A Bernstein Theorem for Minimal Maps with Small Second Fundamental Form
Extracted source text ยท arXiv 1711.09878
\PassOptionsToPackage{svgnames}{xcolor}
\documentclass[a4paper,11pt,fleqn,reqno,oneside]{amsart}
\usepackage{lmodern}
\usepackage{mathtools}
\usepackage{amssymb}
\usepackage{amsthm}
\usepackage{amsfonts}
\usepackage{mathrsfs}
\usepackage[english]{babel}
\usepackage{upgreek}
\usepackage{esvect}
\usepackage{extarrows}
\usepackage[noadjust]{cite}
\usepackage[foot]{amsaddr}
% change to italic greek alphabet
\let\Gamma\varGamma
% support color names (used for hyperlinking)
\usepackage{xcolor}
\usepackage{todonotes}
\usepackage[
paper = a4paper,
top = 32mm, bottom = 32mm,
left = 34mm, right = 34mm,
footskip = 10mm, footnotesep=5mm,
]
{geometry}
\usepackage{doi}
\renewcommand{\doitext}{}
% hyperlinks.
\usepackage{hyperref}
\hypersetup{
pdftitle = {A Bernstein Theorem for Minimal Maps with small Second Fundamental Form},
pdfsubject = {},
pdfauthor = {Felix Lubbe},
pdfkeywords = {Minimal submanifolds, Bernstein theorem, ...},
pdfcreator = {},
pdfproducer = {LaTeX with hyperref},
hidelinks,
colorlinks,
linkcolor=blue,
citecolor=blue,
urlcolor=RoyalBlue
}
\newcommand{\arxiv}[1]{\href{http://www.arxiv.org/abs/#1}{arXiv:#1}}
% amsrefs needs to be loaded after hyperref.
\usepackage[sorted-cites]{amsrefs}
\usepackage{etoolbox}
\patchcmd{\section}{\scshape}{\bfseries}{}{}
\makeatletter
\renewcommand{\@secnumfont}{\bfseries}
\patchcmd{\@startsection}
{\@afterindenttrue}
{\@afterindentfalse}
{}{}
\makeatother
\usepackage{enumitem}
% symbols.
\newcommand{\Hb}{\mathbb{H}}
\newcommand{\Cb}{\mathbb{C}}
\newcommand{\Rb}{\mathbb{R}}
\newcommand{\Nb}{\mathbb{N}}
\newcommand{\Pb}{\mathbb{P}}
\newcommand{\Scal}{\mathrm{Scal}}
\newcommand{\Cscr}{\mathscr{C}}
\newcommand{\Lscr}{\mathscr{L}}
\newcommand{\id}{\mathrm{id}}
\newcommand{\oneds}{\mathds{1}}
\newcommand{\Graph}{\mathrm{Graph}}
\newcommand{\End}{\mathrm{End}}
\newcommand{\Sym}{\mathrm{Sym}}
\newcommand{\Span}{\mathrm{span}}
\newcommand{\Ker}{\mathrm{Ker}}
\newcommand{\vol}{\mathrm{vol}}
\newcommand{\Xx}{\mathfrak{X}}
\newcommand{\Ac}{\mathcal{A}}
\newcommand{\Bc}{\mathcal{B}}
\newcommand{\Kc}{\mathcal{K}}
\newcommand{\Lc}{\mathcal{L}}
\newcommand{\ee}{\mathrm{e}}
\newcommand{\Ncal}{\mathcal{N}}
\renewcommand{\O}{\mathrm{O}}
\newcommand{\SU}{\mathit{SU}}
\newcommand{\SO}{\mathit{SO}}
\newcommand{\Sp}{\mathit{Sp}}
\newcommand{\<}{\langle}
\renewcommand{\>}{\rangle}
\renewcommand{\div}{\mathrm{div}}
\newcommand{\A}{\mathrm{A}}
\newcommand{\Hv}{\vv{H}}
\DeclareMathOperator{\tr}{\mathrm{tr}}
\newcommand{\der}{\mathrm{d}}
\newcommand{\ii}{\mathrm{i}}
\newcommand{\jj}{\mathrm{j}}
\newcommand{\kk}{\mathrm{k}}
\DeclareMathOperator*{\esssup}{\mathrm{ess\;sup}}
\DeclareMathOperator*{\rank}{\mathrm{rank}}
\DeclareMathOperator*{\codim}{\mathrm{codim}}
\DeclareMathOperator*{\spann}{\mathrm{span}}
\newcommand{\gMetric}{\mathrm{g}}
\DeclareMathOperator{\GMetric}{G}
\DeclareMathOperator{\dd}{d\hspace{-3pt}}
\DeclareMathOperator{\Ric}{Ric}
\newcommand{\RMN}{\mathrm{R}_{M\times N}}
\newcommand{\RM}{\mathrm{R}_M}
\newcommand{\RN}{\mathrm{R}_N}
\newcommand{\Rind}{\mathrm{R}}
\newcommand{\gM}{\gMetric_M}
\newcommand{\gN}{\gMetric_N}
\newcommand{\gMN}{\gMetric_{M\times N}}
\newcommand{\gind}{\gMetric}
\newcommand{\sTensor}{\mathrm{s}}
\newcommand{\sMN}{\sTensor_{M\times N}}
\newcommand{\sind}{\sTensor}
\newcommand{\piM}{\pi_M}
\newcommand{\piN}{\pi_N}
\newcommand{\svdf}{\lambda} % singular values of \dd f
\newcommand{\evp}{\nu}
\newcommand{\evs}{\kappa}
% theorems.
\newtheorem{mainthm}{Theorem}
\newtheorem{maincor}[mainthm]{Corollary}
\newtheorem{theorem}{Theorem}
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{corollary}[theorem]{Corollary}
\newtheorem{remark}[theorem]{Remark}
\newtheorem*{remark*}{Remark}
\newtheorem{definition}[theorem]{Definition}
\newtheorem*{definition*}{Definition}
\newtheorem{example}[theorem]{Example}
\newtheorem{claim}{Claim}
% adjust numbering
\numberwithin{equation}{section}
\numberwithin{theorem}{section}
\renewcommand*{\themainthm}{\Alph{mainthm}}
% labels.
\renewcommand{\labelenumi}{(\roman{enumi})}
\setlength{\parindent}{0pt}
\allowdisplaybreaks
% title
\title[Minimal Maps with Small Second Fundamental Form]{A Bernstein Theorem for Minimal Maps\\with small Second Fundamental Form}
\author{Felix Lubbe}
\address{Department of Mathematics, University of Hamburg, Bundesstr.\ 55, D--20146 Hamburg, Germany}
\email{Felix.Lubbe@uni-hamburg.de}
\keywords{Minimal maps, Bernstein theorem}
\subjclass[2010]{Primary 53C42; 53C40, 58J05}
% 53C40: Global submanifolds
% 53C42: Immersions (minimal, prescribed curvature, tight, etc.)
% 58J05: Elliptic equations on manifolds, general theory [See also 35-XX]
% document.
\begin{document}
\begin{abstract}
We consider minimal maps $f:M\to N$ between Riemannian manifolds
$(M,\mathrm{g}_M)$ and $(N,\mathrm{g}_N)$, where $M$ is compact and
where the sectional curvatures satisfy $\sec_N\le \sigma\le \sec_M$ for
some $\sigma>0$. Under certain assumptions on the differential of the
map and the second fundamental form of the graph $\Gamma(f)$ of $f$, we show
that $f$ is either the constant map or a totally geodesic isometric immersion.
\end{abstract}
%
%
%
\maketitle
\setcounter{tocdepth}{1}
\tableofcontents
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% INTRODUCTION
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Introduction}
\setlength{\mathindent}{1.5cm} % !! some arbitrary value, needs to be fixed!
We consider smooth minimal maps $f:M\to N$ between Riemannian
manifolds $(M,\gM)$ and $(N,\gN)$. The map $f$
is called \emph{minimal}, if its graph
\[
\Gamma(f) \coloneqq \bigl\{ (x,f(x)) : x\in M \bigr\} \qquad \subset M\times N
\]
is a minimal submanifold of $\bigl( M\times N, \gM\times\gN \bigr)$ \cite{Sch93}.
The Bernstein theorem asserts that any complete minimal surface in $\Rb^3$ which can
be written as a graph of a function on $\Rb^2$ must be a plane. This result
has been generalized to $\Rb^n$ for $n\le 7$ and to general dimensions
under various growth conditions (see e.\,g.\ \cite{EH90} and references therein). \\
A generalized Bernstein problem is to ask under which additional geometric conditions
the graph $\Gamma(f)$ is totally geodesic. Several results for higher codimension have
been obtained, e.\,g.\ by assuming conditions on the slope of the graph
and by considering its Gauss map \cite{FC80,HJW80,JX99,SWX06,JXY13,JXY16,Wang03}. \\
Let us assume that $M$ is compact.
A further approach to the problem is to demand a volume-decreasing condition for the map $f$.
The map $f$ is called \emph{weakly length-decreasing} if
\[
\|\dd f(v)\|_{\gN} \le \|v\|_{\gM} \qquad \text{for all} \quad v\in\Gamma(TM)
\]
and \emph{weakly area-decreasing} if
\[
\|\dd f(v)\wedge \dd f(w)\|_{\gN} \le \|v\wedge w\|_{\gM} \qquad \text{for all} \quad v,w\in\Gamma(TM) \,.
\]
The length-decreasing property may be expressed in terms of the symmetric
tensor
\[
\sind \coloneqq \gM - f^*\gN \,,
\]
i.\,e.\ $f$ is weakly length-decreasing precisely if $\sind$ is non-negative definite.
This tensor satisfies a nice differential equation,
and allows the application
of a tensorial maximum principle (see \cite{SHS13}, also for the area-decreasing case). \\
Instead of considering the above tensor and its eigenvalues directly,
we will study the behavior of minimal maps in relation to
the trace
\[
\tr(\sind) \coloneqq \sum_{k=1}^m \sind(e_k,e_k) \,,
\]
where $\{e_1,\dotsc,e_m\}$ is a local frame of $TM$, orthonormal with respect to the metric $\gind$
induced by the embedding $\Gamma(f)\subset M\times N$. We remark
that $\frac{1}{m}\tr(\sind)$ is the mean value of the eigenvalues of $\sind$,
i.\,e.\ the trace also may be seen as an average.
After these preparations, we are able to state the main result.
\begin{mainthm}
\label{thm:ThmA}
Let $(M,\gM)$ and $(N,\gN)$ be Riemannian manifolds and
suppose $M$ to be compact of dimension $\dim M \ge 2$.
Assume that there exists $\sigma > 0$,
such that the sectional curvatures $\sec_{M}$ of $M$
and $\sec_{N}$ of $N$ satisfy the relation
\[
\sec_{N} \le \sigma \le \sec_{M} \,.
\]
Let $f:M\to N$ be a minimal map and assume that the tensor $\sind$ satisfies
the condition%
\begin{equation}
\label{eq:trCon}
\tr(\sind) \ge 0 \,.
\end{equation}
Further, assume there is a finite constant $\kappa$, such that
\[
\kappa^{2} > 1 \qquad \text{and} \qquad f^{*}\gN < \kappa^{2} \gM
\]
and such that
the second fundamental tensor $\A$ of the graph satisfies the bound
\begin{equation}
\label{eq:sCondition4}
\|\A\|^{2} \le \frac{\kappa^{2} \sigma}{\kappa^{4}-1} \tr(\sind) \,.
\end{equation}
Then one of the following holds.
\begin{enumerate}[label=(\roman*)]
\item $f$ is the constant map.
\item\label{it:thmAii} $f$ is a totally geodesic isometric immersion,
$\sec_M=\sigma$
and the restriction of $\sec_N$
to $\dd f(TM)$ is equal to $\sigma$. Moreover, $\Gamma(f)$ is a totally
geodesic submanifold of $M\times N$.
\end{enumerate}
\end{mainthm}
If the target space has at most half the dimension of the domain,
the condition on the trace is automatically satisfied, i.\,e.\
the following statement holds.
\begin{maincor}
\label{cor:CorB}
Let $(M,\gM)$ be compact of
dimension $m=\dim M\ge 2$,
let $(N,\gN)$ be of dimension at most $\frac{m}{2}$,
and assume $\sec_N\le \sigma \le \sec_M$ for some $\sigma>0$.
If a minimal map $f:M\to N$
satisfies the condition \eqref{eq:sCondition4},
then $f$ is the constant map.
\end{maincor}
Let us shortly describe the structure of the paper.
In section \ref{sec:SEMP}, we recall the strong elliptic maximum
principle in vector bundles \cite{SHS13}, as required by the subsequent sections.
Section \ref{sec:Maps} describes the geometry of graphs, and
section \ref{sec:Proofs} provides the proof of theorem \ref{thm:ThmA}.
The paper ends with a short discussion of the assumptions
in section \ref{sec:Discussion}. \\
\textbf{Acknowledgements.}
I am grateful to
Andreas Savas-Halilaj for
valuable discussions. This research was initiated while I was
supported
by the Research Training Group 1463 of the DFG
at the Leibniz Universit\"{a}t Hannover.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% THE STRONG ELLIPTIC MAXIMUM PRINCIPLE
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{The Strong Elliptic Maximum Principle}
\label{sec:SEMP}
In \cite{SHS13}, Savas-Halilaj and Smoczyk provided a strong
elliptic maximum principle for vector bundles, which we would
like to apply. We briefly collect the results relevant for us.
All manifolds will be smooth and connected without boundary. \\
Let $(E,\pi,M)$ be a vector bundle of rank $k$ over a smooth
manifold $M$. Suppose $\gMetric_E$ is a bundle metric on $E$
and that $\nabla$ is a metric connection on $E$. A uniformly
elliptic operator $\Lscr$ on $\Gamma(E)$ of second order is
locally given by
\[
\Lscr = \sum_{i,j=1}^m a^{ij} \nabla_{e_i,e_j}^2 + \sum_{j=1}^m b^j \nabla_{e_j} \,,
\]
where $a\in\Gamma(TM\otimes TM)$ is a symmetric, uniformly positive definite
tensor and $b\in\Gamma(TM)$ is a smooth vector field, such that
\[
a = \sum_{i,j=1}^m a^{ij} e_i\otimes e_j \qquad \text{and} \qquad b = \sum_{j=1}^m b^j e_j
\]
in a local frame field $\{e_1,\dotsc,e_m\}$ of $TM$. \\
The following definition goes back to Hamilton \cite[Sec.\ 9]{Ham82}.
\begin{definition}
A fiberwise map $\Psi:\Sym(E^*\otimes E^*) \to \Sym(E^*\otimes E^*)$
is said to satisfy the null-eigenvector condition, if whenever $\vartheta$
is a non-negative symmetric $2$-tensor at a point $x\in M$ and if $v\in T_xM$
is a null-eigenvector of $\vartheta$, then $\Psi(\vartheta)(v,v)\ge 0$.
\end{definition}
The elliptic analogue of Hamilton's maximum principle \cite{Ham86}*{Lemma 8.2}
is given by the next theorem.
\begin{theorem}[{\cite{SHS13}*{Theorem 2.3}}]
\label{thm:SMP}
Let $(M,\gM)$ be a Riemannian manifold and let $(E,\pi,M)$ be a Riemannian
vector bundle over the manifold $M$ equipped with a metric connection.
Suppose that $\phi\in\Sym(E^*\otimes E^*)$ is non-negative definite and
satisfies
\[
\Lscr \phi + \Psi(\phi) = 0 \,,
\]
where $\Psi$ is a smooth fiberwise map satisfying the null-eigenvector condition.
If there is an interior point of $M$ where $\phi$ has a null-eigenvalue, then
$\phi$ must have a null-eigenvalue everywhere.
\end{theorem}
For $\phi\in\Sym(E^*\otimes E^*)$ a real number $\lambda$ is called
\emph{eigenvalue} of $\phi$ with respect to $\gMetric_E$ at the point
$x\in M$, if there exists a non-zero vector $v\in E_x\coloneqq \pi^{-1}(x)$, such that
\[
\phi(v,w) = \lambda \gMetric_E(v,w) \,,
\]
for any $w\in E_x$. Since the tensor $\phi$ is symmetric, it admits
$k$ real eigenvalues $\lambda_1(x),\dotsc,\lambda_k(x)$ at each point $x\in M$.
We will always arrange the eigenvalues, such that
\[
\lambda_1(x)\le\dotsb\le \lambda_k(x)\,.
\]
The following result is also due to Hamilton \cite{Ham82} (see also \cite{SHS13}*{Theorem 2.4}).
\begin{theorem}[Second Derivative Criterion]
\label{thm:SDC}
Suppose that $(M,\gM)$ is a Riemannian manifold
and $(E,\pi,M)$ a Riemannian vector bundle of rank $k$ over $M$ equipped
with a metric connection $\nabla$. Let $\phi\in\Sym(E^*\otimes E^*)$
be a smooth symmetric $2$-tensor. If the largest eigenvalue $\lambda_k$
of $\phi$ admits a local maximum $\lambda$ at an interior point $x_0\in M$, then
\[
(\nabla \phi)(v,v) = 0 \qquad \text{and} \qquad (\Lscr\phi)(v,v) \le 0 \,,
\]
for all vector $v$ in the eigenspace
\[
\{ v \in E_{x_0} : \phi(v,w) = \lambda \gMetric_E(v,w) \;\; \text{for all} \;\; w\in E_{x_0} \}
\]
of $\lambda$ at $x_0$ and for all uniformly
elliptic second order operators $\Lscr$.
\end{theorem}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% MAPS BETWEEN MANIFOLDS
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Maps between Manifolds}
\label{sec:Maps}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% GEOMETRY OF GRAPHS
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\subsection{Geometry of Graphs}
We recall the geometric quantities in a graphical
setting, where we follow the presentation in \cite[Section 3.1]{SHS13}. \\
Let $(M,\gM)$ and $(N,\gN)$ be Riemannian
manifolds of dimensions $m$ and $n$, respectively. The induced
metric on the product manifold will be denoted by
\[
\gMN = \gM\times \gN\,.
\]
A smooth map $f:M\to N$ defines an embedding $F:M\to M\times N$, by
\[
F(x) = \big(x,f(x)\big)\,,\qquad x\in M\,.
\]
The graph of $f$ is defined to be the submanifolds $\Gamma(f) \coloneqq F(M)$.
Since $F$ is an embedding, it induces another Riemannian metric
$\gMetric \coloneqq F^{*}\gMN$ on $M$. The two natural
projections
\[
\pi_{M} : M\times N \to M\,,\qquad \pi_{N}:M\times N\to N\,,
\]
are submersions, that is they are smooth and have maximal rank. Note
that the tangent bundle of the product manifold $M\times N$ splits
as a direct sum
\[
T(M\times N) = TM \oplus TN\,.
\]
The four metrics $\gM,\gN,\gMN$ and $\gMetric$
are related by
\begin{align*}
\gMN &= \pi_{M}^{*} \gM + \pi_{N}^{*}\gN \,, \\
\gMetric &= F^{*}\gMN = \gM + f^{*}\gN \,.
\end{align*}
Now let us define the symmetric $2$-tensors
\begin{align*}
\sTensor_{M\times N} &\coloneqq \pi_{M}^{*}\gM - \pi_{N}^{*}\gN\,, \\
\sTensor &\coloneqq F^{*}\sTensor_{M\times N} = \gM - f^{*}\gN \,.
\end{align*}
Note that $\sTensor_{M\times N}$ is a semi-Riemannian metric of signature $(m,n)$ on the
manifold $M\times N$. \\
The Levi-Civita connection $\nabla^{\gMN}$ associated to the
Riemannian metric $\gMN$ on $M\times N$ is related to the Levi-Civita
connections $\nabla^{\gM}$ on $(M,\gM)$ and $\nabla^{\gN}$
on $(N,\gN)$ by
\[
\nabla^{\gMN} = \pi_{M}^{*}\nabla^{\gM} \oplus \pi_{N}^{*} \nabla^{\gN}\,.
\]
The corresponding curvature operator $\RMN$ on $M\times N$ with respect
to the metric $\gMN$ is related to the curvature operators $\RM$ on
$(M,\gM)$ and $\RN$ on $(N,\gN)$ by
\[
\RMN = \pi_{M}^{*}\RM \oplus \pi_{N}^{*}\RN\,.
\]
We denote the Levi-Civita connection on $M$ with respect to the induced
metric $\gMetric=F^{*}\gMN$ by $\nabla$ and the corresponding curvature
tensor by $\Rind$. \\
The second fundamental tensor $\A$ of the graph $\Gamma(f)$ is
defined as
\[
\A(v,w) \coloneqq \big( \widetilde{\nabla}\dd F\big)(v,w) = \nabla^{\gMN}_{\dd F(v)} \dd F(w) - \dd F(\nabla_{v}w) \,,
\]
where $v,w\in \Gamma(TM)$ and $\widetilde{\nabla}$ is the connection on $F^{*}T(M\times N)\otimes T^{*}M$ induced by the Levi-Civita connection.
The trace of $\A$ with respect to the metric $\gMetric$ is called the
\emph{mean curvature vector field} of $\Gamma(f)$ and it will be denoted by
\[
\Hv \coloneqq \tr \A \,.
\]
Note that $\Hv$ is a section in the normal bundle of the graph.
If $\Hv$ vanishes identically, the graph is said to be minimal.
A map $f:M\to N$ between Riemannian manifolds is called \emph{minimal},
if its graph $\Gamma(f)$ is a minimal submanifold of the product
space $(M\times N,\gMN)$ \cite{Sch93}.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% SINGULAR VALUE DECOMPOSITION
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\subsection{Singular Value Decomposition}
\label{sec:SVD}
We recall the singular value decomposition theorem (see \cite[Section 3.2]{SHS13}
and the references cited there). \\
Fix a point $x\in M$, and let
\[
\svdf_{1}^{2}(x) \le \svdf_{2}^{2} \le \hdots \le \svdf_{m}^{2}(x)
\]
be the eigenvalues of $f^{*}\gN$ with respect to $\gM$. The
corresponding values $\svdf_{i}\ge 0$, $i\in\{1,\hdots,m\}$, are called the
\emph{singular values} of the differential $\dd f$ of $f$ and give rise
to continuous functions on $M$. Let
\[
r \coloneqq r(x) \coloneqq \rank \dd f(x) \,.
\]
Obviously, $r\le \min\{m,n\}$ and $\svdf_{1}(x)=\cdots=\svdf_{m-r}(x)=0$. At the
point $x$ consider an orthonormal basis $\{\alpha_{1},\hdots,\alpha_{m-r};\alpha_{m-r+1},\hdots,\alpha_{m}\}$
with respect to $\gM$ which diagonalizes $f^{*}\gN$. Moreover, at $f(x)$
consider a basis $\{\beta_{1},\hdots,\beta_{n-r};\beta_{n-r+1},\hdots,\beta_{n}\}$ that is
orthonormal
with respect to $\gN$, and such that
\[
\dd f(\alpha_{i}) = \svdf_{i}(x) \beta_{n-m+i}\,,
\]
for any $i\in\{m-r+1,\hdots,m\}$. The above procedure is called the
\emph{singular value decomposition} of the differential $\dd f$. \\
We construct a special basis for the tangent and the normal
space of the graph in terms of the singular values. The vectors
\[
\widetilde{e}_i \coloneqq \begin{cases}
\alpha_{i}\,, & 1\le i\le m-r\,, \\
\frac{1}{\sqrt{1+\svdf_{i}^{2}(x)}}\big( \alpha_{i} \oplus \svdf_{i}(x)\beta_{n-m+i} \big)\,, &m-r+1 \le i \le m\,,
\end{cases}
\]
form an orthonormal basis with respect to the metric $\gMN$ of
the tangent space $\dd F(T_{x}M)$ of the graph $\Gamma(f)$ at $x$.
It follows that with respect to the induced metric $\gind$, the vectors
\[
e_i \coloneqq \frac{1}{\sqrt{1+\svdf_i^2}} \alpha_i
\]
form an orthonormal basis of $T_xM$.
Moreover,
the vectors
\[
\xi_{i} \coloneqq \begin{cases}
\beta_{i}\,, & 1 \le i \le n-r\,, \\
\frac{1}{\sqrt{1+\svdf_{i+m-n}^{2}(x)}}\bigl(-\svdf_{i+m-n}(x)\alpha_{i+m-n} \oplus \beta_{i}\bigr)\,, & n-r+1\le i\le n\,,
\end{cases}
\]
form an orthonormal basis with respect to $\gMN$ of the
normal space $\mathcal{N}_{x}M$ of the graph $\Gamma(f)$ at the point
$x$. From the formulae above, we deduce that
\[
\sTensor_{M\times N}(\widetilde{e}_i,\widetilde{e}_j) = \sind(e_i,e_j) = \frac{1-\svdf_{i}^{2}(x)}{1+\svdf_{i}^{2}(x)}\delta_{ij} \,, \qquad 1\le i,j\le m \,.
\]
Therefore, the eigenvalues of the $2$-tensor $\sTensor$ with respect to $\gMetric$
are given by
\[
\frac{1-\svdf_{1}^{2}(x)}{1+\svdf_{1}^{2}(x)} \ge \cdots \ge \frac{1-\svdf_{m-1}^{2}(x)}{1+\svdf_{m-1}^{2}(x)} \ge \frac{1-\svdf_{m}^{2}(x)}{1+\svdf_{m}^{2}(x)} \,.
\]
Moreover,
\[
\sTensor_{M\times N}(\xi_{i},\xi_{j}) =
\begin{cases}
- \delta_{ij}\,, & 1 \le i \le n-r \,, \\
- \frac{1-\svdf_{i+m-n}^{2}(x)}{1+\svdf_{i+m-n}^{2}(x)}\delta_{ij} \,, & n-r+1\le i \le n\,,
\end{cases}
\]
and
\[
\sTensor_{M\times N}(\widetilde{e}_{m-r+i},\xi_{n-r+j}) = - \frac{2\svdf_{m-r+i}(x)}{1+\svdf_{m-r+i}^{2}(x)} \delta_{ij} \,,\qquad 1\le i,j\le r \,.
\]
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Bernstein-type Theorems
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\section{Bernstein-type Theorems}
\label{sec:Proofs}
In order to show theorem \ref{thm:ThmA}, we need to control the eigenvalues of the tensor $\sTensor$.
For this, following the ideas of \cite{SHS13}, let us define the tensor
\[
\Phi_{c} \coloneqq \sTensor - \frac{1-c}{1+c} \gMetric \,.
\]
The differential equation satisfied by $\Phi_c$
was derived in \cite[Lemma 3.2]{SHS13} and is given by
\begin{align*}
\big( \Delta \Phi_{c} \big) (v,w) &= \sTensor_{M\times N}\big( \nabla_{v} \Hv, \dd F(w) \big) + \sTensor_{M\times N}\big( \nabla_{w}\Hv, \dd F(v) \big) \\
& \qquad + 2 \frac{1-c}{1+c} \gMN\big( \Hv, \A(v,w) \big) \\
& \qquad + \Phi_{c}\big( \Ric v, w \big) + \Phi_{c} \big( \Ric w, v \big) \\
& \qquad + 2 \sum_{k=1}^{m} \left( \sTensor_{M\times N} - \frac{1-c}{1+c} \gMN \right) \big( \A(e_{k},v), \A(e_{k},w) \big) \\
& \qquad + \frac{4}{1+c} \sum_{k=1}^{m} \big( f^{*}\RN(e_{k},v,e_{k},w) - c \RM(e_{k},v,e_{k},w) \big) \,,
\end{align*}
where
\[
\Ric v \coloneqq - \sum_{k=1}^{m} \Rind(e_{k},v) e_{k}
\]
is the Ricci operator on $(M,\gMetric)$ and $\{e_{1},\hdots,e_{m}\}$ is any orthonormal
frame with respect to the induced metric $\gMetric$. \\
Now let $f:M\to N$ be minimal, i.\,e.\ $\Hv=0$.
Setting
\begin{align*}
\Psi_c(\vartheta)(v,w) &\coloneqq - \vartheta(\Ric v,w) - \vartheta(\Ric w,v) \\
& \qquad - 2 \sum_{k=1}^{m} \left( \sTensor_{M\times N} - \frac{1-c}{1+c} \gMN \right) \big( \A(e_{k},v), \A(e_{k},w) \big) \\
& \qquad - \frac{4}{1+c} \sum_{k=1}^{m} \big( f^{*}\RN(e_{k},v,e_{k},w) - c \RM(e_{k},v,e_{k},w) \big) \,,
\end{align*}
we can write
\[
\Delta \Phi_c + \Psi_c(\Phi_c) = 0\,.
\]
To estimate the terms in $\Psi_c$ which contain the second
fundamental tensor, we have the following statement.
\begin{lemma}
\label{lem:AEst}
Let $\eta\in T_{p}^{\perp}M$ be a normal vector on the graph.
Then for any $c\ge\svdf_{m}^{2}$ the estimate
\[
\sMN(\eta,\eta) \le \frac{c-1}{1+c} \gMN(\eta,\eta)
\]
holds. In particular,
\[
\sMN\big( \A(v,w), \A(v,w) \bigr) \le \frac{c-1}{1+c} \|\A(v,w)\|^2 \,.
\]
\end{lemma}
\begin{proof}
Using the formulae for the singular values above, we calculate
\[
\sTensor_{M\times N}(\xi_{i},\xi_{j}) = - \delta_{ij} \le - \frac{1-c}{1+c} \delta_{ij} \,, \qquad 1\le i,j\le n-r
\]
and
\begin{align*}
\sTensor_{M\times N}(\xi_{i},\xi_{j}) &= - \frac{1-\svdf_{i+m-n}^{2}}{1+\svdf_{i+m-n}^{2}} \delta_{ij} \\
&\le - \frac{1-c}{1+c} \delta_{ij} \,, \qquad \qquad n-r+1 \le i,j\le n \,. \qedhere
\end{align*}
\end{proof}
The terms involving the curvatures can be decomposed in the following way.
\begin{lemma}
\label{lem:PhiNullEV}
Let $\{e_1,\dotsc,e_m\}$ be a local $\gind$-orthonormal frame. Then for any $e_l$ we have
\begin{align*}
& 2 \sum_{k=1}^{m}\big( f^{*}\RN(e_{k},e_{l},e_{k},e_{l}) - c \RM(e_{k},e_{l},e_{k},e_{l}) \big) \\
& \quad = -2 \sum_{k\ne l} f^{*}\gN(e_{k},e_{k}) \biggl\{ \big( \sigma - \sec_{N}(\dd f(e_{k})\wedge\dd f(e_{l})) \big) f^*\gN(e_l,e_l) \\
& \qquad \qquad \qquad \qquad \qquad \qquad + \sigma \Phi_c(e_l,e_l) \biggr\} \\
& \qquad - c \gM(e_l,e_l) \sum_{k\ne l} \Phi_c(e_k,e_k) \bigl( \sec_M(e_k\wedge e_l) - \sigma \bigr) \\
& \qquad - \frac{2c}{1+c} \bigl( \Ric_M(e_l,e_l) - (m-1) \sigma \gM(e_l,e_l) \bigr)\\
& \qquad - \frac{2\sigma c}{1+c} \bigl( \tr(\sind) - \sind(e_l,e_l) \bigr) \\
& \qquad - \frac{\sigma(1+c)}{2} \Phi_c(e_l,e_l) \left( \tr\bigl(\Phi_c\bigr) - \Phi_c(e_l,e_l) \right) \,.
\end{align*}
\end{lemma}
\begin{proof}
We calculate
\begin{align*}
& 2 \sum_{k=1}^{m}\big( f^{*}\RN(e_{k},e_{l},e_{k},e_{l}) - c \RM(e_{k},e_{l},e_{k},e_{l}) \big) \\
& \quad = 2 \sum_{k\ne l} \sec_{N}\big( \dd f(e_{k})\wedge\dd f(e_{l}) \big) f^{*}\gN(e_{k},e_{k}) f^{*}\gN(e_{l},e_{l}) \\
& \qquad - 2c \sum_{k\ne l} \sec_{M}(e_{k}\wedge e_{l}) \gM(e_{k},e_{k}) \gM(e_{l},e_{l}) \\
& \quad = -2 \sum_{k\ne l} f^{*}\gN(e_{k},e_{k})\biggl\{ \bigl( \sigma - \sec_{N}(\dd f(e_{k})\wedge\dd f(e_{l})) \bigr) f^{*}\gN(e_{l},e_{l}) \\
& \qquad \qquad \qquad \qquad \qquad \qquad + \sigma\biggl( \gM(e_{l},e_{l}) - f^{*}\gN(e_{l},e_{l}) - \frac{1-c}{1+c} \biggr) \biggr\} \\
& \qquad + 2\sigma \left( \gM(e_{l},e_{l}) - \frac{1-c}{1+c} \right) \sum_{k\ne l} f^{*}\gN(e_{k},e_{k}) \\
& \qquad - 2c\sum_{k\ne l} \bigl( \sec_M(e_k\wedge e_l) - \sigma \bigr) \gM(e_k,e_k) \gM(e_l,e_l) \\
& \qquad - 2c\sigma \sum_{k\ne l} \gM(e_k,e_k) \gM(e_l,e_l) \,.
\end{align*}
Since
\[
2 \gM(e_k,e_k) = \sind(e_k,e_k) + 1 = \left( \sind(e_k,e_k) - \frac{1-c}{1+c} \right) + \frac{2}{1+c} \,,
\]
the term involving the sectional curvatures of $M$ evaluates to
\begin{align*}
& - 2c\sum_{k\ne l} \bigl( \sec_M(e_k\wedge e_l) - \sigma \bigr) \gM(e_k,e_k) \gM(e_l,e_l) \\
& \quad = - c \gM(e_l,e_l) \sum_{k\ne l} \left( \sind(e_k,e_k) - \frac{1-c}{1+c} \right) \bigl( \sec_M(e_k\wedge e_l) - \sigma \bigr) \\
& \qquad - \frac{2c}{1+c} \bigl( \Ric_M(e_l,e_l) - (m-1) \sigma \gM(e_l,e_l) \bigr) \,.
\end{align*}
Further, using
\[
2 \gM(e_k,e_k) = 1 + \sind(e_k,e_k) \,, \qquad 2 f^*\gN(e_k,e_k) = 1 - \sind(e_k,e_k) \,,
\]
for the remaining terms we get
\begin{align*}
& 2\sigma \left( \gM(e_{l},e_{l}) - \frac{1-c}{1+c} \right) \sum_{k\ne l} f^{*}\gN(e_{k},e_{k}) - 2c\sigma \sum_{k\ne l} \gM(e_k,e_k) \gM(e_l,e_l) \\
& \quad = \frac{2\sigma c}{1+c} \sum_{k\ne l} \bigl( f^*\gN(e_k,e_k) - \gM(e_k,e_k) \bigr) \\
& \qquad \quad + 2\sigma \left( \gM(e_l,e_l) - \frac{1}{1+c} \right) \sum_{k\ne l} \bigl( f^*\gN(e_k,e_k) - c\gM(e_k,e_k) \bigr) \\
& \quad = - \frac{2\sigma c}{1+c} \bigl( \tr(\sind) - \sind(e_l,e_l) \bigr) \\
& \qquad \quad + \sigma \left( \sind(e_l,e_l) - \frac{1-c}{1+c} \right) \sum_{k\ne l}\left( \frac{1-c}{2} - \frac{1+c}{2} \sind(e_k,e_k) \right) \\
& \quad = - \frac{2\sigma c}{1+c} \bigl( \tr(\sind) - \sind(e_l,e_l) \bigr) \\
& \qquad \quad + \frac{1+c}{2} \sigma \Phi_c(e_l,e_l) \left( (m-1)\frac{1-c}{1+c} - \tr(\sind) + \sind(e_l,e_l) \right) \,.
\end{align*}
Collecting all terms and using $\sind=\Phi_c+\frac{1-c}{1+c}\gind$, the claim of the lemma follows.
\end{proof}
\begin{corollary}
\label{cor:Psi}
Let $\{e_1,\dotsc,e_m\}$ be a local $\gind$-orthonormal frame. Then for any $e_l$ we have
\begin{align*}
& 2 \sum_{k=1}^{m}\big( f^{*}\RN(e_{k},e_{l},e_{k},e_{l}) - c \RM(e_{k},e_{l},e_{k},e_{l}) \big) \\
& \quad = -2 \sum_{k\ne l} f^{*}\gN(e_{k},e_{k}) \big( \sigma - \sec_{N}(\dd f(e_{k})\wedge\dd f(e_{l})) \big) f^*\gN(e_l,e_l) \\
& \qquad - c \gM(e_l,e_l) \sum_{k\ne l} \Phi_c(e_k,e_k) \bigl( \sec_M(e_k\wedge e_l) - \sigma \bigr) \\
& \qquad - \frac{2c}{1+c} \bigl( \Ric_M(e_l,e_l) - (m-1) \sigma \gM(e_l,e_l) \bigr)\\
& \qquad - \frac{2c\sigma }{1+c} \tr(\sind) \\
& \qquad + \frac{\sigma(1-c)}{2} \Phi_c(e_l,e_l) \left( \tr\bigl(\Phi_c\bigr) - \Phi_c(e_l,e_l) \right) \\
& \qquad - \frac{2c\sigma}{1+c} \left( (m-2) \Phi_c(e_l,e_l) - \frac{1-c}{1+c} \right)\,.
\end{align*}
\end{corollary}
\begin{proof}
We note that
\begin{align*}
& -2 \sigma \Phi_c(e_l,e_l) \sum_{k\ne l} f^*\gN(e_k,e_k) \\
& \quad = - \sigma \Phi_c(e_l,e_l) \sum_{k\ne l} \bigl( 1 - \sind(e_k,e_k) \bigr) \\
&\quad = - \sigma \Phi_c(e_l,e_l) \bigl( (m-1) - \tr(\sind) + \sind(e_l,e_l) \bigr) \\
&\quad = - \sigma \Phi_c(e_l,e_l) \biggl( (m-1) - \tr(\Phi_c) - m\frac{1-c}{1+c} + \Phi_c(e_l,e_l) + \frac{1-c}{1+c} \biggr) \\
&\quad = \sigma \Phi_c(e_l,e_l) \Bigl( \tr(\Phi_c) - \Phi_c(e_l,e_l) \Bigr) - (m-1) \frac{2c\sigma}{1+c} \Phi_c(e_l,e_l) \,.
\end{align*}
Then from lemma \ref{lem:PhiNullEV} and using $\sind=\Phi_c+\frac{1-c}{1+c}\gind$ we get
\begin{align*}
& 2 \sum_{k=1}^{m}\big( f^{*}\RN(e_{k},e_{l},e_{k},e_{l}) - c \RM(e_{k},e_{l},e_{k},e_{l}) \big) \\
& \quad = -2 \sum_{k\ne l} f^*\gN(e_k,e_k) \bigl( \sigma - \sec_N(\dd f(e_k) \wedge \dd f(e_l) \bigr) f^*\gN(e_l,e_l) \\
& \qquad + \sigma \Phi_c(e_l,e_l) \Bigl( \tr(\Phi_c) - \Phi_c(e_l,e_l) \Bigr) - (m-1) \frac{2c \sigma}{1+c} \Phi_c(e_l,e_l) \\
& \qquad - c\gM(e_l,e_l) \sum_{k\ne l} \Phi_c(e_k,e_k) \bigl( \sec_M(e_k\wedge e_l) - \sigma \bigr) \\
& \qquad - \frac{2c}{1+c} \bigl( \Ric_M(e_l,e_l) - (m-1) \sigma \gM(e_l,e_l) \bigr) \\
& \qquad - \frac{2\sigma c}{1+c} \tr(\sind) + \frac{2\sigma c}{1+c} \left( \Phi_c(e_l,e_l) + \frac{1-c}{1+c} \right) \\
& \qquad - \frac{\sigma(1+c)}{2} \Phi_c(e_l,e_l) \bigl( \tr(\Phi_c) - \Phi_c(e_l,e_l) \bigr) \,.
\end{align*}
Rearranging the terms, the claim follows.
\end{proof}
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% PROOF OF THEOREM A
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
\subsection{Proof of Theorem \ref{thm:ThmA}}
From the compactness of $M$ we get that the largest singular
value $\lambda_m$ of $\dd f$ attains
its maximum at some point on $M$.
Let $\svdf_{0}$ denote this maximum,
\[
\svdf_{0}^{2} \coloneqq \max_{x\in M} \svdf_{m}^{2}(x) \,.
\]
Then $\Phi_{\svdf_{0}^{2}}$ is non-negative definite, and at the point where $\svdf_{m}^{2}$
attains its maximum, we know that $e_{m}$ is a null-eigenvector of $\Phi_{\svdf_{0}^{2}}$.
\begin{lemma}
\label{lem:StrictLengthDecr}
Assume $\sec_N\le\sigma\le\sec_M$ for some $\sigma>0$ and assume
$\svdf_{0}^2 < 1$. Then the map $f:M\to N$ is constant.
\end{lemma}
\begin{proof}
We first show that
$\Psi_{\svdf_0^2}$ satisfies the null-eigenvector condition. Let
$v$ be a null-eigenvector of the positive semi-definite tensor $\vartheta$
and extend it to a local $\gind$-orthonormal frame $\{e_1,\dotsc,e_l=v,\dotsc,e_m\}$.
Since $\svdf_0^2<1$, the map $f$ is strictly length-decreasing and accordingly it is
$\sind(w,w)>0$ for any vector $w\ne 0$.
Also noting that $\Phi_{\svdf_0^2}(w,w)\ge 0$ for any $w$
by construction and
\begin{align*}
\tr(\sind) - \sind(v,v) &= \sum_{k\ne l} \underbrace{\sind(e_k,e_k)}_{>0} > 0\,, \\
\tr(\Phi_{\svdf_0^2}) - \Phi_{\svdf_0^2}(v,v) &= \sum_{k\ne l} \underbrace{\Phi_{\svdf_0^2}(e_k,e_k)}_{\ge 0} \ge 0 \,,
\end{align*}
we calculate
\begin{align*}
\Psi_{\svdf_0^2} (\vartheta)(v,v) &\stackrel{\text{Lem.\ \ref{lem:AEst}}}{\ge} \underbrace{4 \frac{1-\svdf_0^2}{1+\svdf_0^2} \sum_{k=1}^m \|\A(e_k,v)\|^2}_{\ge 0} \\
& \qquad \qquad - \frac{4}{1+\svdf_0^2} \sum_{k=1}^m \bigl( f^*\RN(e_k,v,e_k,v) - \svdf_0^2 \RM(e_k,v,e_k,v) \bigr) \\
&\stackrel{\hphantom{\text{Lem.\ \ref{lem:AEst}}}}{\ge} \frac{4\sigma \svdf_0^2}{(1+\svdf_0^2)^2}\left( \tr(\sind) - \sind(v,v) \right) \\
& \qquad \qquad + \sigma \Phi_{\svdf_0^2}(v,v) \bigl( \tr(\Phi_{\svdf_0^2}) - \Phi_{\svdf_0^2}(v,v) \bigr) \\
&\stackrel{\hphantom{\text{Lem.\ \ref{lem:AEst}}}}{\ge} 0\,,
\end{align*}
where we have used lemma \ref{lem:PhiNullEV} and the curvature assumptions.
Thus, the null-eigenvector condition is satisfied. \\
By the definition
of $\svdf_0^2$, at the point $x\in M$ where the largest singular value $\svdf_m$
of $\dd f$ attains its maximal value, we have $\Phi_{\svdf_0^2}(e_m,e_m)=0$,
so that $\Phi_{\svdf_0^2}$ has a null-eigenvector at that point.
Consequently, by the strong
maximum principle (see theorem \ref{thm:SMP}),
$\Phi_{\svdf_0^2}$
has a null-eigenvector at any point of $M$. \\
We will now show that if $\Phi_{\svdf_0^2}$ admits a null-eigenvector at some point $x$,
then $\Phi_{\svdf_0^2}$ vanishes at $x$. Since by definition $\Phi_{\svdf_0^2}\ge 0$,
we can apply the second derivative test criterion at an arbitrary point $x\in M$.
Using lemmas \ref{lem:AEst} and \ref{lem:PhiNullEV}, at $x$ we calculate
\begin{align*}
0 &\stackrel{\substack{\hphantom{\text{Lem.\ \ref{lem:PhiNullEV}}}\\\text{min}}}{\le} (\Delta \Phi_{\svdf_0^2})(e_m,e_m) \\
&\stackrel{\text{Lem.\ \ref{lem:AEst}}}{\le} \underbrace{ 4 \frac{\svdf_0^2-1}{1+\svdf_0^2} \sum_{k=1}^m \| \A(e_k,e_m) \|^2}_{\le 0} \\
& \qquad \qquad + \frac{4}{1+\svdf_0^2} \sum_{k=1}^m \bigl( f^*\RN(e_k,e_m,e_k,e_m) - \svdf_0^2 \RM(e_k,e_m,e_k,e_m) \bigr) \\
& \stackrel{\text{Lem.\ \ref{lem:PhiNullEV}}}{\le} \underbrace{- \frac{4\sigma\svdf_0^2}{(1+\svdf_0^2)^2}}_{\le 0} \sum_{k\ne m} \underbrace{\sind(e_k,e_k)}_{>0} = 0 \,.
\end{align*}
Consequently, since $\sigma>0$,
we must have $\svdf_0^2=0$. Since $\svdf_0^2$ is the maximum
of the largest singular value, it is $0 \le \svdf_1^2(x) = \dotsb = \svdf_m^2(x) \le \svdf_0^2 = 0$.
\end{proof}
To prove the remaining case of theorem \ref{thm:ThmA},
we only need to consider the case $\svdf_0^2\ge 1$.
\begin{lemma}
\label{lem:PhiNullEV2}
Assume $\sec_{N} \le \sigma \le \sec_{M}$ for some $\sigma > 0$.
Further, assume $\tr(\sind)\ge 0$ and $\lambda_0^2\ge 1$.
If there exists $\kappa>1$, such that
$f^{*}\gN < \kappa^{2} \gM$ and
\[
\|\A\|^{2} \le \frac{\kappa^{2} \sigma}{\kappa^{4}-1} \tr(\sTensor) \,,
\]
then $\Psi_{\svdf_{0}^{2}}$ satisfies the null-eigenvector condition.
\end{lemma}
\begin{proof}
Let
$v$ be a null-eigenvector of the positive semi-definite tensor $\vartheta$
and extend it to a local $\gind$-orthonormal frame $\{e_1,\dotsc,e_l=v,\dotsc,e_m\}$.
Note that $\Phi_{\svdf_0^2}(v,v)\ge 0$ by construction
and consequently
\[
\tr(\Phi_{\svdf_0^2}) - \Phi_{\svdf_0^2}(v,v) = \sum_{k\ne l} \underbrace{\Phi_{\svdf_0^2}(e_k,e_k)}_{\ge 0} \ge 0 \,.
\]
From the curvature assumptions, lemma \ref{lem:AEst} and corollary \ref{cor:Psi}
we obtain
\begin{align*}
\Psi_{\svdf_0^2}(\vartheta)(v,v) &\stackrel{\text{Lem.\ \ref{lem:AEst}}}{\ge} -4 \frac{\svdf_0^2-1}{1+\svdf_0^2} \sum_{k=1}^m \| \A(e_k,v) \|^2 \\
&\qquad \qquad - \frac{4}{1+\svdf_0^{2}} \sum_{k=1}^{m}\big( f^{*}\RN(e_{k},v,e_{k},v) - \svdf_0^{2}\RM(e_{k},v,e_{k},v) \big) \\
& \stackrel{\substack{\hphantom{Lem.\ \ref{lem:AEst}}\\\svdf_0^2\ge 1\\\text{Cor.\ \ref{cor:Psi}}}}{\ge} - \frac{4}{1+\svdf_0^{2}}\left\{ (\svdf_0^{2}-1) \|\A\|^{2} - \frac{\svdf_0^{2}}{1+\svdf_0^{2}} \sigma \tr(\sind) \right\} \,.
\end{align*}
If $\lambda_0^2=1$, then
\[
\Psi_{\svdf_0^2}(\vartheta)(v,v) \ge \sigma \tr(\sind) \ge 0 \,.
\]
so that the null-eigenvector condition is satisfied. If $\lambda_0^2>1$,
also using $\svdf_{0}^{2} < \kappa^{2}$, it is
\[
\frac{\kappa^{2}}{\kappa^{4}-1} < \frac{\svdf_{0}^{2}}{\svdf_{0}^{4}-1}
\]
and accordingly
\begin{align*}
\Psi_{\svdf_0^2}(\vartheta)(v,v) & \ge - \frac{4}{1+\svdf_0^2}\left\{ (\svdf_0^{2}-1) \|\A\|^{2} - \frac{\svdf_0^{2}}{1+\svdf_0^{2}} \sigma \tr(\sTensor) \right\} \\
& = - 4 \frac{\svdf_0^2-1}{1+\svdf_0^{2}}\left\{ \|\A\|^{2} - \frac{\svdf_0^{2}}{\svdf_0^4-1} \sigma \tr(\sTensor) \right\} \\
& \ge - 4 \frac{\svdf_0^2-1}{1+\svdf_0^2} \left\{ \frac{\kappa^2}{\kappa^4-1} - \frac{\svdf_0^2}{\svdf_0^4-1} \right\} \sigma \tr(\sind) \\
&\ge 0 \,,
\end{align*}
so that the null-eigenvector condition is satisfied.
\end{proof}
Let us now finish the proof of theorem \ref{thm:ThmA}.
By construction, the non-negative definite tensor
$\Phi_{\svdf_0^2}$ admits a null-eigenvector at some point.
By lemma \ref{lem:PhiNullEV2}, we can apply the strong
maximum principle (see theorem \ref{thm:SMP})
to conclude that
$\Phi_{\svdf_{0}^{2}}$ admits a null-eigenvector everywhere on $M$,
i.\,e.\ $\Phi_{\svdf_0^2}$ attains its minimal value at every point on $M$.
Thus, we may apply the second derivative test criterion to $\Phi_{\svdf_0^2}$ at
an arbitrary point $x\in M$. At $x$ consider a basis $\{e_1,\dotsc,e_m\}$,
orthonormal with respect to $\gind$ consisting of eigenvectors of $\Phi_{\svdf_0^2}$,
such that $e_m$
is a null-eigenvector of $\Phi_{\svdf_0^2}$ and $\svdf_m^2(x)=\svdf_0^2$.
From lemma \ref{lem:AEst} and corollary \ref{cor:Psi}
we conclude
\begin{align}
\nonumber 0 &\stackrel{\text{min}}{\le} \big( \Delta \Phi_{\svdf_{0}^{2}} \big)(e_{m},e_{m}) \\
\nonumber & \stackrel{\text{\phantom{min}}}{\le} \underbrace{\frac{4}{1+\svdf_{0}^{2}}}_{> 0} \Biggl( \underbrace{ (\svdf_{0}^{2}-1) \|\A\|^{2} - \frac{\svdf_{0}^{2}}{1+\svdf_{0}^{2}}\sigma \tr(\sTensor)}_{\le 0} \Biggr) \\
\nonumber &\quad - \underbrace{\frac{4}{1+\svdf_0^2}}_{> 0} \sum_{k\ne m} \underbrace{f^{*}\gN(e_{k},e_{k})}_{\ge 0} \big( \underbrace{\sigma-\sec_{N}(\dd f(e_{k})\wedge \dd f(e_{m}))}_{\ge 0}\big) \\
\nonumber &\quad - \underbrace{\frac{4\svdf_{0}^{2}}{(1+\svdf_0^2)^2}}_{\ge 0} \sum_{k\ne m} \underbrace{\Phi_{\svdf_0^2}(e_m,e_m)}_{\ge 0} \big( \underbrace{\sec_{M}(e_{k}\wedge e_{m}) - \sigma}_{\ge 0}\big) \\
\nonumber &\quad - \underbrace{\frac{4\svdf_{0}^{2}}{(1+\svdf_{0}^{2})^{2}}}_{\ge 0}\Big( \underbrace{\Ric_{M}(e_{m},e_{m}) - (m-1)\sigma \gM(e_{m},e_{m})}_{\ge 0}\Big) \\
\nonumber &\quad + \underbrace{\frac{1-\svdf_0^2}{1+\svdf_0^2}}_{\le 0} \sigma \underbrace{\Phi_{\svdf_0^2}(e_m,e_m)}_{\ge 0} \Bigl( \underbrace{\tr(\Phi_{\svdf_0^2}) - \Phi_{\svdf_0^2}(e_m,e_m)}_{\ge 0} \Bigr) \\
\label{eq:LapPhi} &\quad - \underbrace{\frac{4\svdf_0^2}{(1+\svdf_0^2)^2}}_{> 0} \sigma \biggl( \underbrace{(m-2) \Phi_{\svdf_0^2}(e_m,e_m)}_{\ge 0} \underbrace{-\frac{1-\svdf_0^2}{1+\svdf_0^2}}_{\ge 0} \biggr) = 0 \,.
\end{align}
Since all terms have the same sign, they have to vanish independently.
Using $\sigma>0$ and $\lambda_0^2\ge 1$, the second line and the
assumptions on $\|\A\|^2$ imply
\begin{align*}
0 &= \bigl( \svdf_0^2-1 \bigr) \|\A\|^2 - \frac{\svdf_0^2}{1+\svdf_0^2} \sigma \tr(\sind) \\
& \le \left( \bigl( \svdf_0^2-1 \bigr) \frac{\kappa^2}{\kappa^4-1} - \frac{\svdf_0^2}{1+\svdf_0^2} \right) \sigma \tr(\sind) \stackrel{\svdf_0^2<\kappa^2}{\le} 0 \,.
\end{align*}
Since $\sigma>0$, this yields $\tr(\sind)=0$. From
\[
0 = - \underbrace{\frac{4\svdf_0^2}{(1+\svdf_0^2)^2}}_{> 0} \sigma \biggl( \underbrace{(m-2) \Phi_{\svdf_0^2}(e_m,e_m)}_{\ge 0} \underbrace{-\frac{1-\svdf_0^2}{1+\svdf_0^2}}_{\ge 0} \biggr)
\]
we infer $\svdf_0^2=1$.
Consequently, $\tr(\sind)=0$
and $\svdf_1^2(x) \le \dotsb \le \svdf_m^2(x)=\lambda_0^2 = 1$ force
all singular values to be equal to one at any point $x\in M$,
$\svdf_1^2(x) = \dotsb = \svdf_m^2(x) = 1$. Now the assumption on the second
fundamental form reads
\[
\|\A\|^2 \le \frac{\kappa^2 \sigma}{\kappa^4-1} \tr(\sind) = 0 \,,
\]
so that $\Gamma(f)$ is totally geodesic. Further, it follows that $\gind=2\gM$
and the second fundamental form also satisfies the equation
\[
0 = \A = \A_I \oplus \A_f \,,
\]
where $\A_I$ is the second fundamental form of the map
$I:(M,\gind)\to (M,\gM)$, and $\A_f$ is the second fundamental form
of the map $f:(M,\gind)\to (N,\gN)$. Consequently, $\A_f=0$, so that
$f$ is a totally geodesic isometric immersion. \\
Since all singular values are the same and all vectors are null-eigenvectors
of $\Phi_{\svdf_0^2}$, we can evaluate $0 = (\Delta \Phi_{\svdf_0^2})(v,v)$
for an arbitrary vector $v$. Then
the third line in \eqref{eq:LapPhi} yields $\sec_N=\sigma$ on $\dd f(TM)$, and
the fifth line implies $\Ric_M=\sigma\gM$. In view of the curvature
assumptions, the latter means $\sec_M=\sigma$.
Thus, the claim of the theorem follows. \hfill \qed
\subsection{Proof of Corollary \ref{cor:CorB}}
Using the assumptions on the dimensions and the notation
from section \ref{sec:SVD}, we have $r=r(x)=\rank(\dd f) \le \min\{m,n\}=n$
and $\svdf_j(x) = 0$ for $j=1,\dotsc,m-r$. Consequently,
\begin{align*}
\tr(\sind) &= \underbrace{\sum_{j=1}^{m-r}1}_{\ge m-n} + \underbrace{\sum_{j=m-r+1}^m \frac{1-\svdf_{n-m+j}^2(x)}{1+\svdf_{n-m+j}^2(x)}}_{>-r} > m-n-r \ge m-2n \ge 0 \,,
\end{align*}
so that condition \eqref{eq:trCon} of theorem \ref{thm:ThmA} is fulfilled.
Condition \eqref{eq:sCondition4} is satisfied
by assumption, so that we can draw the conclusions of theorem \ref{thm:ThmA}.
Further, since the above inequality for $\tr(\sind)$ is strict,
the map $f:M\to N$ must be constant and the claim follows. \hfill \qed
%
%
%
\section{Discussion}
\label{sec:Discussion}
We conclude by giving some remarks on the assumptions made in the theorem.
\begin{remark}
\begin{enumerate}
\item Since $M$ is compact, a constant $\kappa^{2}>1$ satisfying
the inequality $f^{*}\gN < \kappa^{2} \gM$ always exists.
\item If $\dim M\ge 3$, the condition on the trace
\eqref{eq:trCon} is strictly weaker than the area-decreasing condition.
In view of Eq.\ \eqref{eq:sCondition4}, this is compensated by an additional requirement on the second fundamental form.
\end{enumerate}
\end{remark}
\begin{remark}
In some situations, we can exclude case \ref{it:thmAii} of the theorem. For example,
if $\dim M > \dim N$, a map $f$ satisfying the assumptions of theorem \ref{thm:ThmA}
cannot be an isometric immersion, so that $f$ necessarily has to be constant.
\end{remark}
To draw the conclusion of theorem \ref{thm:ThmA}, the assumptions
can be weakened in various situations. Below, we will always assume
the curvature conditions of theorem \ref{thm:ThmA} to hold.
\begin{remark}[Low dimensions]
In low dimensions, one can remove the assumption on the second fundamental form.
\begin{enumerate}
\item $\dim N=1$. In this case, any smooth minimal map $f:M\to N$ is constant \cite[Theorem C]{SHS13}.
\item $\dim M=2$. Here, the condition $\tr(\sind)\ge 0$ means that $f$ is a weakly area-decreasing map.
Then \cite[Theorem B]{SHS13} implies that $f$ is either constant, or it is an isometric immersion on a non-empty, closed subset $D$ of $M$ and
strictly area-decreasing on the complement of $D$.
\item $\dim M=\dim N=2$. Let us consider the Jacobian of the projection map $\pi_M : \Gamma(f)\to M$,
which may be expressed as $v\coloneqq\star\Omega_{\gind}$, where $\Omega_{\gind}$ is the
volume form on $\Gamma(f)$ induced by the metric $\gind$ and $\star:\Omega^k(M)\to\Omega^{2-k}(M)$
is the Hodge star with respect to the induced metric $\gind$.
The differential equation satisfied by $v$
is essentially calculated in \cite{Wang02}*{Proposition 3.1} and is given by
\begin{align*}
\Delta \ln v &= -\|\A\|^2 - \svdf_1^2 \Bigl\{ (\A_{11}^1)^2 + (\A_{12}^1)^2 \Bigr\} - \svdf_2^2 \Bigl\{ (\A_{12}^2)^2 + (\A_{22}^2)^2 \Bigr\} \\
& \quad - 2 \svdf_1\svdf_2 \Bigl\{ \A_{11}^2 \A_{21}^1 + \A_{12}^2 \A_{22}^1 \Bigr\} \\
& \quad - \frac{1}{(1+\svdf_1^2)(1+\svdf_2^2)} \Bigl\{ (\svdf_1^2+\svdf_2^2) \sec_M - 2 \svdf_1^2\svdf_2^2 \sec_N \Bigr\} \,.
\end{align*}
From the minimality of $\Gamma(f)$ we also get
$\A_{11}^2=-\A_{22}^2$ and $\A_{22}^1=-\A_{11}^1$, so that we estimate
\begin{align*}
\Delta \ln v &\le -\|\A\|^2 - \svdf_1^2 \Bigl\{ (\A_{11}^1)^2 + (\A_{12}^1)^2 \Bigr\} - \svdf_2^2 \Bigl\{ (\A_{12}^2)^2 + (\A_{22}^2)^2 \Bigr\} \\
& \quad + 2 |\svdf_1\svdf_2| \Bigl\{ \bigl|\A_{22}^2 \A_{21}^1\bigr| + \bigl|\A_{12}^2 \A_{11}^1 \bigr| \Bigr\} \\
& \quad - \frac{1}{(1+\svdf_1^2)(1+\svdf_2^2)} \Bigl\{ (\svdf_1^2+\svdf_2^2) \sec_M - 2 \svdf_1^2\svdf_2^2 \sec_N \Bigr\} \\
&= -\|\A\|^2 - \Bigl( \bigl|\svdf_1\A_{11}^1\bigr| - \bigl|\svdf_2\A_{12}^2 \bigr| \Bigr)^2 - \Bigl( \bigl|\svdf_1\A_{21}^1 \bigr| - \bigl|\svdf_2\A_{22}^2\bigr| \Bigr)^2 \\
&\quad - \frac{1}{(1+\svdf_1^2)(1+\svdf_2^2)}\Bigl\{ (\svdf_1^2+\svdf_2^2)(\sec_M-\sigma) + 2 \svdf_1^2\svdf_2^2 (\sigma-\sec_N) \\
& \qquad \qquad \qquad \qquad \qquad + \sigma(\svdf_1^2+\svdf_2^2-2\svdf_1^2\svdf_2^2) \Bigr\} \,.
\end{align*}
The assumption $\tr(\sind)\ge 0$ is equivalent to $\svdf_1^2\svdf_2^2\le 1$,
which in turn implies the estimate
\[
\svdf_1^2+\svdf_2^2-2\svdf_1^2\svdf_2^2 \ge \svdf_1^2+\svdf_2^2 - 2 |\svdf_1\svdf_2| = \bigl( |\svdf_1|-|\svdf_2| \bigr)^2 \ge 0
\]
for the third curvature term.
Therefore, all terms in the above differential equation are non-negative.
By the maximum principle, we conclude that $\ln v$ must be constant on $M$.
Evaluating $0=\Delta\ln v$, we see that $\A=0$ everywhere.
The estimate for the curvature term yields $\svdf_1^2=\svdf_2^2$,
and then $\svdf_1^2=\svdf_2^2=0$ or $\svdf_1^2=\svdf_2^2=1$ everywhere.
In the latter case, the other curvature terms imply $\sec_M=\sigma$ and
$\sec_N=\sigma$ on $\dd f(TM)$.
\end{enumerate}
\end{remark}
\begin{remark}[Weakly length-decreasing maps]
\label{rem:WeaklyLengthDecreasing}
By examining lemma \ref{lem:StrictLengthDecr} and the proof
of theorem \ref{thm:ThmA},
we note that if $\svdf_0^2\le 1$
one does not need to impose a condition on
the second fundamental form.
Thus, in the case at hand, we obtain an alternative proof of \cite[Theorem A]{SHS13}.
\end{remark}
Let us remark that there exists an abundance
of weakly length-decreasing minimal maps
which are not totally geodesic. In particular,
these maps satisfy condition \eqref{eq:trCon}.
\begin{example}[Holomorphic Maps, {\cite[Example 2(a)]{SHS13}}]
For complex manifolds $M$ and $N$, the graph of a holomorphic map $f:M\to N$ is
automatically minimal. By a result due to Ahlfors \cite{Ahlfors38} and its extension by Yau \cite{Yau78},
for every holomorphic map $f:M\to N$ between a complete K\"{a}hler manifold
$M$ with Ricci curvature bounded from below by a negative constant $-a$ and $N$
a Hermitian manifold with holomorphic bisectional curvature bounded
from above by a negative constant $-b$, it is $f^*\gN\le\frac{a}{b}\gM$.
\end{example}
\begin{bibdiv}
\begin{biblist}
\bib{Ahlfors38}{article}{
author={Ahlfors, Lars V.},
title={An extension of Schwarz's lemma},
journal={Trans. Amer. Math. Soc.},
volume={43},
date={1938},
number={3},
pages={359--364},
issn={0002-9947},
% review={\MR{1501949}},
}
\bib{EH90}{article}{
author={Ecker, Klaus},
author={Huisken, Gerhard},
title={A Bernstein result for minimal graphs of controlled growth},
journal={J. Differential Geom.},
volume={31},
date={1990},
number={2},
pages={397--400},
issn={0022-040X},
% review={\MR{1037408}},
}
\bib{FC80}{article}{
author={Fischer-Colbrie, D.},
title={Some rigidity theorems for minimal submanifolds of the sphere},
journal={Acta Math.},
volume={145},
date={1980},
number={1-2},
pages={29--46},
issn={0001-5962},
% review={\MR{558091}},
}
\bib{Ham82}{article}{
author={Hamilton, Richard S.},
title={Three-manifolds with positive Ricci curvature},
journal={J. Differential Geom.},
volume={17},
date={1982},
number={2},
pages={255--306},
issn={0022-040X},
% review={\MR{664497}},
}
\bib{Ham86}{article}{
author={Hamilton, Richard S.},
title={Four-manifolds with positive curvature operator},
journal={J. Differential Geom.},
volume={24},
date={1986},
number={2},
pages={153--179},
}
\bib{HJW80}{article}{
author={Hildebrandt, S.},
author={Jost, J.},
author={Widman, K.-O.},
title={Harmonic mappings and minimal submanifolds},
journal={Invent. Math.},
volume={62},
date={1980/81},
number={2},
pages={269--298},
issn={0020-9910},
% review={\MR{595589}},
}
\bib{JX99}{article}{
author={Jost, J.},
author={Xin, Y. L.},
title={Bernstein type theorems for higher codimension},
journal={Calc. Var. Partial Differential Equations},
volume={9},
date={1999},
number={4},
pages={277--296},
issn={0944-2669},
% review={\MR{1731468}},
}
\bib{JXY12}{article}{
author={Jost, J\"urgen},
author={Xin, Yuanlong},
author={Yang, Ling},
title={The regularity of harmonic maps into spheres and applications to
Bernstein problems},
journal={J. Differential Geom.},
volume={90},
date={2012},
number={1},
pages={131--176},
issn={0022-040X},
% review={\MR{2891479}},
}
\bib{JXY13}{article}{
author={Jost, J.},
author={Xin, Y. L.},
author={Yang, Ling},
title={The Gauss image of entire graphs of higher codimension and
Bernstein type theorems},
journal={Calc. Var. Partial Differential Equations},
volume={47},
date={2013},
number={3-4},
pages={711--737},
issn={0944-2669},
% review={\MR{3070561}},
}
\bib{JXY16}{article}{
author={Jost, J\"urgen},
author={Xin, Yuan Long},
author={Yang, Ling},
title={The geometry of Grassmannian manifolds and Bernstein-type theorems
for higher codimension},
journal={Ann. Sc. Norm. Super. Pisa Cl. Sci. (5)},
volume={16},
date={2016},
number={1},
pages={1--39},
issn={0391-173X},
% review={\MR{3524663}},
}
\bib{LS10}{article}{
author={Li, Guanghan},
author={Salavessa, Isabel M. C.},
title={Bernstein-Heinz-Chern results in calibrated manifolds},
journal={Rev. Mat. Iberoam.},
volume={26},
date={2010},
number={2},
pages={651--692},
issn={0213-2230},
% review={\MR{2677011}},
}
\bib{SHS13}{article}{
author = {Savas-Halilaj, Andreas},
author = {Smoczyk, Knut},
title = {Bernstein theorems for length and area decreasing minimal maps},
journal = {Calc. Var. Partial Differential Equations},
year = {2014},
volume = {50},
pages = {549--577},
number = {3-4},
doi = {\doi{10.1007/s00526-013-0646-0}},
% fjournal = {Calculus of Variations and Partial Differential Equations},
issn = {0944-2669},
% mrclass = {53C40 (53A07 58J05)},
% mrnumber = {3216824},
% mrreviewer = {Mircea Crasmareanu},
% url = {http://dx.doi.org/10.1007/s00526-013-0646-0}
}
\bib{Sch93}{article}{
author={Schoen, Richard M.},
title={The role of harmonic mappings in rigidity and deformation
problems},
conference={
title={Complex geometry},
address={Osaka},
date={1990},
},
book={
series={Lecture Notes in Pure and Appl. Math.},
volume={143},
publisher={Dekker, New York},
},
date={1993},
pages={179--200},
% review={\MR{1201611}},
}
\bib{SWX06}{article}{
author={Smoczyk, Knut},
author={Wang, Guofang},
author={Xin, Y. L.},
title={Bernstein type theorems with flat normal bundle},
journal={Calc. Var. Partial Differential Equations},
volume={26},
date={2006},
number={1},
pages={57--67},
issn={0944-2669},
% review={\MR{2217483}},
}
\bib{Wang02}{article}{
author={Wang, Mu-Tao},
title={Long-time existence and convergence of graphic mean curvature flow
in arbitrary codimension},
journal={Invent. Math.},
volume={148},
date={2002},
number={3},
pages={525--543},
issn={0020-9910},
% review={\MR{1908059}},
}
\bib{Wang03}{article}{
author={Wang, Mu-Tao},
title={On graphic Bernstein type results in higher codimension},
journal={Trans. Amer. Math. Soc.},
volume={355},
date={2003},
number={1},
pages={265--271},
issn={0002-9947},
% review={\MR{1928088}},
}
\bib{Yau78}{article}{
author={Yau, Shing Tung},
title={A general Schwarz lemma for K\"ahler manifolds},
journal={Amer. J. Math.},
volume={100},
date={1978},
number={1},
pages={197--203},
issn={0002-9327},
% review={\MR{0486659}},
}
\end{biblist}
\end{bibdiv}
\end{document}