implies that ${\mathcal{𝒱}}_{\overline{g}}^{-1}\left(1\right)$ is an embedded submanifold of ${\mathcal{𝒞}}_{+}^{k,\alpha }\left(M\right)$.

Now, for $\overline{g}\in {\left(M\right)}^{k,\alpha }{\left(M\right)}_{1}$, define the smooth maps $S:{\mathcal{𝒱}}_{\overline{g}}^{-1}\left(1\right)\to {\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1}$ and $T:{\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1}\to {\mathcal{𝒞}}^{k,\alpha }\left(M\right)$ by

$S\left(\varphi \right)=\varphi \overline{g}\mathit{ }\text{and}\mathit{ }T\left(\stackrel{~}{g}\right)=\frac{1}{m}{\mathrm{tr}}_{\overline{g}}\left(\stackrel{~}{g}\right).$

Observe that S is an injective immersion and T is a smooth left-inverse for S. Moreover, $\mathrm{Im}{\left(\mathrm{d}S\right)}_{\varphi }$ has a closed complement. Therefore, ${\left[\overline{g}\right]}_{1}=\mathrm{Im}S$ is an embedded submanifold of ${\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1}$.

Finally, if we take $\varphi =1$ and $S\left(1\right)=\overline{g}$, we have

${T}_{1}{\mathcal{𝒱}}_{\overline{g}}^{-1}\left(1\right)=\left\{\psi \in {\mathcal{𝒞}}^{k,\alpha }\left(M\right):{\int }_{M}\psi {\omega }_{\overline{g}}=0\right\}.$

Since S is an immersion, we have ${T}_{\overline{g}}{\left[\overline{g}\right]}_{1}=\mathrm{Im}{\left(dS\right)}_{1}$. But since ${\left(\mathrm{d}S\right)}_{\varphi }\left(\psi \right)=\psi \overline{g}$ for all $\varphi \in {\mathcal{𝒱}}_{\overline{g}}^{-1}\left(1\right)$, $\psi \in {T}_{\varphi }{\mathcal{𝒱}}_{\overline{g}}^{-1}\left(1\right)$, we obtain the desired expression for the tangent space, which can also be identified with

${T}_{\overline{g}}{\left[\overline{g}\right]}_{1}=\left\{\psi \in {\mathcal{𝒞}}^{k,\alpha }\left(M\right):{\int }_{M}\psi {\omega }_{\overline{g}}=0\right\}.\mathit{∎}$

Note that ${\left[\overline{g}\right]}_{1}={\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1}\cap \left[\overline{g}\right]$. In [11], it is also proved that ${\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1}$ is transverse to $\left[\overline{g}\right]$, so ${\left[\overline{g}\right]}_{1}$ is an embedded submanifold of ${\mathcal{ℳ}}^{k,\alpha }\left(M\right)$.

Now, define the ${\mathcal{𝒞}}^{k,\alpha }$-normalized conformal class of a metric $\overline{g}\in {\mathcal{ℳ}}^{k,\alpha }\left(M\right)$ by

${\left[\overline{g}\right]}^{0}=\left\{\stackrel{~}{g}\in \left[\overline{g}\right]:{H}_{\stackrel{~}{g}}=0\right\}.$

This is a non-empty set. Indeed, a result obtained by Escobar [8] assures that there is at least one metric with vanishing mean curvature in each conformal class, so given a conformal class $\left[\overline{g}\right]$, we can assume that ${H}_{\overline{g}}=0$.

#### Proposition 2.4.

The ${\mathcal{C}}^{k\mathrm{,}\alpha }$-normalized conformal class of $\overline{g}$ can be identified with

which is a closed subset of ${\mathcal{C}}^{k\mathrm{,}\alpha }\mathit{}\mathrm{\left(}M\mathrm{\right)}$.

#### Proof.

Let $\stackrel{~}{g}={\varphi }^{\frac{4}{m-2}}\overline{g}\in {\left[\overline{g}\right]}^{0}$. We denote with a tilde all quantities related with $\stackrel{~}{g}$. Then, $\stackrel{~}{\eta }={\varphi }^{\frac{-2}{m-2}}{\eta }_{\overline{g}}$ and

$\stackrel{~}{H}={\stackrel{~}{g}}^{ij}\stackrel{~}{g}\left(\stackrel{~}{\eta },{\stackrel{~}{\nabla }}_{{\partial }_{i}}{\partial }_{j}\right)={\stackrel{~}{g}}^{ij}\stackrel{~}{g}\left(\stackrel{~}{\eta },{\stackrel{~}{\mathrm{\Gamma }}}_{ij}^{r}{\partial }_{r}\right),$

but ${\stackrel{~}{\mathrm{\Gamma }}}_{ij}^{r}={\mathrm{\Gamma }}_{ij}^{r}+\frac{2}{m-2}{\varphi }^{-1}\left({\delta }_{j}^{r}{\partial }_{i}\varphi +{\delta }_{i}^{r}{\partial }_{j}\varphi -{\overline{g}}_{ij}{\nabla }^{r}\varphi \right)$ and ${\stackrel{~}{g}}^{ij}={\varphi }^{\frac{-4}{m-2}}{\overline{g}}^{ij}$. Then,

$\stackrel{~}{H}={\varphi }^{\frac{-2}{m-2}}\left({H}_{\overline{g}}+\frac{2\left(m-1\right)}{m-2}{\varphi }^{-1}{\partial }_{{\eta }_{\overline{g}}}\varphi \right)={\varphi }^{-\frac{m}{m-2}}\left(\varphi {H}_{\overline{g}}+\frac{2\left(m-1\right)}{m-2}{\partial }_{{\eta }_{\overline{g}}}\varphi \right).$

Since ${H}_{\overline{g}}=0$ and $\varphi >0$, it follows that $\stackrel{~}{H}=0$ if and only if ${\partial }_{{\eta }_{\overline{g}}}\varphi =0$. ∎

We want to show that the ${\mathcal{𝒞}}^{k,\alpha }$-normalized conformal class ${\left[\overline{g}\right]}^{0}$ is a submanifold of the ${\mathcal{𝒞}}^{k,\alpha }$-conformal class $\left[\overline{g}\right]$. To this aim, we need the following proposition which is an elementary version of a more general result that can be found in [15, Chapter IV, Theorem 4].

#### Proposition 2.5.

There exists a continuous linear map

$\mathcal{ℱ}:{\mathcal{𝒞}}^{k,\alpha }\left(\partial M\right)\to {\mathcal{𝒞}}^{k+1,\alpha }\left(M\right)$

such that, for $\xi \mathrm{\in }{\mathcal{C}}^{k\mathrm{,}\alpha }\mathit{}\mathrm{\left(}\mathrm{\partial }\mathit{}M\mathrm{\right)}$, the following properties are satisfied:

• (i)

$\mathcal{ℱ}\left(\xi \right)$ vanishes on $\partial M$ ;

• (ii)

${\partial }_{\eta }\mathcal{ℱ}\left(\xi \right)=\xi$.

#### Proof.

Choose a finite set of local charts $\left({U}_{r},{\phi }_{r}\right)$ on M, $r=1,\mathrm{\dots },n$, that satisfy the following properties:

• (i)

${U}_{r}$ is an open subset of M with ${U}_{r}\cap \partial M\ne \mathrm{\varnothing }$ for all r;

• (ii)

$U={\bigcup }_{r=1}^{n}{U}_{r}$ is an open neighborhood of $\partial M$;

• (iii)

${\phi }_{r}$ is a diffeomorphism from ${U}_{r}$ to ${ℝ}^{m-1}×\left[0,+\mathrm{\infty }\right)$ carrying ${U}_{r}\cap \partial M$ onto ${ℝ}^{m-1}×\left\{0\right\}$;

• (iv)

${\left(\mathrm{d}{\phi }_{r}\right)}_{p}\left(\eta \left(p\right)\right)=\frac{\partial }{\partial {x}_{m}}$ for all $p\in {U}_{r}\cap \partial M$.

Set ${U}_{0}=M\setminus \partial M$, so that ${\left({U}_{r}\right)}_{r=0,\mathrm{\dots },n}$ is an open cover of M, and let ${\left({\rho }_{r}\right)}_{r=0}^{n}$ be a smooth partition of unity subordinated to such a cover. Given $\xi \in {\mathcal{𝒞}}^{k,\alpha }\left(\partial M\right)$, consider, for all r, the function ${\xi }_{r}=\xi \circ {\phi }_{r}^{-1}:{ℝ}^{m-1}\to ℝ$, which is of class ${\mathcal{𝒞}}^{k,\alpha }$. Let ${F}_{{\xi }_{r}}:{ℝ}^{m}\to ℝ$ be defined by

${F}_{{\xi }_{r}}\left(x\right)=\frac{1}{{x}_{m}^{m-1}}{\int }_{Q\left(x\right)}{\xi }_{r}\left(z\right)𝑑z,$

where $x=\left({x}_{1},\mathrm{\dots },{x}_{m-1},{x}_{m}\right)$, $Q\left(x\right)={\prod }_{i=1}^{m-1}\left[{x}_{i}-\frac{1}{2}{x}_{m},{x}_{i}+\frac{1}{2}{x}_{m}\right]$, $z=\left({z}_{1},\mathrm{\dots },{z}_{m-1}\right)$, and ${x}_{m}\ne 0$. Note that ${F}_{{\xi }_{r}}\left({x}_{1},\mathrm{\dots },{x}_{m-1},0\right)=0.$ A straightforward calculation shows that ${F}_{{\xi }_{r}}\in {\mathcal{𝒞}}^{k+1,\alpha }\left({ℝ}^{m}\right)$. Let now ${\mathcal{ℱ}}_{r}={F}_{{\xi }_{r}}\circ {\phi }_{r}$. Clearly, ${\mathcal{ℱ}}_{r}\in {\mathcal{𝒞}}^{k+1,\alpha }\left({U}_{r}\right).$ Finally, define $\mathcal{ℱ}\left(\xi \right):M\to ℝ$ as

$\mathcal{ℱ}\left(\xi \right):=\sum _{r=1}^{n}{\rho }_{r}\cdot {\mathcal{ℱ}}_{r}.$

It is easy to see that $\mathcal{ℱ}\left(\xi \right)$ satisfies the desired properties. ∎

#### Proposition 2.6.

The ${\mathcal{C}}^{k\mathrm{,}\alpha }$-normalized conformal class ${\mathrm{\left[}\overline{g}\mathrm{\right]}}^{\mathrm{0}}$ is an embedded submanifold of $\mathrm{\left[}\overline{g}\mathrm{\right]}$.

#### Proof.

Given $\overline{g}\in {\mathcal{ℳ}}^{k,\alpha }\left(M\right)$, let ${\eta }_{\overline{g}}$ be the unit (inward) vector field normal to the boundary $\partial M$. Define

${\mathcal{𝒩}}_{\overline{g}}:\left[\overline{g}\right]\to {\mathcal{𝒞}}^{k-1,\alpha }\left(\partial M\right),$$\varphi \overline{g}↦{\partial }_{{\eta }_{\overline{g}}}\varphi .$

So, ${\mathcal{𝒩}}_{\overline{g}}^{-1}\left(\left\{0\right\}\right)={\left[\overline{g}\right]}^{0}$ and the differential ${\left(\mathrm{d}{\mathcal{𝒩}}_{\overline{g}}\right)}_{\varphi \overline{g}}:{\mathcal{𝒞}}^{k,\alpha }\left(M\right)\to {\mathcal{𝒞}}^{k-1,\alpha }\left(\partial M\right)$ is given by

${\left(\mathrm{d}{\mathcal{𝒩}}_{\overline{g}}\right)}_{\varphi \overline{g}}\left(\psi \right)={\partial }_{{\eta }_{\overline{g}}}\psi$

for all $\varphi \overline{g}\in \left[\overline{g}\right]$ and $\psi \in {\mathcal{𝒞}}^{k,\alpha }\left(M\right)$. Now, by the last proposition, ${\left(\mathrm{d}{\mathcal{𝒩}}_{\overline{g}}\right)}_{\varphi \overline{g}}$ admits a bounded right-inverse for all $\varphi \overline{g}\in \left[\overline{g}\right]$. Therefore, the differential is surjective and its kernel, given by

$\mathrm{ker}{\left(\mathrm{d}{\mathcal{𝒩}}_{\overline{g}}\right)}_{\varphi \overline{g}}=\left\{\psi \in {\mathcal{𝒞}}^{k,\alpha }\left(M\right):{\partial }_{{\eta }_{\overline{g}}}\psi =0\right\},$

has a closed complement in ${\mathcal{𝒞}}^{k,\alpha }\left(M\right)$. It follows that ${\left[\overline{g}\right]}^{0}$ is an embedded submanifold of $\left[\overline{g}\right]$. ∎

We can also combine both features of interest in the same conformal class, defining the ${\mathcal{𝒞}}^{k,\alpha }$-normalized conformal class consisting of metrics of unit volume as

${\left[\overline{g}\right]}_{1}^{0}=\left\{\varphi \overline{g}:\varphi \in {\mathcal{𝒞}}_{+}^{k,\alpha }\left(M\right),{\partial }_{{\eta }_{\overline{g}}}\varphi =0,{\int }_{M}{\varphi }^{\frac{m}{2}}{\omega }_{\overline{g}}=1\right\}.$

This is an embedded submanifold of ${\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1}$ and an embedded submanifold of $\left[\overline{g}\right]$. For instance, we can express ${\left[\overline{g}\right]}_{1}^{0}$ as ${\left[\overline{g}\right]}^{0}\cap {\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1}$. The corresponding tangent space is identified with

${T}_{\overline{g}}{\left[\overline{g}\right]}_{1}^{0}=\left\{\psi \in {\mathcal{𝒞}}^{k,\alpha }{\left(M\right)}^{0}:{\int }_{M}\psi {\omega }_{\overline{g}}=0\right\}.$

## 2.2 The Hilbert–Einstein functional

Consider the Hilbert–Einstein functional $F:{\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1}\to ℝ$ given by

$F\left(\overline{g}\right)={\int }_{M}{R}_{\overline{g}}{\omega }_{\overline{g}}.$(2.2)

It is well known that F is a smooth functional over ${\mathcal{ℳ}}^{k,\alpha }\left(M\right)$ and over $\left[\overline{g}\right]$. Let $\overline{g}\left(t\right)$ be a variation of a metric $\overline{g}\in {\mathcal{ℳ}}^{k,\alpha }\left(M\right)$ in the direction of $h\in {T}_{\overline{g}}{\mathcal{ℳ}}^{k,\alpha }\left(M\right)$, that is, $\overline{g}\left(0\right)=\overline{g}$ and ${\frac{d}{dt}\overline{g}\left(t\right)|}_{t=0}=h$. Then, we can calculate the first variation of the functional F as

$\delta F\left(\overline{g}\right)h:={\frac{d}{dt}|}_{t=0}F\left(\overline{g}\left(t\right)\right)={\int }_{M}\delta {R}_{\overline{g}\left(t\right)}{\omega }_{\overline{g}}+{R}_{\overline{g}}\delta {\omega }_{\overline{g}\left(t\right)},$

where the first variation of the scalar curvature is given by

$\delta {R}_{\overline{g}\left(t\right)}=\delta \left(\overline{g}{\left(t\right)}^{ij}{R}_{ij}\left(t\right)\right)=\delta \overline{g}{\left(t\right)}^{ij}{R}_{ij}\left(t\right)+\overline{g}{\left(t\right)}^{ij}\delta {R}_{ij}\left(t\right)=-{h}^{ij}{R}_{ij}\left(t\right)+{g}^{ij}\delta {R}_{ij}\left(t\right)$

with ${R}_{ij}\left(t\right)$ denoting the coordinates of the Ricci tensor of the metric $\overline{g}\left(t\right)$. We have

${g}^{ij}\delta {R}_{ij}\left(t\right)={\overline{\nabla }}_{i}\left({\overline{\nabla }}_{j}{h}^{ij}-{\overline{\nabla }}^{i}\left({\overline{g}}^{lm}{h}_{lm}\right)\right)$

and

$\delta {\omega }_{\overline{g}\left(t\right)}=\frac{1}{2}{\overline{g}}^{ij}{h}_{ij}{\omega }_{\overline{g}}.$

Then, we have

$\delta F\left(\overline{g}\right)h={\int }_{M}\left(-{h}_{ij}{R}_{ij}+\frac{1}{2}{\overline{g}}^{ij}{h}_{ij}{R}_{\overline{g}}\right){\omega }_{\overline{g}}+{\int }_{M}{\overline{\nabla }}_{i}\left({\overline{\nabla }}_{j}{h}^{ij}-{\overline{\nabla }}^{i}\left({\overline{g}}^{lm}{h}_{lm}\right)\right){\omega }_{\overline{g}}$

and, by the divergence theorem,

$\delta F\left(\overline{g}\right)h=-{\int }_{M}\left({h}_{ij}{R}_{ij}-\frac{1}{2}{\overline{g}}^{ij}{h}_{ij}{R}_{\overline{g}}\right){\omega }_{\overline{g}}-{\int }_{\partial M}\left({\overline{\nabla }}_{j}{h}^{ij}-{\overline{\nabla }}^{i}\left({\overline{g}}^{lm}{h}_{lm}\right)\right){\eta }_{i}{\sigma }_{\overline{g}},$

where ${\eta }_{i}$ denote the coordinates, with respect to $\overline{g}$, of the inward unit normal vector field on the boundary. This expression can be written in a concise form as

$\delta F\left(\overline{g}\right)h=-{\int }_{M}{〈{\mathrm{Ric}}_{\overline{g}}-\frac{1}{2}{R}_{\overline{g}}\overline{g},h〉}_{\overline{g}}{\omega }_{\overline{g}}-{\int }_{\partial M}\left({\overline{\nabla }}_{j}{h}^{ij}\right){\eta }_{i}{\sigma }_{\overline{g}}+{\int }_{\partial M}{〈\overline{\nabla }{\mathrm{tr}}_{\overline{g}}h,\eta 〉}_{\overline{g}}{\sigma }_{\overline{g}}$

and, after some further calculations, we get

$\delta F\left(\overline{g}\right)h=-{\int }_{M}{〈{\mathrm{Ric}}_{\overline{g}}-\frac{1}{2}{R}_{\overline{g}}\overline{g},h〉}_{\overline{g}}{\omega }_{\overline{g}}-2{\int }_{\partial M}\left(\delta {H}_{\overline{g}}+\frac{1}{2}〈I{I}_{\overline{g}},h〉\right){\sigma }_{\overline{g}}.$

For compact Riemannian manifolds (without boundary), it is well known that the critical points of F on ${\mathcal{ℳ}}^{k}{\left(M\right)}_{1}$ are the Einstein metrics of unit volume on M and that if F is restricted to the ${\mathcal{𝒞}}^{k,\alpha }$-conformal classes ${\left[\overline{g}\right]}_{1}$ of unit volume, then the critical points are those metrics conformal to $\overline{g}$ which have unit volume and constant scalar curvature. Since we are dealing with manifolds with boundary, we are interested in critical points for the restriction to the ${\mathcal{𝒞}}^{k,\alpha }$-normalized conformal classes.

The critical points of F on ${\left[\overline{g}\right]}_{1}^{0}$ are those metrics conformal to $\overline{g}$ which have unit volume, constant scalar curvature, and vanishing mean curvature. Indeed, when we take $h=\psi \overline{g}\in {T}_{\overline{g}}{\left[\overline{g}\right]}_{1}^{0}$, the first variation becomes

$\delta F\left(\overline{g}\right)\psi \overline{g}=-{\int }_{M}{〈{\mathrm{Ric}}_{\overline{g}}-\frac{1}{2}{R}_{\overline{g}}\overline{g},\psi \overline{g}〉}_{\overline{g}}{\omega }_{\overline{g}}=-{\int }_{M}\left(\psi {R}_{\overline{g}}-\frac{1}{2}m\psi {R}_{\overline{g}}\right){\omega }_{\overline{g}}=\frac{m-2}{2}{\int }_{M}\psi {R}_{\overline{g}}{\omega }_{\overline{g}}.$

So, by the fundamental lemma of the calculus of variations, we have $\delta F\left(\overline{g}\right)\psi \overline{g}=0$ for all $\psi \in {\mathcal{𝒞}}^{k,\alpha }{\left(M\right)}^{0}$ with null integral if and only if ${R}_{\overline{g}}$ is constant.

If $\overline{g}\in {\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1}$ is a critical point of F on ${\left[\overline{g}\right]}_{1}^{0}$, then the second variation of F is given by the quadratic form (see [10])

${\delta }^{2}F\left(\overline{g}\right)\left(\psi \right)=\frac{m-2}{2}{\int }_{M}\left(\left(m-1\right){\mathrm{\Delta }}_{\overline{g}}\psi -{R}_{\overline{g}}\psi \right)\psi {\omega }_{\overline{g}},$

where $\psi \in {\mathcal{𝒞}}^{k,\alpha }{\left(M\right)}^{0}$ and has vanishing integral, and $\overline{g}$ is non-degenerate if ${R}_{\overline{g}}=0$ or if $\frac{{R}_{\overline{g}}}{m-1}$ is not an eigenvalue of ${\mathrm{\Delta }}_{\overline{g}}$ with Neumann boundary conditions.

In fact, note that $\left(m-1\right){\mathrm{\Delta }}_{g}-{R}_{g}$, as an operator from ${\mathcal{𝒞}}^{k,\alpha }\left(M\right)$ to ${\mathcal{𝒞}}^{k-2,\alpha }$, is Fredholm of index zero. Note also that this operator carries the subspace

$\left\{\psi \in {\mathcal{𝒞}}^{k,\alpha }\left(M\right):{\partial }_{{\eta }_{\overline{g}}}\psi =0,{\int }_{M}\psi =0\right\}$

into the subspace of ${\mathcal{𝒞}}^{k-2,\alpha }$ consisting of functions with vanishing integral. It follows that the operator

$\left(m-1\right){\mathrm{\Delta }}_{\overline{g}}-{R}_{\overline{g}}:{T}_{\overline{g}}{\left[\overline{g}\right]}_{1}\to {\mathcal{𝒞}}^{k-2,\alpha }\left(M\right)$

is Fredholm of index zero. So, the quadratic form ${\delta }^{2}F\left(\overline{g}\right)\left(\psi \overline{g},\psi \overline{g}\right)$ is non-degenerate if and only if the kernel $\mathrm{ker}\left(\left(m-1\right){\mathrm{\Delta }}_{g}-{R}_{g}\right)=\left\{0\right\}$. Indeed, the kernel is non-trivial if and only if $\frac{{R}_{\overline{g}}}{m-1}$ is a non-zero eigenvalue of ${\mathrm{\Delta }}_{\overline{g}}$.

## 3.1 Local rigidity

In this section, we present a result obtained in [11], but written in a slightly different way to better suit our context. We refer to [11, Proposition 3 and Corollary 4] for details. The proofs are essentially identical. We begin with the following definition.

#### Definition 3.1.

Let $\overline{g}\in {\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1}$ have constant scalar curvature ${R}_{\overline{g}}$ in M. We say that $\overline{g}$ is non-degenerate if either ${R}_{\overline{g}}=0$ or if $\frac{{R}_{\overline{g}}}{m-1}$ is not an eigenvalue of ${\mathrm{\Delta }}_{\overline{g}}$ with Neumann boundary condition ${\partial }_{{\eta }_{\overline{g}}}f=0$. In other words, $\frac{{R}_{\overline{g}}}{m-1}$ is not a solution of the eigenvalue problem

(3.1)

#### Proposition 3.2.

Let ${\overline{g}}_{\mathrm{*}}\mathrm{\in }{\mathcal{M}}^{k\mathrm{,}\alpha }\mathit{}{\mathrm{\left(}M\mathrm{\right)}}_{\mathrm{1}}$ be a non-degenerate constant scalar curvature metric. Then, there exists an open neighborhood U of ${\overline{g}}_{\mathrm{*}}$ in ${\mathcal{M}}^{k\mathrm{,}\alpha }\mathit{}{\mathrm{\left(}M\mathrm{\right)}}_{\mathrm{1}}$ such that the set

$S=\left\{\overline{g}\in U:{R}_{\overline{g}}\mathit{\text{is constant}}\right\}$

is a smooth embedded submanifold of ${\mathcal{M}}^{k\mathrm{,}\alpha }\mathit{}{\mathrm{\left(}M\mathrm{\right)}}_{\mathrm{1}}$ which is strongly transverse to the ${\mathcal{C}}^{k\mathrm{,}\alpha }$-normalized conformal classes.

#### Proof.

The proof is a direct application of [11, Proposition 1]. ∎

#### Corollary 3.3.

Let ${\overline{g}}_{\mathrm{*}}\mathrm{\in }{\mathcal{M}}^{k\mathrm{,}\alpha }\mathit{}{\mathrm{\left(}M\mathrm{\right)}}_{\mathrm{1}}$ be a non-degenerate metric on M with constant scalar curvature and vanishing mean curvature. Then, there is an open neighborhood U of ${\overline{g}}_{\mathrm{*}}$ in ${\mathcal{M}}^{k\mathrm{,}\alpha }\mathit{}{\mathrm{\left(}M\mathrm{\right)}}_{\mathrm{1}}$ such that every ${\mathcal{C}}^{k\mathrm{,}\alpha }$-normalized conformal class of metrics in ${\mathcal{M}}^{k\mathrm{,}\alpha }\mathit{}{\mathrm{\left(}M\mathrm{\right)}}_{\mathrm{1}}$ has at most one metric of constant scalar curvature and unit volume in U.

#### Proof.

The fact that the manifold S is transverse to the normalized conformal class guarantees the local uniqueness of intersections. ∎

## 3.2 Bifurcation of solutions

Let M be an m-dimensional compact Riemannian manifold with boundary for $m\ge 3$. Define a continuous path

$\left[a,b\right]\to {\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1},k\ge 3,$$s↦{\overline{g}}_{s},$

of Riemannian metrics on M with constant scalar curvature ${R}_{{\overline{g}}_{s}}$ and vanishing mean curvature ${H}_{{\overline{g}}_{s}}$ for all $s\in \left[a,b\right]$.

#### Definition 3.4.

An instant ${s}_{*}\in \left[a,b\right]$ is called a bifurcation instant for the family ${\left\{{\overline{g}}_{s}\right\}}_{s\in \left[a,b\right]}$ if there exists a sequence ${\left({s}_{n}\right)}_{n\ge 1}\subset \left[a,b\right]$ and a sequence ${\left({\overline{g}}_{n}\right)}_{n\ge 1}\subset {\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1}$ of Riemannian metrics on M satisfying the following:

• (i)

${lim}_{n\to \mathrm{\infty }}{s}_{n}={s}_{*}$ and ${lim}_{n\to \mathrm{\infty }}{\overline{g}}_{n}={\overline{g}}_{{s}_{*}}\in {\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1}$;

• (ii)

${\overline{g}}_{n}\in \left[{\overline{g}}_{{s}_{n}}\right]$ but ${\overline{g}}_{n}\ne {\overline{g}}_{{s}_{n}}$ for all $n\ge 1$;

• (iii)

${\overline{g}}_{n}$ has constant scalar curvature and vanishing mean curvature for all $n\ge 1$.

If ${s}_{*}\in \left[a,b\right]$ is not a bifurcation instant, then the family ${\left\{{\overline{g}}_{s}\right\}}_{s\in \left[a,b\right]}$ is called locally rigid at ${s}_{*}$.

An instant $s\in \left[a,b\right]$ for which $\frac{{R}_{{\overline{g}}_{s}}}{m-1}$ is a non-vanishing solution of the eigenvalue problem (3.1) is called a degeneracy instant for the family ${\left\{{g}_{s}\right\}}_{s\in \left[a,b\right]}$.

#### Theorem 3.5.

Let M be an m-dimensional compact manifold with boundary $\mathrm{\partial }\mathit{}M\mathrm{\ne }\mathrm{\varnothing }$ for $m\mathrm{\ge }\mathrm{3}$ and let

$\left[a,b\right]\to {\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1},k\ge 3,$$s↦{\overline{g}}_{s},$

be a ${C}^{\mathrm{1}}$-path of Riemannian metrics on M having constant scalar curvature ${R}_{{\overline{g}}_{s}}$ and vanishing mean curvature ${H}_{{\overline{g}}_{s}}$. Denote by ${n}_{s}$ the number of eigenvalues of the Laplace–Beltrami operator ${\mathrm{\Delta }}_{{\overline{g}}_{s}}$ with Neumann boundary condition (counted with multiplicity) that are less than $\frac{{R}_{{\overline{g}}_{s}}}{m\mathrm{-}\mathrm{1}}$. Assume that if $s\mathrm{=}a$ or $s\mathrm{=}b$, then $\frac{{R}_{{\overline{g}}_{s}}}{m\mathrm{-}\mathrm{1}}\mathrm{=}\mathrm{0}$ or it is not an eigenvalue of ${\mathrm{\Delta }}_{{\overline{g}}_{s}}$, and ${n}_{a}\mathrm{\ne }{n}_{b}$. Then, there exists a bifurcation instant ${s}_{\mathrm{*}}\mathrm{\in }\mathrm{\left(}a\mathrm{,}b\mathrm{\right)}$ for the family ${\mathrm{\left\{}{\overline{g}}_{s}\mathrm{\right\}}}_{s\mathrm{\in }\mathrm{\left[}a\mathrm{,}b\mathrm{\right]}}$.

#### Proof.

The result follows from the non-equivariant bifurcation theorem [12, Theorem A.2]. ∎

#### Remark 3.6.

Given a Riemannian manifold $\left(M,\overline{g}\right)$ with minimal boundary $\partial M\ne \mathrm{\varnothing }$, it is important to stress that, for all $s\in {ℝ}^{+}$, the manifold $\left(M,s\overline{g}\right)$ also has minimal boundary, since ${H}_{s\overline{g}}=\frac{1}{\sqrt{s}}{H}_{\overline{g}}$. Moreover, we have ${\mathrm{\Delta }}_{s\overline{g}}=\frac{1}{s}{\mathrm{\Delta }}_{\overline{g}}$, ${R}_{s\overline{g}}=\frac{1}{s}{R}_{\overline{g}}$, and ${\eta }_{s\overline{g}}=\frac{1}{\sqrt{s}}\eta$. This means that the spectrum of the operator ${\mathrm{\Delta }}_{\overline{g}}-\frac{{R}_{\overline{g}}}{m-1}$ with Neumann boundary condition is invariant by homothety of the metric. On the other hand, ${\omega }_{s\overline{g}}={s}^{\frac{m}{2}}{\omega }_{\overline{g}}$. When necessary, we will normalize metrics to have unit volume without change in the spectral theory of the operator ${\mathrm{\Delta }}_{\overline{g}}-\frac{{R}_{\overline{g}}}{m-1}$ with Neumann boundary condition.

## 4 Bifurcation of solutions of the Yamabe problem in product manifolds

Let $\left({M}_{1},{g}^{\left(1\right)}\right)$ be a compact Riemannian manifold with $\partial {M}_{1}=\mathrm{\varnothing }$ and constant scalar curvature, and let $\left({M}_{2},{\overline{g}}^{\left(2\right)}\right)$ be a compact Riemannian manifold with minimal boundary and constant scalar curvature. Consider the product manifold $M={M}_{1}×{M}_{2}$ whose boundary is given by $\partial M={M}_{1}×\partial {M}_{2}$. Let ${m}_{1}$ and ${m}_{2}$ be the dimensions of ${M}_{1}$ and ${M}_{2}$, respectively, and assume that $\mathrm{dim}\left(M\right)=m={m}_{1}+{m}_{2}\ge 3$. For each $s\in \left(0,+\mathrm{\infty }\right)$, define a family ${\overline{g}}_{s}={g}^{\left(1\right)}\oplus s{\overline{g}}^{\left(2\right)}$ of metrics on M. Then, ${\left\{{\overline{g}}_{s}\right\}}_{s}\subset {\mathcal{ℳ}}^{k,\alpha }\left(M\right)$.

The following statements, briefly justified, are valid.

• (i)

$\left(M,{\overline{g}}_{s}\right)$ has constant scalar curvature for all $s>0$ and its scalar curvature is given by

${R}_{{\overline{g}}_{s}}={R}_{{g}^{\left(1\right)}}+{R}_{s{\overline{g}}^{\left(2\right)}}={R}_{{g}^{\left(1\right)}}+\frac{1}{s}{R}_{{\overline{g}}^{\left(2\right)}}.$

• (ii)

Since we can identify the tangent space of the product manifold ${T}_{\left(p,q\right)}M$ with the direct sum ${T}_{p}{M}_{1}\oplus {T}_{q}{M}_{2}$ for $p\in {M}_{1}$ and $q\in {M}_{2}$, the interior vector field ${\eta }_{s}$ normal to $\partial M$ can be written as

${\eta }_{s}=0+\frac{1}{\sqrt{s}}{\eta }_{2},$

where ${\eta }_{2}$ is the interior vector field normal to $\partial {M}_{2}$.

• (iii)

The mean curvature of $\partial M$ is zero, since we have

${H}_{{\overline{g}}_{s}}={H}_{s{\overline{g}}^{\left(2\right)}}=\frac{1}{\sqrt{s}}{H}_{{\overline{g}}^{\left(2\right)}}.$

• (iv)

The Laplace–Beltrami operator with respect to ${\overline{g}}_{s}$ is given by

${\mathrm{\Delta }}_{{\overline{g}}_{s}}=\left({\mathrm{\Delta }}_{{g}^{\left(1\right)}}\otimes I\right)+\frac{1}{s}\left(I\otimes {\mathrm{\Delta }}_{{\overline{g}}^{\left(2\right)}}\right).$

• (v)

Consider the family ${\left\{{\overline{g}}_{s}\right\}}_{s>0}$. For the purpose of studying bifurcation instants, this causes no loss of generality. In fact, ${s}_{*}$ is a bifurcation instant for the family ${\left\{{g}^{\left(1\right)}\oplus s{\overline{g}}^{\left(2\right)}\right\}}_{s>0}$ if and only if ${s}_{*}$ is a bifurcation instant for the family ${\left\{\frac{1}{s}{g}^{\left(1\right)}\oplus {\overline{g}}^{\left(2\right)}\right\}}_{s>0}$ on M. The same is valid for degeneracy instants.

Now, considering the remark at the end of the previous section, ${\left\{{\overline{g}}_{s}\right\}}_{s>0}$ is a family of critical points of the functional

$F:{\mathcal{ℳ}}^{k,\alpha }{\left(M\right)}_{1}\to ℝ,$

restricted to ${\left[\overline{g}\right]}_{1}^{0}$, so it is a family of solutions with minimal boundary of the Yamabe problem in product manifolds.

In order to investigate the existence of bifurcation instants for the family ${\left\{\overline{g}\right\}}_{s>0}$, we are interested in studying the spectrum of the operator

${\mathcal{𝒥}}_{s}={\mathrm{\Delta }}_{{\overline{g}}_{s}}-\frac{{R}_{{\overline{g}}_{s}}}{m-1},$

whose domain is

$\left\{\psi \in {\mathcal{𝒞}}^{k,\alpha }{\left(M\right)}^{0}:{\int }_{M}\psi {\omega }_{{\overline{g}}_{s}}=0\right\}.$

Denote by $0={\rho }_{0}^{\left(1\right)}<{\rho }_{1}^{\left(1\right)}<{\rho }_{2}^{\left(1\right)}<\mathrm{\cdots }$ the sequence of all distinct eigenvalues of ${\mathrm{\Delta }}_{{g}^{\left(1\right)}}$ with geometric multiplicity ${\mu }_{i}^{\left(1\right)}$, $i\ge 0$, and by $0={\rho }_{0}^{\left(2\right)}<{\rho }_{1}^{\left(2\right)}<{\rho }_{2}^{\left(2\right)}<\mathrm{\cdots }$ the sequence of all distinct eigenvalues of ${\mathrm{\Delta }}_{{\overline{g}}^{\left(2\right)}}$ subject to Neumann boundary condition, that is,

(4.1)

where $j\ge 0$ and ${\mu }_{j}^{\left(2\right)}$ is the geometric multiplicity of ${\rho }_{j}^{\left(2\right)}$, $j\ge 0$. Then, the spectrum of ${\mathcal{𝒥}}_{s}$ is given by

$\mathrm{\Sigma }\left({\mathcal{𝒥}}_{s}\right)=\left\{{\sigma }_{i,j}:i,j\ge 0,i+j>0\right\},$

where

${\sigma }_{i,j}\left(s\right)={\rho }_{i}^{\left(1\right)}+\frac{1}{s}{\rho }_{j}^{\left(2\right)}-\frac{1}{m-1}\left({R}_{{g}^{\left(1\right)}}+\frac{1}{s}{R}_{{\overline{g}}^{\left(2\right)}}\right)$

are the eigenvalues of ${J}_{s}$ with Neumann boundary condition on $\partial M$ and with geometric multiplicity equal to the product ${\mu }_{i}^{\left(1\right)}{\mu }_{j}^{\left(2\right)}$.

We emphasize that the ${\sigma }_{i,j}$ are not necessarily all distinct.

#### Definition 4.1.

Let ${i}_{*}$ and ${j}_{*}$ be the smallest non-negative integers that satisfy

${\rho }_{{i}_{*}}^{\left(1\right)}\ge \frac{{R}_{{g}^{\left(1\right)}}}{m-1}\mathit{ }\text{and}\mathit{ }{\rho }_{{j}_{*}}^{\left(2\right)}\ge \frac{{R}_{{\overline{g}}^{\left(2\right)}}}{m-1}.$

We say that the pair of metrics $\left({g}^{\left(1\right)},{\overline{g}}^{\left(2\right)}\right)$ is degenerate if equalities hold in both cases, that is, if ${\sigma }_{{i}_{*},{j}_{*}}\left(s\right)=0$ for all s. In this case, the operator ${\mathcal{𝒥}}_{s}$ is also called degenerate.

We can state that, if ${R}_{{g}^{\left(1\right)}}<0$ or if ${R}_{{\overline{g}}^{\left(2\right)}}<0$ and ${H}_{{\overline{g}}^{\left(2\right)}}=0$, then $\left({g}^{\left(1\right)},{\overline{g}}^{\left(2\right)}\right)$ is certainly non-degenerate. Observe also that if $\left({g}^{\left(1\right)},{\overline{g}}^{\left(2\right)}\right)$ is degenerate, then zero is an eigenvalue of ${\mathcal{𝒥}}_{s}$ for all $s\in \left(0,+\mathrm{\infty }\right)$, otherwise there is only a discrete countable set $S\subset \left(0,+\mathrm{\infty }\right)$ of instants s for which the operator ${\mathcal{𝒥}}_{s}$ is singular. First, we consider the case when both scalar curvatures are positive.

## 4.1 The case of positive scalar curvature

We are interested in studying the zeros of the function $s↦{\sigma }_{i,j}\left(s\right)$ as $i,j$ vary. At first glance, we can already draw some conclusions. For instance, if the function ${\sigma }_{i,j}$ is not identically zero, for fixed $i,j$, it has at most one zero in $\left(0,+\mathrm{\infty }\right)$. Let us write ${\sigma }_{i,j}$ as

${\sigma }_{i,j}\left(s\right)={\rho }_{i}^{\left(1\right)}-\frac{{R}_{{g}^{\left(1\right)}}}{m-1}+\frac{1}{s}\left({\rho }_{j}^{\left(2\right)}-\frac{{R}_{{\overline{g}}^{\left(2\right)}}}{m-1}\right).$

Derive

${\sigma }_{i,j}^{\prime }\left(s\right)=-\frac{1}{{s}^{2}}\left({\rho }_{j}^{\left(2\right)}-\frac{{R}_{{\overline{g}}^{\left(2\right)}}}{m-1}\right),$

so ${\sigma }_{i,j}^{\prime }\left(s\right)=0$ if and only if ${\rho }_{j}^{\left(2\right)}=\frac{{R}_{{\overline{g}}^{\left(2\right)}}}{m-1}$ if and only if ${R}_{{\overline{g}}^{\left(2\right)}}$ is a solution of (4.1). If ${\sigma }_{i,j}^{\prime }\left(s\right)=0$ for some s, then ${\sigma }_{i,j}^{\prime }\left(s\right)=0$ for all $s\in \left(0,+\mathrm{\infty }\right)$. Hence, σ is strictly monotone or constant, so if it is not identically null (constant), it certainly has at most one zero (strictly monotone).

#### Lemma 4.2.

Assume that the pair $\mathrm{\left(}{g}^{\mathrm{\left(}\mathrm{1}\mathrm{\right)}}\mathrm{,}{\overline{g}}^{\mathrm{\left(}\mathrm{2}\mathrm{\right)}}\mathrm{\right)}$ is non-degenerate and that ${R}_{{g}^{\mathrm{\left(}\mathrm{1}\mathrm{\right)}}}\mathrm{,}{R}_{{\overline{g}}^{\mathrm{\left(}\mathrm{2}\mathrm{\right)}}}\mathrm{>}\mathrm{0}$ with ${H}_{{\overline{g}}^{\mathrm{\left(}\mathrm{2}\mathrm{\right)}}}\mathrm{=}\mathrm{0}$. Then, the functions ${\sigma }_{i\mathrm{,}j}\mathit{}\mathrm{\left(}s\mathrm{\right)}$ satisfy the following properties:

• (i)

For all $i,j\ge 0$ , the map $s↦{\sigma }_{i,j}\left(s\right)$ is strictly monotone in $\left(0,+\mathrm{\infty }\right)$ except for the maps ${\sigma }_{i,{j}_{*}}$ which are constant and equal to ${\rho }_{i}^{\left(1\right)}-\frac{{R}_{{g}^{\left(1\right)}}}{m-1}$ when ${\rho }_{{j}_{*}}^{\left(2\right)}=\frac{{R}_{{\overline{g}}^{\left(2\right)}}}{m-1}$.

• (ii)

For $i\ne {i}_{*}$ and $j\ne {j}_{*}$ , the map ${\sigma }_{i,j}\left(s\right)$ admits a zero if and only if

• either $j<{j}_{*}$ and $i>{i}_{*}$ , in which case ${\sigma }_{i,j}$ is strictly increasing,

• or if $j>{j}_{*}$ and $i<{i}_{*}$ , in which case ${\sigma }_{i,j}$ is strictly decreasing.

• (iii)

If ${\rho }_{{i}_{*}}^{\left(1\right)}=\frac{{R}_{{g}^{\left(1\right)}}}{m-1}$ , then ${\sigma }_{{i}_{*},j}$ does not have zeros for any $j\in \left(0,+\mathrm{\infty }\right)$ . If ${\rho }_{{i}_{*}}^{\left(1\right)}>\frac{{R}_{{g}^{\left(1\right)}}}{m-1}$ , then ${\sigma }_{{i}_{*},j}$ has a zero if and only if $j<{j}_{*}$.

• (iv)

If ${\rho }_{{j}_{*}}^{\left(2\right)}=\frac{{R}_{\overline{g}}^{\left(2\right)}}{m-1}$ , then ${\sigma }_{i,{j}_{*}}$ does not have zeros for any $i\in \left(0,+\mathrm{\infty }\right)$ . If ${\rho }_{{j}_{*}}^{\left(2\right)}>\frac{{R}_{{\overline{g}}^{\left(2\right)}}}{m-1}$ , then ${\sigma }_{i,{j}_{*}}$ has a zero if and only if $i<{i}_{*}$.

#### Proof.

The entire statement follows directly from a straightforward analysis of the expression

${\sigma }_{i,j}\left(s\right)=\left({\rho }_{i}^{\left(1\right)}-\frac{{R}_{{g}^{\left(1\right)}}}{m-1}\right)+\frac{1}{s}\left({\rho }_{j}^{\left(2\right)}-\frac{{R}_{{\overline{g}}^{\left(2\right)}}}{m-1}\right).\mathit{∎}$

#### Corollary 4.3.

If $\mathrm{\left(}{g}^{\mathrm{\left(}\mathrm{1}\mathrm{\right)}}\mathrm{,}{\overline{g}}^{\mathrm{\left(}\mathrm{2}\mathrm{\right)}}\mathrm{\right)}$ is non-degenerate, then the set $S\mathrm{\subset }\mathrm{\left(}\mathrm{0}\mathrm{,}\mathrm{+}\mathrm{\infty }\mathrm{\right)}$ of instants s at which ${\mathcal{J}}_{s}$ is singular, is countable and discrete. More precisely, it consists of a strictly decreasing sequence ${\mathrm{\left(}{s}_{n}^{\mathrm{\left(}\mathrm{1}\mathrm{\right)}}\mathrm{\right)}}_{n}$ tending to 0 and a strictly increasing unbounded sequence ${\mathrm{\left(}{s}_{n}^{\mathrm{\left(}\mathrm{2}\mathrm{\right)}}\mathrm{\right)}}_{n}$. For all other values of s, ${\mathcal{J}}_{s}$ is an isomorphism and, in particular, the family ${\mathrm{\left\{}{\overline{g}}_{s}\mathrm{\right\}}}_{s\mathrm{>}\mathrm{0}}$ is locally rigid at these instants.

#### Proof.

By Lemma 4.2, each function ${\sigma }_{i,j}$ has at most one zero, thus there is only a countable number of degeneracy instants for ${\mathcal{𝒥}}_{s}$. Let ${s}_{ij}$ be the zero of ${\sigma }_{ij}$. Then,

$0<{s}_{ij}=-\frac{{\rho }_{j}^{\left(2\right)}-\frac{{R}_{{\overline{g}}^{\left(2\right)}}}{m-1}}{{\rho }_{i}^{\left(1\right)}-\frac{{R}_{{g}^{\left(1\right)}}}{m-1}}.$

Now, we study the behavior of these zeros in the following two cases:

• If $j>{j}_{*}$ and $i<{i}_{*}$, then

${s}_{ij}=\frac{{\rho }_{j}^{\left(2\right)}-\frac{{R}_{{\overline{g}}^{\left(2\right)}}}{m-1}}{\frac{{R}_{{g}^{\left(1\right)}}}{m-1}-{\rho }_{i}^{\left(1\right)}}\ge \left({\rho }_{j}^{\left(2\right)}-\frac{{R}_{{\overline{g}}^{\left(2\right)}}}{m-1}\right)\frac{1}{\frac{{R}_{{g}^{\left(1\right)}}}{m-1}-{\rho }_{{i}_{*}-1}^{\left(1\right)}}\to +\mathrm{\infty }$

as $j\to +\mathrm{\infty }$.

• If $i>{i}_{*}$ and $j<{j}_{*}$, then

${s}_{ij}=\frac{\frac{{R}_{{\overline{g}}^{\left(2\right)}}}{m-1}-{\rho }_{j}^{\left(2\right)}}{{\rho }_{i}^{\left(1\right)}-\frac{{R}_{{g}^{\left(1\right)}}}{m-1}}\le \frac{1}{{\rho }_{i}^{\left(1\right)}-\frac{{R}_{{g}^{\left(1\right)}}}{m-1}}\left(\frac{{R}_{{\overline{g}}^{\left(2\right)}}}{m-1}-{\rho }_{1}^{\left(2\right)}\right)\to 0$

as $i\to +\mathrm{\infty }$.

Therefore, all the zeros of the eigenvalues ${\sigma }_{i,j}$ accumulate only at 0 and at $+\mathrm{\infty }$. Let ${s}_{*}\in \left(0,+\mathrm{\infty }\right)\setminus S$. Then, ${\mathcal{𝒥}}_{{s}_{*}}$ is an isomorphism, that is, $0\notin \mathrm{\Sigma }\left({J}_{s⁣*}\right)$ or, equivalently, $\frac{{R}_{{\overline{g}}_{{s}_{*}}}}{m-1}$ is not an eigenvalue of ${\mathrm{\Delta }}_{{\overline{g}}_{{s}_{*}}}$. So, ${\overline{g}}_{{s}_{*}}\in {\left\{{\overline{g}}_{s}\right\}}_{s>0}$ is a non-degenerate metric. It follows from Proposition 3.2 that the family ${\left\{{\overline{g}}_{s}\right\}}_{s>0}$ is locally rigid at ${s}_{*}$. ∎

Note that ${R}_{\overline{g}}$ is obviously different from zero, since we are considering only positive scalar curvature.

#### Theorem 4.4.

Let $\mathrm{\left(}{M}_{\mathrm{1}}\mathrm{,}{g}^{\mathrm{\left(}\mathrm{1}\mathrm{\right)}}\mathrm{\right)}$ be a compact Riemannian manifold with positive constant scalar curvature and let $\mathrm{\left(}{M}_{\mathrm{2}}\mathrm{,}{\overline{g}}^{\mathrm{\left(}\mathrm{2}\mathrm{\right)}}\mathrm{\right)}$ be a compact manifold with boundary and with positive constant scalar curvature and minimal boundary $\mathrm{\partial }\mathit{}{M}_{\mathrm{2}}$. Assume that the pair $\mathrm{\left(}{g}^{\mathrm{\left(}\mathrm{1}\mathrm{\right)}}\mathrm{,}{\overline{g}}^{\mathrm{\left(}\mathrm{2}\mathrm{\right)}}\mathrm{\right)}$ is non-degenerate. For all $s\mathrm{\in }\mathrm{\left(}\mathrm{0}\mathrm{,}\mathrm{+}\mathrm{\infty }\mathrm{\right)}$, let ${\overline{g}}_{s}\mathrm{=}{g}^{\mathrm{\left(}\mathrm{1}\mathrm{\right)}}\mathrm{\oplus }s\mathit{}{\overline{g}}^{\mathrm{\left(}\mathrm{2}\mathrm{\right)}}$ be the metric on the product manifold with boundary $M\mathrm{=}{M}_{\mathrm{1}}\mathrm{×}{M}_{\mathrm{2}}$. Then, there exists a sequence tending to 0 and a sequence tending to $\mathrm{+}\mathrm{\infty }$ consisting of bifurcation instants for the family ${\mathrm{\left\{}{\overline{g}}_{s}\mathrm{\right\}}}_{s\mathrm{>}\mathrm{0}}$.

#### Proof.

We prove that bifurcation instants consist of subsequences of the two sequences of instants where ${\mathcal{𝒥}}_{s}$ is singular, whose existence was proved above.

With the same notation used in Corollary 4.3, let ${n}_{0}>0$ be such that ${s}_{n}^{\left(1\right)}<{s}_{1}^{\left(2\right)}$ and ${s}_{1}^{\left(1\right)}<{s}_{n}^{\left(2\right)}$ for all $n>{n}_{0}$. Then, there is $\epsilon >0$ for all $n>{n}_{0}$ such that the operator ${\mathcal{𝒥}}_{\left(\cdot \right)}$ is an isomorphism on the intervals $\left[{s}_{n}^{\left(1\right)}-\epsilon ,{s}_{n}^{\left(1\right)}+\epsilon \right]$ and $\left[{s}_{n}^{\left(2\right)}-\epsilon ,{s}_{n}^{\left(2\right)}+\epsilon \right]$, except for the instants ${s}_{n}^{\left(1\right)}$ and ${s}_{n}^{\left(2\right)}$ themselves.

As the zeros of the increasing eigenvalue functions accumulate at 0 and the zeros of the decreasing eigenvalue functions accumulate at $+\mathrm{\infty }$, if ${\sigma }_{p,q}$ is a non-increasing eigenvalue function, for all $s\in \left(0,{s}_{n}^{\left(1\right)}+\epsilon \right]$, $n>{n}_{0}$, we have ${\sigma }_{p,q}\left(s\right)\ne 0$. So, ${\sigma }_{p,q}\left({s}_{n}^{\left(1\right)}-\epsilon \right)<0$ if and only if ${\sigma }_{p,q}\left({s}_{n}^{\left(1\right)}+\epsilon \right)<0$. On the other hand, if we consider an increasing eigenvalue function ${\sigma }_{i,j}$, for all $n>{n}_{0}$, we have ${\sigma }_{i,j}\left({s}_{n}^{\left(1\right)}\right)=0$, ${\sigma }_{i,j}\left({s}_{n}^{\left(1\right)}-\epsilon \right)<0$, ${\sigma }_{i,j}\left({s}_{n}^{\left(1\right)}+\epsilon \right)>0$, and the fact that ${s}_{n}^{\left(1\right)}<{s}_{1}^{\left(2\right)}$ ensures that there is no decreasing function that vanishes at ${s}_{n}^{\left(1\right)}$. Hence, we can surely conclude that ${n}_{{s}_{n}^{\left(1\right)}-\epsilon }\ne {n}_{{s}_{n}^{\left(1\right)}+\epsilon }$. By Theorem 3.5, it follows that the subsequence ${\left({s}_{n}^{\left(1\right)}\right)}_{n>{n}_{0}}$ is the sought after sequence of bifurcation instants tending to 0.

Now, analyzing the non-decreasing eigenvalue function and the decreasing eigenvalue function in a similar way, we obtain ${n}_{{s}_{n}^{\left(2\right)}-\epsilon }\ne {n}_{{s}_{n}^{\left(2\right)}+\epsilon }$ and we can apply Theorem 3.5 to conclude that the subsequence ${\left({s}_{n}^{\left(2\right)}\right)}_{n>{n}_{0}}$ is the sought after sequence of bifurcation instants tending to $+\mathrm{\infty }$. ∎

Note that the case of degenerate pairs cannot be treated with Theorem 3.5, because in this case the zero is present in the spectrum of the operator ${\mathcal{𝒥}}_{s}$ for all $s\in \left(0,+\mathrm{\infty }\right)$ and thus the hypotheses of the theorem are never satisfied. Another interesting observation is that we do not know, by Theorem 4.4, if bifurcation occurs at degeneracy instants s between ${s}_{1}^{\left(1\right)}$ and ${s}_{1}^{\left(2\right)}$.

## 4.2 The case of non-positive scalar curvature

Now, consider the family ${\left\{{\overline{g}}_{s}\right\}}_{s>0}$ on the product manifold ${M}_{1}×{M}_{2}$ in the case when one or both scalar curvatures ${R}_{{g}^{\left(1\right)}}$ or ${R}_{{\overline{g}}^{\left(2\right)}}$ are non-positive, maintaining the Neumann boundary condition on $\partial {M}_{2}$. Observe that if ${R}_{{g}^{\left(1\right)}}$ and ${R}_{{\overline{g}}^{\left(2\right)}}$ are both non-positive, then the pair $\left({g}^{\left(1\right)},{\overline{g}}^{\left(2\right)}\right)$ is non-degenerate.

If ${R}_{{g}^{\left(1\right)}}\le 0$ and ${R}_{{\overline{g}}^{\left(2\right)}}>0$, then the pair $\left({g}^{\left(1\right)},{\overline{g}}^{\left(2\right)}\right)$ is degenerate if and only if ${R}_{{g}^{\left(1\right)}}=0$ and ${\rho }_{{j}_{*}}^{\left(2\right)}=\frac{{R}_{{\overline{g}}^{\left(2\right)}}}{m-1}$ as ${\rho }_{i}^{\left(1\right)}\ge 0$ for all $i\in \left\{0,1,2,\mathrm{\dots }\right\}$.

#### Theorem 4.5.

The following statements are true.

• (i)

If ${R}_{{g}^{\left(1\right)}}\le 0$ and ${R}_{{\overline{g}}^{\left(2\right)}}\le 0$ , then the family ${\left\{{\overline{g}}_{s}\right\}}_{s>0}$ has no degeneracy instants and thus it is locally rigid at every $s\in \left(0,+\mathrm{\infty }\right)$.

• (ii)

If ${R}_{{g}^{\left(1\right)}}\le 0$, ${R}_{{\overline{g}}^{\left(2\right)}}>0$ , and the pair $\left({g}^{\left(1\right)},{\overline{g}}^{\left(2\right)}\right)$ is non-degenerate, then the set of degeneracy instants for ${\mathcal{𝒥}}_{s}$ is a strictly decreasing sequence ${\left({s}_{n}\right)}_{n\in ℕ}$ that converges to 0 as $n\to \mathrm{\infty }$ . Moreover, every degeneracy instant is a bifurcation instant for the family ${\left\{{\overline{g}}_{s}\right\}}_{s>0}$

• (iii)

Symmetrically, if ${R}_{{g}^{\left(1\right)}}>0$, ${R}_{{\overline{g}}^{\left(2\right)}}\le 0$ , and the pair $\left({g}^{\left(1\right)},{\overline{g}}^{\left(2\right)}\right)$ is non-degenerate, then the set of degeneracy instants for ${\mathcal{𝒥}}_{s}$ is a strictly increasing unbounded sequence ${\left({s}_{n}\right)}_{n\in ℕ}$ and every degeneracy instant is a bifurcation instant for the family ${\left\{{\overline{g}}_{s}\right\}}_{s>0}$.

#### Proof.

The result follows from an analysis of the functions

${\sigma }_{i,j}\left(s\right)={\rho }_{i}^{\left(1\right)}+\frac{1}{s}{\rho }_{j}^{\left(2\right)}-\frac{1}{m-1}\left({R}_{{g}^{\left(1\right)}}+\frac{1}{s}{R}_{{\overline{g}}^{\left(2\right)}}\right).$

For (i), it is straightforward to see that ${\sigma }_{i,j}\left(s\right)>0$ for all $i,j=0,1,2,\mathrm{\dots }$ with $i+j>0$, so ${\mathcal{𝒥}}_{s}$ has no vanishing eigenvalues and the result follows.

For (ii), the functions ${\sigma }_{i,j}$ admit a zero only if $i\ge 0$ and $j<{j}_{*}$. If ${s}_{i,j}$ denotes the zero of such a function, then we have

$0<{s}_{i,j}=|\frac{{\rho }_{j}^{\left(2\right)}-\frac{{R}_{{\overline{g}}^{\left(2\right)}}}{m-1}}{{\rho }_{i}^{\left(1\right)}-\frac{{R}_{{g}^{\left(1\right)}}}{m-1}}|\le \frac{{R}_{{\overline{g}}^{\left(2\right)}}}{{\rho }_{i}^{\left(1\right)}-\frac{{R}_{{g}^{\left(1\right)}}}{m-1}}\to 0$

as $i\to +\mathrm{\infty }$. Hence, there is a decreasing sequence ${\left({s}_{n}\right)}_{n\in ℕ}$ of degeneracy instants that accumulates at 0. Since ${\left({s}_{n}\right)}_{n\in ℕ}$ accumulates only at 0, for each $n\in ℕ$, there exists $\epsilon >0$ such that the interval $\left[{s}_{n}-\epsilon ,{s}_{n}+\epsilon \right]$ contains only ${s}_{n}$ as a degeneracy instant. Arguing as in the proof of Theorem 4.4, we have ${n}_{{s}_{n}-\epsilon }\ne {n}_{{s}_{n}+\epsilon }$. The conclusion follows from Theorem 3.5.

Proceeding in a similar way, for (iii), the functions ${\sigma }_{i,j}$ admit a zero only if $i<{i}^{*}$ and $j\ge 0$, and we have an unbounded increasing sequence of degeneracy instants, each of which is a bifurcation instant for the family ${\left\{{\overline{g}}_{s}\right\}}_{s>0}$. ∎

## Acknowledgements

The authors gratefully acknowledge the supervision of Professor Paolo Piccione throughout the development of this work.

## References

• [1]

S. M. Almaraz, An existence theorem of conformal scalar-flat metrics on manifolds with boundary, Pacific J. Math. 248 (2010), no. 1, 1–22.  Google Scholar

• [2]

T. Aubin, Équations différentielles non linéaires et problème de Yamabe concernant la courbure scalaire, J. Math. Pures Appl. (9) 55 (1976), no. 3, 269–296.  Google Scholar

• [3]

R. G. Bettiol and P. Piccione, Bifurcation and local rigidity of homogeneous solutions to the Yamabe problem on spheres, Calc. Var. Partial Differential Equations 47 (2013), no. 3–4, 789–807.

• [4]

R. G. Bettiol and P. Piccione, Multiplicity of solutions to the Yamabe problem on collapsing Riemannian submersions, Pacific J. Math. 266 (2013), no. 1, 1–21.  Google Scholar

• [5]

M. M. Disconzi and M. A. Khuri, Compactness and non-compactness of the Yamabe problem on manifolds with boundary, J. Reine Angew. Math. (2014), DOI 10.1515/crelle-2014-0083.

• [6]

J. F. Escobar, Uniqueness theorems on conformal deformation of metrics, Sobolev inequalities, and an eigenvalue estimate, Comm. Pure Appl. Math. 43 (1990), no. 7, 857–883.  Google Scholar

• [7]

J. F. Escobar, Conformal deformation of a Riemannian metric to a scalar flat metric with constant mean curvature on the boundary, Ann. of Math. (2) 136 (1992), no. 1, 1–50.  Google Scholar

• [8]

J. F. Escobar, The Yamabe problem on manifolds with boundary, J. Differential Geom. 35 (1992), no. 1, 21–84.  Google Scholar

• [9]

Z.-C. Han and Y. Lin, The Yamabe problem on manifolds with boundary: existence and compactness results, Duke Math. J. 99 (1999), no. 3, 489–542.  Google Scholar

• [10]

N. Koiso, On the second derivative of the total scalar curvature, Osaka J. Math. 16 (1979), no. 2, 413–421.  Google Scholar

• [11]

L. L. de Lima, P. Piccione and M. Zedda, A note on the uniqueness of solutions for the Yamabe problem, Proc. Amer. Math. Soc. 140 (2012), no. 12, 4351–4357.

• [12]

L. L. de Lima, P. Piccione and M. Zedda, On bifurcation of solutions of the Yamabe problem in product manifolds, Ann. Inst. H. Poincaré Anal. Non Linéaire 29 (2012), no. 2, 261–277.  Google Scholar

• [13]

R. M. Schoen, Conformal deformation of a Riemannian metric to constant scalar curvature, J. Differential Geom. 20 (1984), no. 2, 479–495.  Google Scholar

• [14]

R. M. Schoen, Variational theory for the total scalar curvature functional for Riemannian metrics and related topics, in: Topics in Calculus of Variations (Montecatini Terme 1987), Lecture Notes in Math. 1365, Springer, Berlin (1989), 120–154.  Google Scholar

• [15]

E. M. Stein, Singular Integrals and Differentiability Properties of Functions, Princeton Math. Ser. 30, Princeton University Press, Princeton, 1970.  Google Scholar

• [16]

N. S. Trudinger, Remarks concerning the conformal deformation of Riemannian structures on compact manifolds, Ann. Sc. Norm. Super. Pisa Cl. Sci. (3) 22 (1968), no. 2, 265–274.  Google Scholar

• [17]

H. Yamabe, On a deformation of Riemannian structures on compact manifolds, Osaka J. Math. 12 (1960), 21–37.  Google Scholar

## Footnotes

• 1

Let $T:E\to F$ be a linear bounded operator between Banach spaces. If T admits a continuous right-inverse $S:F\to E$, then $\mathrm{ker}T$ admits a closed complement. Moreover, this complement is $\mathrm{Im}S$.

Accepted: 2015-12-12

Published Online: 2016-01-19

Published in Print: 2018-02-01

The first author is partially sponsored by CAPES-Brazil via the Instituto de Matemática e Estatística, Universidade de São Paulo.

Citation Information: Advances in Nonlinear Analysis, Volume 7, Issue 1, Pages 1–14, ISSN (Online) 2191-950X, ISSN (Print) 2191-9496,

Export Citation

© 2018 Walter de Gruyter GmbH, Berlin/Boston.