On the coupling of regularization techniques and the boundary element method for a hemivariational inequality modelling a delamination problem
Extracted source text · arXiv 1510.06343
%%%%%%%%%%%%%%%%%%%%%%% file template.tex %%%%%%%%%%%%%%%%%%%%%%%%%
%
% This is a general template file for the LaTeX package SVJour3
% for Springer journals. Springer Heidelberg 2010/09/16
%
% Copy it to a new file with a new name and use it as the basis
% for your article. Delete % signs as needed.
%
% This template includes a few options for different layouts and
% content for various journals. Please consult a previous issue of
% your journal as needed.
%
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%
% First comes an example EPS file -- just ignore it and
% proceed on the \documentclass line
% your LaTeX will extract the file if required
\begin{filecontents*}{example.eps}
%!PS-Adobe-3.0 EPSF-3.0
%%BoundingBox: 19 19 221 221
%%CreationDate: Mon Sep 29 1997
%%Creator: programmed by hand (JK)
%%EndComments
gsave
newpath
20 20 moveto
20 220 lineto
220 220 lineto
220 20 lineto
closepath
2 setlinewidth
gsave
.4 setgray fill
grestore
stroke
grestore
\end{filecontents*}
%
\RequirePackage{fix-cm}
%
%\documentclass{svjour3} % onecolumn (standard format)
%\documentclass[smallcondensed]{svjour3} % onecolumn (ditto)
\documentclass[smallextended]{svjour3} % onecolumn (second format)
%\documentclass[twocolumn]{svjour3} % twocolumn
%
\smartqed % flush right qed marks, e.g. at end of proof
%
\usepackage{graphicx}
\usepackage{amsmath,amssymb}
\usepackage{subfigure}
\usepackage[percent]{overpic}
\usepackage{color}
%
% \usepackage{mathptmx} % use Times fonts if available on your TeX system
%
% insert here the call for the packages your document requires
%\usepackage{latexsym}
% etc.
%
% please place your own definitions here and don't use \def but
% \newcommand{}{}
\newcommand{\R}{\mathbb{R}}
\newcommand{\ds}{\displaystyle}
\spnewtheorem{assumption}{Assumption}{\bf}{\it}
\spnewtheorem{rem}{Remark}{\bf}{\it}
\spnewtheorem{alg}{Algorithm}{\bf}{\rm}
%
% Insert the name of "your journal" with
% \journalname{myjournal}
%
\begin{document}
\title{Coupling regularization and adaptive $hp$-BEM for the solution of %hemivariational inequality modeling
a delamination problem%\thanks{Grants or other notes
%about the article that should go on the front page should be
%placed here. General acknowledgments should be placed at the end of the article.}
}
%\subtitle{Do you have a subtitle?\\ If so, write it here}
\titlerunning{$hp$-adaptive BEM for regularized hemivariational inequalities} % if too long for running head
\author{Nina~Ovcharova \and
Lothar~Banz %etc.
}
%\authorrunning{Short form of author list} % if too long for running head
\institute{N. Ovcharova \at
Universit\"at der Bundeswehr M\"unchen, D-85577 Neubiberg/Munich, Germany \\
\email{nina.ovcharova@unibw.de} % \\
% \emph{Present address:} of F. Author % if needed
\and
L. Banz \at
Department of Mathematics, University of Salzburg, Hellbrunner Stra{\ss}e 34, 5020 Salzburg, Austria\\
\email{lothar.banz@sbg.ac.at}
}
\date{Received: date / Accepted: date}
% The correct dates will be entered by the editor
\maketitle
\begin{abstract}
In this paper, we couple regularization techniques %the smoothing of the data %regularization techniques of nondifferentiable optimization
with the adaptive $hp$-version of the boundary element method ($hp$-BEM) for the efficient numerical solution of linear elastic problems with nonmonotone contact boundary conditions. As a model example we treat the delamination of composite structures with a contaminated interface layer. This problem has a weak formulation in terms of a nonsmooth variational inequality. The resulting hemivariational inequality (HVI)
%The variational formulation of this problem leads to hemivariational inequality with a nonsmooth functional defined on the contact boundary. This problem
is first regularized and then, discretized by an adaptive $hp$-BEM. We give conditions for the uniqueness of the solution and provide an a-priori error estimate. Furthermore, we derive an a-posteriori error estimate
for the nonsmooth variational problem based on a novel regularized mixed formulation, thus enabling $hp$-adaptivity.
%prove convergence of the BEM Galerkin solution of the regularized problem in the energy norm, and obtain an a-priori error estimate for the regularized problem based on a novel C\'{e}a-Falk approximation lemma. Furthermore, we derive an a-posteriori error estimate based on an equivalent regularized mixed formulation thus enabling $hp$-adaptivity.
Various numerical experiments illustrate the behavior, strengths and weaknesses of the proposed high-order approximation scheme.
\keywords{ Hemivariational inequality \and Regularization techniques \and Delamination problem \and a-priori and a-posteriori error estimates \and $hp$-BEM}
% \PACS{PACS code1 \and PACS code2 \and more}
\subclass{65N38 \and 74M15 }
\end{abstract}
\section{Introduction}
The interest in robust and efficient numerical methods for the solution of nonsmooth problems in nonmonoton contact like adhesion and delamination problems increases constantly. The nonsmoothness comes from the nonsmooth data of the problems itself, in particular from nonmonotone multivalued physical laws involved in the boundary conditions that lead to nonsmooth functionals in the variational formulation.
%Efficient numerical methods for simulation of mechanical problems with nonsmooth nonmonotone contact like the adhesive contact in composite structures is of ever increasing importance in the last years. We are motivated by the benchmark delamination problems in material sciences that come from the double cantilever beam (DCB) test problem \cite{Gudladt}. The result of a typical experiment is shown in Figure \ref{01}(a) from \cite{Gudladt}, where three probes with different contamination of the interface layer have been exposed. The mathematical formulation of these mechanical phenomena leads to boundary value problems with operator subdifferential inclusions in the boundary conditions. In particular, they involve nonmonotone, possibly multivalued laws in the boundary conditions, which can be expressed by means of the Clarke subdifferential of a nonconvex, nonsmooth locally Lipschitz function. As a result, a nonsmooth functional defined on the contact part appears in the variational formulation of these problems.
There are several approaches to treat this non-differentiability. We can first
discretize by finite elements, and then solve the nonconvex optimization problems by novel nonsmooth optimization methods \cite{Noll_Ovcharova}, or first regularize the nonsmooth functional, then discretize by finite element methods and finally use standard optimization solvers \cite{Ovcharova2012}. Since the interesting
nonmonotone behavior takes place only in a boundary part of the involved linear elastic bodies, the domain hemivariational inequaliy can also be formulated as
%is transformed to
a boundary hemivariational inequality %by means of the symmetric
via the Poincar\'{e}-Steklov operator.
Therefore, the boundary element methods (BEM) are the methods of choice for discretization. There are number of works exploiting the $h$- and the high-order $hp$-version of the BEM for monotone contact problems.
%Another option to treat these problems numerically is to use the boundary element method. To this end, the contact problem with a nonmonotone law has to be recast into a boundary integral formulation by making use of the Poincar\'{e}-Steklov operator. There exist numerous works that employ high-order $hp$-boundary elements for monotone contact problems.
Advanced $hp$-BEM discretizations, a-priori and a-posteriori error estimates for unilateral Signorini problems or contact problems with monotone friction are established in \cite{banz2013posteriori,Banz2015,chernov2008hp,MS-1,maischak2007adaptive,stephan2009hp}. However, successfully applied $hp$-BEM techniques for nonsmooth nonconvex variational inequalities are still missing. Pure $h$-BEM with lowest polynomial degree has been investigated for the first time by Nesemann and Stephan \cite{Nesemann} for unilateral contact with adhesion. % but using the simplest piecewise linear approximation.
They study the existence and uniqueness, and introduce a residual error indicator.
In this paper, we extend the approximation scheme from \cite{Ovcharova2015} (dealing with the $h$ - version of the BEM) to the $hp$-version of the BEM for the regularized problem. % solve nonsmooth, nonmonotone variational inequalities by combining regularization with the $hp$-adaptive BEM.
More precisely, %To this end, we use the equivalent formulation of the adhesive contact problem via the Poincar\'{e}-Steklov operator.
% which is first regularized and then discretized by a $hp$-BEM.
the nonsmooth variational inequality is first regularized and then discretized by $hp$-adaptive BEM. % high-order boundary elements.
%after an $\varepsilon$-regularization of the nonsmooth functional, we discretize the Poincar\'{e}-Steklov operator by boundary elements and obtain a finite dimensional variational problem defined only on the boundary.
For the Galerkin solution of the regularized problem we provide an a-priori error estimate %using a new C\'{e}a-Falk approximation lemma for the regularized problem. We also
and derive a reliable a-posteriori error estimate based on a {\it novel} regularized mixed formulation and thus, enabling $hp$-adaptivity. All our theoretical results are illustrated with various numerical experiments for a contact problem with adhesion.
%namely, we combine the regularization techniques with the boundary element method. More precisely, after $\varepsilon$-regularization of the nonsmooth functional, we discretize the Poincar\'{e}-Steklov operator by boundary elements and obtain a finite dimensional variational problem defined only on the boundary that can be solved numerically e.g. by a Newton method.
We also provide conditions for the uniqueness of the solution that sharpen the results due to \cite{Nesemann}. %establish $\varepsilon h$-norm convergence of the discrete solution in the energy norm.\\
\textbf{Notation:} We denote with $C_i$, $c_i$ and such alike generic constants, which can take different values at different positions. We use bold font to indicate vector-valued variables, e.g.~$\mathbf {u} \in \mathbf{H}^1(\Omega):=[H^1(\Omega)]^d$. If there are too many indices, one of them
is written as a superscript, % might be shifted to the exponent,
e.g.~$\mathbf{u}_{\varepsilon, hp}=\mathbf{u}^{\varepsilon}_{hp}$.
\section{A nonmonotone boundary value problem from delamination}
Let $\Omega \subset \R^d \, (d= 2, 3)$ be a bounded domain with Lipschitz boundary $\Gamma=\partial \Omega$. We assume that the boundary is decomposed into three disjoint open parts $\Gamma_D, \Gamma_N$,
and $\Gamma_C$ such that $\Gamma = \overline{\Gamma}_D \cup
\overline{\Gamma}_N \cup \overline{\Gamma}_C$ and, moreover, the measures of $\Gamma_C$ and $\Gamma_D$ are positive. We fix an elastic body occupying $\overline{\Omega}$. The body is subject to volume force ${\mathbf f} \in [L^2(\Omega)]^d$ and $g\in H^{1/2}(\Gamma_C)$, $g\geq 0$, is a gap function associating every point $x\in \Gamma_C$ with its distance to the rigid obstacle measured in the direction of the unit outer normal vector $\mathbf{n}(x)$. The body is fixed along $\Gamma_D$, surface tractions
%Zero displacements are prescribed on $\Gamma_D$, surface tractions
$\mathbf {t} \in [L^2(\Gamma_N)]^d$ act on $\Gamma_N$, and on the part $\Gamma_C$ a nonmonotone, generally multivalued boundary condition holds. %Moreover, the body is subject to volume force ${\mathbf f} \in [L^2(\Omega)]^2$ and $g\in H^{1/2}(\Gamma_C)$, $g\geq 0$, is a gap function associating every point $x\in \Gamma_C$ with its distance to the rigid obstacle measured in the direction of the unit outer normal vector $\mathbf{n}(x)$.
Further, $\varepsilon (\mathbf {u}) = \frac{1}{2}(\nabla {\mathbf u} + \nabla {\mathbf u}^T)$ denotes the linearized strain tensor and $\sigma ( \mathbf {u}) = \mathcal{C} \,: \, \varepsilon (\mathbf {u})$ stands for the stress tensor, where $\mathcal{C}$ is the Hooke tensor, assumed to be
uniformly positive definite with $L^\infty$ coefficients. The boundary stress vector can be decomposed further into the normal, respectively, the tangential stress:
\[
\sigma_n= \sigma(\mathbf{u}) {\mathbf n} \cdot {\mathbf n}, \quad \sigma_t=\sigma(\mathbf{u}) {\mathbf n} - \sigma_n {\mathbf n}.
\]
%where ${\mathbf n}$ denotes the unit outer normal vector on $\partial \Omega$.
%Our benchmark problem is a two- or three-dimensional symmetric laminated structure with an interlayer adhesive under loading
%(see Figure \ref{delpr} for the 2D benchmark problem).
By assuming that the structure is symmetric and the forces applied to the upper and lower part are the same, it suffices to consider only the upper half of the specimen represented by $\overline{\Omega}$, see Fig. \ref{delpr} left for the 2D benchmark problem.
\begin{figure}[tbp]
\centering \mbox{
\subfigure[Load-displacement curve for different contamination concentrations, see \cite{Gudladt}]{
\includegraphics[trim = 0cm 9cm 0cm 0cm, clip,width=0.5\textwidth]{load-displ-curve}} \
\subfigure[A nonmonotone adhesion law]{
\includegraphics[trim = 50mm 150mm 0mm 0mm, clip,width=65.0mm, keepaspectratio]{adhesion}} }
\caption{Nonmonotone adhesion law} \label{01}
\end{figure}
\begin{figure}[b]
\centering
\includegraphics[trim = 0cm 0cm 0cm 0cm, clip,width=0.9\textwidth]{2Dbenchmark}
\caption{Reference configuration for the 2D benchmark under loading. Under applied traction force $\mathbf{t}$ the crack front propagates to the left} \label{delpr}
\end{figure}
The delamination problem under consideration is the following: Find ${\mathbf u} \in \mathbf{H}^1(\Omega):=[H^1(\Omega)]^d$ such that
\begin{subequations}
\begin{alignat}{2}
-\mbox{div} \; \sigma ({\mathbf u})&= {\mathbf f} & \quad & \mbox{in} \; \Omega \label{eq1} \\
{\mathbf u}&=0 & \quad & \mbox{on} \; \Gamma_D \\
\sigma({\mathbf u}){\mathbf n} &= {\mathbf t} & \quad & \mbox{on}\; \Gamma_N \\
u_n &\leq g & \quad & \mbox{on}\; \Gamma_C \label{unilateral} \\
\sigma_t({\mathbf u})&=0 & \quad & \mbox{on} \; \Gamma_C\label{eq:notangstress} \\
-\sigma_n ({\mathbf u}) &\in \partial f(u_n) & \quad & \mbox{on} \; \Gamma_C . \label{nbc}
\end{alignat}
\end{subequations}
The contact law (\ref{nbc}), written as a differential inclusion by means of the Clarke subdifferential $\partial f$ \cite{Clarke} of a locally Lipschitz function $f$, describes the nonmonotone, multivalued behavior of the adhesive. More precisely, $\partial f$ is the physical law between the normal component $\sigma_n$ of the stress boundary vector and the normal component $u_n= \mathbf{u} \cdot \mathbf{n}$ of the displacement $\mathbf{u}$ on $\Gamma_C$. A typical zig-zagged nonmonotone adhesion law is shown in Figure \ref{01}(b).\\
To give a variational formulation of the above boundary value problem we define
\begin{align*}
\mathbf{H}^1_D(\Omega)=\left\{\mathbf{v} \in \mathbf{H}^1(\Omega) \, : \, \mathbf{v}|_{\Gamma_D}=0 \right\}, \quad
\mathcal{K}=\{ \mathbf{v} \in \mathbf{H}^1_D(\Omega) \, :\, v_n \leq g \text{ a.e.~on } \Gamma_C\}
\end{align*}
and introduce the $\mathbf{H}^1_D(\Omega)$-coercive and continuous bilinear form of linear elasticity
\[
a(\mathbf{u},\mathbf{v})= \int_{\Omega} \sigma (\mathbf{u}) : \varepsilon (\mathbf{v}) \, dx.
\]
Multiplying the equilibrium equation (\ref{eq1}) by $\mathbf{v-u}$, integrating over $\Omega$ and applying the divergence theorem yields
%\small
\[
\int_{\Omega} \sigma (\mathbf{u}) : \varepsilon (\mathbf{v}-\mathbf{u}) \, dx = \int_{\Omega} \mathbf{f} \cdot (\mathbf{v}-\mathbf{u})\, dx + \int_{\Gamma} \sigma (\mathbf{u}) \mathbf{n} \cdot (\mathbf{v}-\mathbf{u})\, ds.
\]
From the definition of the Clarke subdifferential, the nonmonotone boundary condition (\ref{nbc}) is equivalent to
\[
-\sigma_n (\mathbf{u})(v_n-u_n)\leq f^0(u_n;v_n-u_n).
\]
Here, the notation $f^0(x;z)$ stands for the generalized directional derivative of $f$ at $x$ in direction $z$.
Substituting $\sigma(\mathbf{u})\mathbf{n}$ by $\mathbf{t}$ on $\Gamma_N$, using on $\Gamma_C$ the decomposition
\[
\sigma (\mathbf{u})\mathbf{n} \cdot (\mathbf{v}-\mathbf{u}) = \sigma_t (\mathbf{u}) \cdot (\mathbf{v_t}-\mathbf{u_t}) +\sigma_n (\mathbf{u})(v_n-u_n)
\]
and taking into account that on $\Gamma_C$ no tangential stresses are assumed, c.f.~\eqref{eq:notangstress},
we obtain the hemivariational inequality (HVI): Find $\mathbf {u} \in \mathcal{K}$ such that
\begin{align}
a(\mathbf{u},\mathbf{v}-\mathbf{u}) + \int_{\Gamma_C} f^0(u_n(s);v_n(s)-u_n(s)) \, ds \geq \int_{\Omega} \mathbf{f} \cdot (\mathbf{v}-\mathbf{u})\, dx \nonumber \\
+ \int_{\Gamma_N} \mathbf{t} \cdot (\mathbf{v}-\mathbf{u})\, ds \quad \forall \, \mathbf{v} \in \mathcal{K}. \label{HemVar}
\end{align}
\section{Boundary integral operator formulation}
Since the main difficulties of the boundary value problem, namely the nonmonotone adhesion law (\ref{nbc}) and the unilateral contact condition (\ref{unilateral}) appear on the boundary,
%$f^0(u_n(s);v_n(s)-u_n(s))$ and $v_n \leq g$, lie on the boundary,
it might be viewed advantageous to formulate \eqref{HemVar} as a boundary integral problem.
To this end, we introduce the free boundary part $\Gamma_0=\Gamma \backslash \overline{\Gamma_D}=\Gamma_N \cup \Gamma_C $ and recall the Sobolev spaces \cite{HsWendl}:
\[
\begin{array}{lll}
H^{1/2}(\Gamma)&=&\{v\in L^2(\Gamma)\, :\, \exists\, v' \in H^1(\Omega), \, \mbox{tr}\, v'=v\},\\[0.2cm]
H^{1/2}(\Gamma_0)&=&\{v=v'|_{\Gamma_0} \, :\, \exists\, v'\in H^{1/2}(\Gamma) \},\\[0.2cm]
\tilde{H}^{1/2}(\Gamma_0)&=&\{v=v'|_{\Gamma_0} \,:\, \exists\, v'\in H^{1/2}(\Gamma),\; \mbox{supp} \, v' \subset {\Gamma}_0 \} %,\\[0.2cm]
%H^{-1/2}(\Gamma_0)&=& (\tilde{H}^{1/2}(\Gamma_0))', \quad
%\tilde{H}^{-1/2}(\Gamma_0) \; = \; ({H}^{\frac{1}{2}}(\Gamma_0))'.
\end{array}
\]
with the standard norms
\[\|u\|_{H^{1/2}(\Gamma_0)}= \displaystyle\inf_{ v\in H^{1/2}(\Gamma), v|_{\Gamma_0}=u}\|v\|_{H^{1/2}(\Gamma)} \quad \text{and} \quad
\|u\|_{\tilde{H}^{1/2}(\Gamma_0)}=\|u_0\|_{H^{1/2}(\Gamma)},
\]
where $u_0$ is the extension of $u$ onto $\Gamma$ by zero. The Sobolev space of negative order on $\Gamma_0$ are defined by duality as
\[
H^{-1/2}(\Gamma_0)=(\tilde{H}^{1/2}(\Gamma_0))^* \quad \text{and} \quad
\tilde{H}^{-1/2}(\Gamma_0) \; = \; ({H}^{1/2}(\Gamma_0))^*.
\]
Moreover, from \cite[Lemma 4.3.1]{HsWendl} we have the inclusions
\[
\tilde{H}^{1/2}(\Gamma_0)\subset H^{1/2}(\Gamma_0)\subset L^2(\Gamma_0)\subset \tilde{H}^{-1/2}(\Gamma_0)\subset H^{-1/2}(\Gamma_0).
\]
%For the spaces of vector-valued functions we use the bold symbols, e.g.~$\mathbf{H}^{1/2}(\Gamma)=[H^{1/2}(\Gamma)]^d$.\\
For the solution $\mathbf{u}(\mathbf{x})$ of \eqref{eq1} with $\mathbf{x} \in \Omega\backslash \Gamma$
we have the following representation formula, also known as Somigliana's identity, see e.g.~\cite{Kleiber}:
{\small \begin{align} \label{eq2}
\mathbf{u}(\mathbf{x})=\int_{\Gamma} G(\mathbf{x},\mathbf{y}) \left(\mathbf{T}_y \mathbf{u}(\mathbf{y})\right)\, ds_y -\int_\Gamma \mathbf{T}_y G(\mathbf{x},\mathbf{y})\mathbf{u}(\mathbf{y}) \,ds_y + \int_\Omega G(\mathbf{x},\mathbf{y})\mathbf{f}(\mathbf{y}) \, d\mathbf{y},
\end{align}
}
where $G(\mathbf{x},\mathbf{y})$ is the fundamental solution of the Navier-Lam\'{e} equation defined by
\begin{align*}
G(\mathbf{x},\mathbf{y})=\begin{cases}\frac{\lambda +3\mu}{4\pi \mu(\lambda+2\mu)}\left(\log |\mathbf{x}-\mathbf{y}|{\rm I} + \frac{\lambda + \mu}{\lambda +3\mu}\frac{(\mathbf{x}-\mathbf{y})(\mathbf{x}-\mathbf{y})^\top}{|\mathbf{x}-\mathbf{y}|^2}\right), \quad &\text{ if d=2}\\[0.2cm]
\frac{\lambda +3\mu}{8\pi \mu(\lambda+2\mu)}\left( |\mathbf{x}-\mathbf{y}|^{-1}{\rm I} + \frac{\lambda + \mu}{\lambda +3\mu}\frac{(\mathbf{x}-\mathbf{y})(\mathbf{x}-\mathbf{y})^\top}{|\mathbf{x}-\mathbf{y}|^3}\right), \quad &\text{ if d=3}
\end{cases}
\end{align*}
% \[
% G(\mathbf{x}, \mathbf{y})=\frac{1}{8 \pi \mu (\lambda+ 2 \mu)}\left (
% \frac{(\lambda+3\mu)\delta_{ij}}{\|\mathbf{x}-\mathbf{y}\|}+(\lambda+\mu)\frac{(x_i-y_i)(x_j-y_j)}{\|\mathbf{x}-\mathbf{y}\|^3}
% \right )
% \]
with the Lam\'{e} constants $\lambda, \mu >0$ depending on the material parameters, i.e.~the modulus of elasticity~$E$ and the Poisson's ratio~$\nu$:
$$
\lambda = \frac{E\nu}{1-\nu^2} \,, \qquad \mu= \frac{E}{1+\nu}\, .
$$
Here, $\mathbf{T}_y$ stands for the traction operator with respect to $\mathbf{y}$ defined by $\mathbf{T}_y (\mathbf{u}) := \sigma(\mathbf{u}(\mathbf{y}))\cdot \mathbf{n}_y$, where $\mathbf{n}_y$ is the unit outer normal vector at $\mathbf{y}\in \Gamma$.
Letting $\Omega\backslash \Gamma \ni \mathbf{x} \to \Gamma$ in (\ref{eq2}), we obtain the well-known %system of boundary integral equations %
Calder\'{o}n operator
\[
\left (\begin{array}{c}
\mathbf{u}\\[0.2cm]
\mathbf{T}_x \mathbf{u}
\end{array}
\right )
= \left ( \begin{array} {cc}
\frac{1}{2}I-K & V\\[0.2cm]
W & \frac{1}{2}I+K'
\end{array}
\right)
\left( \begin{array}{c}
\mathbf{u}\\[0.2cm]
\mathbf{T}_x \mathbf{u}
\end{array}
\right ) + \left( \begin{array}{c} N_0 \mathbf{f} \\[0.2cm]
N_1 \mathbf{f}
\end{array}
\right),
\]
with the single layer potential $V$, the double layer potential
$K$, its formal adjoint $K'$, and the hypersingular integral operator $W$ defined for $\mathbf{x} \in \Gamma$ as follows:
\begin{alignat*}{2}
\left( V \phi\right ) (\mathbf{x}) & := \int\limits_\Gamma G(\mathbf{x},\mathbf{y}) \phi (\mathbf{y}) \, ds_y, & \qquad
\left(K \phi \right ) (\mathbf{x}) &:=
\int_{\Gamma} \mathbf{T}_y G^\top(\mathbf{x},\mathbf{y}) \phi (\mathbf{y}) \, ds_y \\
\left( K' \phi \right) (\mathbf{x}) & := \mathbf{T}_x \int\limits_\Gamma G
(\mathbf{x},\mathbf{y}) \phi (\mathbf{y}) \, ds_y, & \qquad
\left(W \phi \right ) (\mathbf{x}) &:= - \mathbf{T}_x \left (K \phi\right ) (\mathbf{x}).
\end{alignat*}
The Newton potentials $N_0, N_1$ are given for $x\in \Gamma$ by
\[
N_0 \mathbf{f} = \int\limits_\Gamma G
(\mathbf{x},\mathbf{y}) \mathbf{f} (\mathbf{y}) \, ds_y, \quad
N_1 \mathbf{f} = \mathbf{T}_x \int\limits_\Gamma G
(\mathbf{x},\mathbf{y}) \mathbf{f} (\mathbf{y}) \, ds_y.
\]
From \cite{Co-1} it is known that the linear operators
\begin{alignat*}{2}
V: &\;\mathbf{H} ^{-1/2 + \sigma} ( \Gamma) \rightarrow \mathbf{H}^{1/2 + \sigma} ( \Gamma)
, &\qquad K:&\; \mathbf{H}^{1/2 + \sigma} ( \Gamma ) \rightarrow \mathbf{H}^{1/2 + \sigma} ( \Gamma)
\\
K' : &\; \mathbf{H}^{-1/2 + \sigma} ( \Gamma ) \rightarrow \mathbf{H}^{-1/2 + \sigma} (
\Gamma) , &\qquad
W:&\; \mathbf{H}^{1/2 + \sigma} ( \Gamma ) \rightarrow \mathbf{H}^{-1/2 + \sigma} (\Gamma)
\end{alignat*}
are well-defined and continuous for $| \sigma | \leq \ds\frac{1}{2}$.
Moreover, $V$ is symmetric and positive definite (elliptic on $\mathbf{H}^{-1/2}(\Gamma)$) in $\mathbb{R}^3$ and, if the capacity of $\Gamma$ is smaller than 1, also in $\mathbb{R}^2$. That can always be achieved by scaling, since the capacity (or conformal radius or transfinite diameter) of $\Gamma$ is smaller than 1, if $\Omega$ is
contained in a disc with radius $<1$
(see e.g. \cite{SlSp,Steinbach}). $W$ is symmetric and positive semidefinite with kernel $\R$ (elliptic on $\tilde{\mathbf{H}}^{1/2}(\Gamma_0)$).
Hence, since $V : \mathbf{H}^{-1/2}(\Gamma) \to \mathbf{H}^{1/2}(\Gamma)$ is invertible, we obtain by taking the Schur complement of the Calderon projector that
\begin{equation}\label{mainformula}
\mathbf{T}_x \mathbf{u}= P\mathbf{u} - N \mathbf{f}
\end{equation}
with the symmetric Poincar\'e-Steklov operator $P : \mathbf{H}^{1/2}(\Gamma) \to \mathbf{H}^{-1/2}(\Gamma)$ and its Newton potential $N$ given by
\begin{align*}
P\mathbf{u} = W \mathbf{u} + \left( K' + \ds\frac{1}{2} I \right) V^{-1} \left( K + \ds\frac{1}{2} I \right) \mathbf{u} , \
N \mathbf{f} = \left( K' + \ds\frac{1}{2} I \right) V^{-1} N_0 \mathbf{f} -N_1 \mathbf{f}.
\end{align*}
Note that if $\mathbf{f}=0$, $P$ maps $\mathbf{u}$ to its traction and, therefore, the Poincar\'e-Steklov operator $P$ is sometimes called the Dirichlet-to-Neumann mapping.
Moreover, $P$ induces a symmetric bilinear form on $\mathbf{H}^{1/2} ( \Gamma )$, and is continuous and $\tilde{\mathbf{H}}^{1/2}(\Gamma_0)$-elliptic, see e.g.~\cite{cf}, i.e.~there exist constants $c_P$, $C_P>0$ such that
\[
\begin{array}{rllll}
\|P \mathbf{v}\| _{\mathbf{H}^{-1/2}(\Gamma)}&\leq& C_P \|\mathbf{v}\| _{\mathbf{H}^{1/2}(\Gamma)} &\quad &\forall \, \mathbf{v} \in {\mathbf{H}^{1/2}(\Gamma)}, \\[0.2cm]
\langle P\mathbf{v}, \mathbf{v} \rangle_{\Gamma_0} &\geq & c_P\|\mathbf{v}\|_{\mathbf{\tilde{H}}^{1/2}(\Gamma_0)} &\quad & \forall \, \mathbf{v} \in {\mathbf{\tilde{H}}^{1/2}(\Gamma_0)}.
\end{array}
\]
Here, $\langle \cdot, \cdot \rangle_{\Gamma_0}$ is the $L^2(\Gamma_0)$-duality pairing between the involved spaces.
Using the boundary functional sets
$$
\mathcal{V}=\tilde{\mathbf{H}}^{1/2} (\Gamma_0) \quad \text{and} \quad \mathcal{K}^{\Gamma}=\{ \mathbf{v} \in \mathcal{V} : v_n \leq g \text{ a.e.~on } \Gamma_C \}
$$
we obtain as in the domain based cases the boundary integral hemivariational inequality (BIHVI), Problem ($\mathcal{P})$: Find $\mathbf{u} \in \mathcal{K}^{\Gamma}$ such that
\begin{align}
\langle P \mathbf{u}, \mathbf{v}-\mathbf{u} \rangle_{\Gamma_0} +
\int_{\Gamma_C} f^0(u_n(s);v_n(s)-u_n(s)) ds \geq \langle \mathbf{t},\mathbf{v}-\mathbf{u} \rangle_{\Gamma_N} \nonumber \\
+ \langle N \mathbf{f}, \mathbf{v}-\mathbf{u} \rangle _{\Gamma_0} \quad \forall \, \mathbf{v} \in \mathcal{K}^{\Gamma}. \label{BdHemVar}
\end{align}
To shorten the right hand side we introduce the linear functional
\[
\langle \mathbf{F},\mathbf{v}\rangle_{\Gamma_0}=\int _{\Gamma_N} \mathbf{t} \cdot \mathbf{v}\, ds + \langle N \mathbf{f}, \mathbf{v} \rangle_{\Gamma_0}.
\]
\section{Regularization of the nonsmooth functional}
%{\color{blue}
%We impose the following growth condition on $\partial f$, Assumption~\ref{ass:growthCond}, and introduce the functional $\varphi: \mathbf{H}^{1/2} (\Gamma)\times \mathbf{H}^{1/2} (\Gamma) \to \R$ defined by
%\begin{equation} \label{varphi}
%\varphi(\mathbf{u},\mathbf{v})=\displaystyle
%{\int_{\Gamma_C}} f^0(u_n(s); v_n(s)- u_n(s)) \, ds \quad \forall \,
%\mathbf{u}, \mathbf{v} \in \mathbf{H}^{1/2} (\Gamma).
%\end{equation}
%}
%\begin{assumption} \label{ass:growthCond}
%There exist positive constants $c_1$ and $c_2$ such that for all $\xi \in \R$ and $\eta \in \partial f(\xi)$ the following inequalities hold
%\begin{subequations}
% \begin{alignat}{2}
% |\eta| &\leq c_1 (1+|\xi|) & \quad &\forall\, \xi \in \R,\ \eta \in \partial f(\xi)\label{gr_cond1}\\
% \eta ^T \xi &\geq -c_2 |\xi| & \quad &\forall\, \xi \in \R,\ \eta \in \partial f(\xi).
% \end{alignat}
%\end{subequations}
%\end{assumption}
% \begin{description}
% \item (a) $ |\eta| \leq c_1 (1+|\xi|)$ $\forall\, \xi \in \R,\ \eta \in \partial f(\xi)$
% \item (b) $\eta ^T \xi \geq -c_2 |\xi|$ $\forall\, \xi \in \R,\ \eta \in \partial f(\xi)$
% \end{description}
%Due to (\ref{gr_cond1}) the functional $\varphi$ is well-defined, and according to \cite{Ovcharova2012}, pseudomonotone and weakly upper semicontinuous.
%We recall that the functional $\varphi : X\times X \to \R$, where $X$ is a real reflexive Banach space, is pseudomonotone if $u_n\rightharpoonup u$ (weakly) in $X$ and $\displaystyle \liminf _{n \to \infty} \varphi (u_n,u) \geq 0$ imply that, for all $v\in X$, we have $\limsup _{n \to \infty} \varphi (u_n,v)\leq \varphi (u,v).$
In this section, we review from \cite{Ovcharova2012,Ov_Gw} a class of smoothing approximations for nonsmooth functions that can be re-expressed by means of the plus function $\mathrm{p}(t)=t^+=\max \{t, 0 \}$. The approximation is based on smoothing of the plus function via convolution. % $\mathrm{p}(t)=t^+=\max \{t, 0 \}$. % and state some tools that will be used throughout this paper.
%In order to smooth the functional $\varphi$ we introduce from \cite{Ovcharova2012}
%We introduce the smoothing approximation
Let $\varepsilon >0$ be a small regularization parameter.
%We introduce $\hat{f}:\, \R \times \R_{++} \rightarrow \R$,
The smoothing approximation $\hat{f}:\, \R \times \R_{++} \rightarrow \R$ of a locally Lipschitz function $f:\, \R \to \R$ is defined via convolution
by
\[
\hat{f}(x, \varepsilon)=\int_{\R} f(x-\varepsilon t) \rho(t) \, dt,
\]
where %$\varepsilon >0$ is a small regularization parameter,
$\rho \, :\, \R \to \R_{+}$ is a probability density function such that
$$
\kappa= \int_{\R}|t|\, \rho(t)\, dt < \infty
$$
and
\[
\R_{+}=\{\varepsilon\in \R \, :\, \varepsilon \geq 0\}, \quad \R_{++}=\{\varepsilon\in \R \, :\, \varepsilon > 0\}.
\]
We consider a class of nonsmooth functions that can be reformulated by means of the plus function. %$\mathrm{p}(t)=t^+=\max \{t, 0 \}$.
To this class of functions belong maximum, minimum or nested max-min function. %functions which are relevant for aour apllications.
%In general, $\hat{f}$ is not easily applicable in practice, but for a special class of functions like maximum, minimum or nested max-min function, it can be computed explicitly.
If $f$ is a maximum function, for example, $f(x)=\max\{g_1(x),g_2(x)\}$, then $f$ can be reformulated by using the plus function as %$\mathrm{p}(x)=x^+=\max \{x, 0 \}$ as
\begin{equation}\label{max2}
f(x) = g_1(x) + \mathrm{p}( g_2(x)- g_1(x)),
\end{equation}
The smoothing function of $f$ is then defined by
\begin{equation}\label{smooth_S}
S(x,\varepsilon) =g_1(x)+ \hat{p}(g_2(x)-g_1(x),\varepsilon),
\end{equation}
where $\hat{p}( t, \varepsilon)$ is the smoothing function of $\mathrm{p}(t)$ via convolution.
%where $\mathrm{p}: \R \to \R_+$ is the plus function defined by $\mathrm{p}(x)=x^+=\max \{x, 0 \}$.
Choose, for example, the Zang probability density function
\[
\rho(t)= \left \{ \begin{array} {ll} 1 & \mbox{if} \, - \frac{1}{2}\leq t \leq
\frac{1}{2} \\[0.2cm]
0 & \mbox{otherwise} \, .
\end{array} \right.
\]
Then
%we obtain for the smoothing approximation $\hat{\mathrm{p}}(t,\varepsilon)$ of $\mathrm{p}(t)$ defined via convolution,
\[
\hat{\mathrm{p}}(t,\varepsilon) = \int _\R \mathrm{p}(t-\varepsilon s)\rho(s) \, ds =\left \{ \begin{array} {ll} 0 & \mbox{if} \quad t < -\frac {\varepsilon}{2}\\ [0.1cm]
\frac{1}{2\varepsilon}(t+ \frac{\varepsilon}{2})^2 & \mbox{if} \, -
\frac{\varepsilon}{2} \leq t \leq \frac{\varepsilon}{2} \\[0.1cm]
t &\mbox{if} \quad t > \frac
{\varepsilon}{2}
\end{array}
\right.
\]
%is a smoothing function of $\mathrm{p}(t)$.
%Replacing $\mathrm{p}(t)$ by $\hat{\mathrm{p}}(t, \varepsilon)$ in (\ref{max2}) yields % %the smoothing function
%$S:\R\times \R_{++} \to \R$,
%\begin{equation}\label{smooth_S}
%\end{equation}
%the smoothing function of the maximum function $f$, which in this simple case takes the explicit form
and hence,
\begin{equation}\label{Zang}
{\small S(x,\varepsilon)=
\left \{ \begin{array} {ll} g_1(x) & \; \mbox{if } g_2(x) -g_1(x) \leq - \frac{\varepsilon}{2}
\\[0.1cm]
\frac{1}{2\varepsilon} [g_2(x) -g_1(x)]^2 + \frac{1}{2} (g_2(x) +
g_1(x)) + \frac{\varepsilon}{8} & \; \mbox{if } |
g_2(x) -g_1(x)| \leq \frac{\varepsilon}{2}
\\[0.1cm]
g_2(x) & \; \mbox{if } g_2(x) -g_1(x) \geq \frac{\varepsilon}{2} \; .
\end{array}
\right.}
\end{equation}
The relation (\ref{smooth_S})
%We can extend (\ref{smooth_S})
can be extended to the maximum function $f:\R \to \R$ of $m$ continuous functions $g_1, \ldots, g_m$, i.e.
\begin{align} \label{eq:bsp_f_function}
f(x)=\max \{g_1(x), g_2(x), \ldots, g_m(x)\}
\end{align}
by
%of $m$ continuous functions $g_1, \ldots, g_m$.
%The smoothing function $S :\R \times \R_{++} \to \R$ is then given by
\begin{equation} \label{S_general}
S(x, \varepsilon)=
g_1(x) + \hat{\mathrm{p}}\left( g_2(x)-g_1(x)+\ldots +
\hat{\mathrm{p}}\left(
g_m(x)-g_{m-1}(x), \varepsilon\right), \varepsilon\right).
\end{equation}
The major properties of the function $S$ in (\ref{S_general}) are listed in the following lemma:
\begin{lemma}\cite[lemma 10] {Ovcharova2012}
Let $f$ and $S$ be defined as in \eqref{eq:bsp_f_function}, \eqref{S_general}, respectively, with $g_i \in C^1$. Then there holds:
\begin {description}
\item (i) For any $\varepsilon >0$ and for all $x\in \R$,
\[
|S(x,\varepsilon)-f(x)|\leq (m-1)\kappa \varepsilon.
\]
\item(ii) The function $S$ is continuously differentiable on $\R\times \R_{++}$ and for any $x\in \R$ and $\varepsilon>0$ there exist $\Lambda_i \in [0,1]$ such that $\displaystyle \sum_{i=1}^m \Lambda_i=1$ and
\begin {equation}\label{f1}
\frac{\partial S(x,\varepsilon)}{\partial x} = S_x(x, \varepsilon)=\displaystyle \sum_{i=1}^m \Lambda_i g'_i(x).
\end{equation}
Moreover,
\begin{equation} \label{f1_1}
\left\{\lim_{z\to x, \varepsilon \to 0^+} S_x(z, \varepsilon)\right\}\subseteq \partial f(x).
\end{equation}
\end {description}
\end{lemma}
\begin{rem}
Since $f(x)=\min\{g_1(x), \ldots, g_m(x)\} = -\max \{-g_1(x), \ldots, -g_m(x)\}$ these type of functions can be handled as above.
\end{rem}
\begin{assumption}
Assume that there exists positive constants $c_i, d_i$ such that for all $x\in \R$
\begin{subequations}
\begin{alignat}{2}
| g'_i(x)| &\leq c_i(1+|x|) \label{ass_1}\\
g'_i(x) \cdot (-x) &\leq d_i|x|. \label{ass_2}
\end{alignat}
\end{subequations}
\end{assumption}
%For the proof of the following lemma we refer to \cite{Ovcharova2012}.
By using (\ref{f1}) - (\ref{f1_1}) and (\ref{ass_1}) - (\ref{ass_2}), the following auxiliary result can be easily deduced.
\begin{lemma} For any $x, z, \xi \in \R$ it holds that
%Under the assumptions (\ref{ass_1}) - (\ref{ass_2}), the following estimates hold for any $x, z\in \R$:
\begin{subequations}
\begin{alignat}{2}
\left| S_x (x, \varepsilon) \cdot z \right | &\leq c(1+|x|)\,|z| \label{ass_3}\\
S_x (x, \varepsilon) \cdot (-x) &\leq d|x| \label{ass_4}\\
\limsup_{z\to x, \varepsilon\to 0^+} S_x(\varepsilon,x)\, \xi &\leq f^0(x;\xi). %\quad \forall \xi \in \R.
%\left| S_x (x, \varepsilon) \right | &\leq \sqrt{m} \, c(1+|x|).
%\limsup
\label{ass_5}
\end{alignat}
\end{subequations}
\end{lemma}
Further, we introduce the functional $J_\varepsilon :\mathbf{H}^{1/2} (\Gamma) \to \R$ defined by
$$
J_\varepsilon(\mathbf{u})=\int_{\Gamma_C} S( u_n(s), \varepsilon) \, ds.
$$
The regularized problem $(\mathcal{P}_{\varepsilon})$ of ($\mathcal{P}$) is given by:~Find $\mathbf{u}_\varepsilon \in \mathcal{K}^{\Gamma}$ such that
\begin{equation} \label{reg}
\langle P \mathbf{u}_{\varepsilon},\mathbf{v}-\mathbf{u}_{\varepsilon} \rangle_{\Gamma_0} +
\langle D J_\varepsilon(\mathbf{u}_\varepsilon), \mathbf{v} -\mathbf{u}_\varepsilon \rangle_{\Gamma_C}
\geq \langle \mathbf{F} , \mathbf{v}-\mathbf{u}_{\varepsilon} \rangle_{\Gamma_0} \quad \forall \, \mathbf{v} \in \mathcal{K}^{\Gamma},
\end{equation}
where $D J_\varepsilon :\mathbf{H}^{1/2} (\Gamma) \to \mathbf{H}^{1/2} (\Gamma)$ is the G\^{a}teaux derivative of the functional $J_\varepsilon$ and is given by
\[
\langle D J_\varepsilon(\mathbf{u}), \mathbf{v} \rangle_{\Gamma_C} = \int_{\Gamma_C} S_x ( u_n (s),\varepsilon) v_n(s) \, ds.
\]
To simplify the notations we
introduce $\varphi: \mathbf{H}^{1/2} (\Gamma)\times \mathbf{H}^{1/2} (\Gamma) \to \R$ defined by
\begin{equation} \label{varphi}
\varphi(\mathbf{u},\mathbf{v})=\displaystyle
{\int_{\Gamma_C}} f^0(u_n(s); v_n(s)- u_n(s)) \, ds \quad \forall \,
\mathbf{u}, \mathbf{v} \in \mathbf{H}^{1/2} (\Gamma).
\end{equation}
The regularized version of $\varphi(\cdot,\cdot)$ is therewith $\varphi_{\varepsilon} : \mathbf{H}^{1/2} (\Gamma) \times \mathbf{H}^{1/2} (\Gamma) \to \R$, %defined by
\[
\varphi_{\varepsilon}(\mathbf{u}, \mathbf{v}):=\langle D J_\varepsilon(\mathbf{u}), \mathbf{v}- \mathbf{u} \rangle_{\Gamma_C}.
\]
The existence result for (\ref{BdHemVar}), resp. (\ref{reg}), is due to \cite{Gwinner_PhD} and relies in both cases on the pseudomonotonicity of $\varphi$ and $\varphi_{\varepsilon}$, respectively. For details we refer the reader to \cite{Ovcharova2012,Ov_Gw}.
Finally, we recall that the functional $\varphi : X\times X \to \R$, where $X$ is a real reflexive Banach space, is pseudomonotone if $u_n\rightharpoonup u$ (weakly) in $X$ and $\displaystyle \liminf _{n \to \infty} \varphi (u_n,u) \geq 0$ imply that, for all $v\in X$, we have $\limsup _{n \to \infty} \varphi (u_n,v)\leq \varphi (u,v).$
\section{Uniqueness results} \label{section4}
In this section, we discuss the uniqueness of solution of the boundary hemivariational inequality %, which in general does not hold,
and of the corresponding regularization problem. The main results are presented in Theorem \ref{theo2} which is based on the abstract uniqueness Theorem \ref{theo0} from \cite{Ovcharova2015} that gives a sufficient condition for uniqueness. % of the solution of the boundary hemivariational inequality. % to have a unique solution.
Similar uniqueness result but for the regularized problem is derived in Theorem \ref{theouniq}. We are also dealing with the question which classes of smoothing functions preserve the property of unique solvability of the original problem as $\varepsilon \to 0^+$. % of the hemivariational inequality.
\subsection{Uniqueness of the BIHVI}
\begin{assumption}
Assume that there exists an $\alpha_0 \in [0,c_P)$ such that for any $\mathbf{u}, \mathbf{v} \in \mathcal{V}$ it holds
\begin{equation}
\varphi(\mathbf{u},\mathbf{v})+\varphi(\mathbf{v},\mathbf{u})\leq \alpha_0 \|\mathbf{u}-\mathbf{v}\|^2_{\mathcal{V}}. \label{unique}
\end{equation}
To make the results of uniqueness self-consistent, we introduce from \cite{Ovcharova2015} the following theorem.
\end{assumption}
\begin{theorem}\cite[Theorem 5.1]{Ovcharova2015}\label{theo0}
Under the assumption (\ref{unique}) with $\alpha_0<c_P$, there exists a unique solution to the BIHVI problem $(\mathcal{P})$, which depends Lipschitz continuously on the right hand side $\mathbf{F} \in \mathcal{V}^*$.
\end{theorem}
{\bf Proof}\, Assume that $\mathbf{u}$, $\tilde{\mathbf{u}}$ are two solutions of $(\mathcal{P})$. Then the inequalities below hold:
\begin{align*}
\langle P\mathbf{u}-\mathbf{F}, \mathbf{v} -\mathbf{u}\rangle_{\Gamma_0} +
\varphi(\mathbf{u},\mathbf{v}) &\geq 0
\quad \forall \, \mathbf{v}\in \mathcal{K}^{\Gamma} \\
%$$
%$$
\langle P \tilde{\mathbf{u}}-\mathbf{F}, \mathbf{v} -\tilde{\mathbf{u}}\rangle_{\Gamma_0} +
\varphi(\tilde{\mathbf{u}}, \mathbf{v}) &\geq 0
\quad \forall \, \mathbf{v}\in \mathcal{K}^{\Gamma}.
\end{align*}
Setting $\mathbf{v}=\tilde{\mathbf{u}}$ in the first inequality and
$\mathbf{v}=\mathbf{u}$ in the second one, and summing up the resulting inequalities, we get
\begin{eqnarray} \label{unique01}
\langle P\mathbf{u} -P \tilde{\mathbf{u}}, \tilde{\mathbf{u}}
-\mathbf{u} \rangle_{\Gamma_0} +
\varphi (\mathbf{u},\tilde{\mathbf{u}}) +
\varphi (\tilde{\mathbf{u}},\mathbf{u}) \geq 0.
\end{eqnarray}
From the coercivity of the operator $P$ and the assumption (\ref{unique}) we obtain
$$
c_P\|\mathbf{u}-\tilde{\mathbf{u}}\|_{\mathcal{V}}^2 \leq \varphi (\mathbf{u} ,\tilde{\mathbf{u}}) +
\varphi (\tilde{\mathbf{u}} ,\mathbf{u} )\leq
\alpha_0 \|\mathbf{u}-\tilde{\mathbf{u}}\|^2_{\mathcal{V}}.
$$
Hence, since $\alpha_0 \in [0, c_P)$, if $\mathbf{u} \neq \tilde{\mathbf{u}}$ we receive a
contradiction. \\
Now let $\mathbf{F}_i \in \mathcal{V}^*$ and denote $\mathbf{u}_i=\mathbf{u}_{\mathbf{F}_i}, \; i=1,2.$
Analogously to (\ref{unique01}), we find that
$$
\langle P\mathbf{u}_1 -\mathbf{F}_1-P \mathbf{u}_2+\mathbf{F}_2, \mathbf{u}_2
-\mathbf{u}_1 \rangle_{\Gamma_0} +
\varphi (\mathbf{u}_1 ,\mathbf{u}_2) +
\varphi (\mathbf{u}_2 ,\mathbf{u}_1 ) \geq 0.
$$
Hence,
$$
c_P\|\mathbf{u}_1 -\mathbf{u}_2 \|^2_{\mathcal{V}}\leq \varphi (\mathbf{u}_1 ,\mathbf{u}_2 ) +
\varphi (\mathbf{u}_2 ,\mathbf{u}_1) + \langle \mathbf{F}_1-\mathbf{F}_2, \mathbf{u}_2-\mathbf{u}_1 \rangle_{\Gamma_0}
$$
and by (\ref{unique}),
$$
(c_P-\alpha_0)\|\mathbf{u}_1-\mathbf{u}_2\|^2_{\mathcal{V}}\leq \langle \mathbf{F}_1-\mathbf{F}_2, \mathbf{u}_2-\mathbf{u}_1 \rangle_{\Gamma_0} \leq \|\mathbf{F}_1-\mathbf{F}_2\|_{\mathcal{V}^*} \|\mathbf{u}_1 -\mathbf{u}_2 \|_{\mathcal{V}} \, .
$$
Also, since $\alpha_0 <c_P$ we deduce that
$$
\|\mathbf{u}_1 -\mathbf{u}_2 \|_{\mathcal{V}} \leq \frac{1}{c_P-\alpha_0} \, \|\mathbf{F}_1-\mathbf{F}_2\|_{\mathcal{V}^*},
$$
which concludes the proof of the theorem.
\qed
Further, we present a class of locally Lipschitz functions for which the crucial assumption (\ref{unique}) is satisfied.
Let $f:\R \to \R$ be a function such that
\begin{equation} \label{clmont}
( \xi^*-\eta^*)\, (\xi- \eta) \geq -\alpha_0 |\xi -\eta |^2 \quad
\forall \, \xi^*\in \partial f(\xi), \; \forall \, \eta^*\in \partial f(\eta)
\end{equation}
for any $\xi, \eta \in \R$ and some $\alpha_0 \geq 0$.
From the definition of the Clarke generalized derivative \cite{Clarke} we get
$$
f^0(\xi;\eta-\xi)= \max _{\xi^*\in \partial f(\xi)} \xi^* \, (\eta -\xi).
$$
Rewriting (\ref{clmont}) as
$$
\xi^* \, (\eta- \xi) + \eta^* \, (\xi- \eta) \leq \alpha_0 |\xi -\eta |^2
$$
we find
\begin{equation} \label{unique_f}
f^0(\xi;\eta-\xi)+ f^0(\eta;\xi-\eta) \leq \alpha_0|\xi-\eta|^2.
\end{equation}
%We introduce $\Pi : \mathbf{H}^{1/2}(\Gamma) \to L^2(\Gamma_C)$ defined by $\Pi(\mathbf{u})=\mathbf{u}\cdot \mathbf{n}=u_n$. The mapping $\Pi$ is linear and continuous, i.e.
%\begin{equation} \label{Pi}
%\|\Pi \mathbf{v} \|_{L^2(\Gamma_C)}\leq \|\mathbf{v}\|_{H^{1/2}(\Gamma)} \quad \forall \, \mathbf{v} \in \mathbf{H}^{1/2}(\Gamma).
%\end{equation}
%From (\ref{unique_f}) and (\ref{Pi}), we obtain
Hence,
\begin{align}
\varphi(\mathbf{u},\mathbf{v})+\varphi(\mathbf{v},\mathbf{u})&=
\int_{\Gamma_C} f^0( u_n; v_n - u_n )\, ds + \int_{\Gamma_C} f^0( v_n; u_n - v_n )\, ds
\nonumber \\
& \leq \alpha_0 \| u_n - v_n\|^2_{L^2(\Gamma_C)} \leq \alpha_0 \|\mathbf{u}-\mathbf{v}\|^2_{\mathcal{V}} \label{est1_ex1}
\end{align}
and consequently the assumption (\ref{unique}) is satisfied provided that $\alpha_0$ is sufficiently small ($\alpha_0<c_P$).
Next, we show that if $\partial f$ includes only non-negative jumps, then the condition (\ref{clmont}) is globally satisfied. Whereas for negative jumps, (\ref{clmont}) holds only locally. %For the sake of simplicity, we consider a function with one jump only.
%Provided that $\alpha_0>0$ is sufficiently small ( $\alpha_0 < c_P$), then the assumption (\ref{unique}) is satisfied. Here, $c_0>0$ is the operator norm of the mapping $\Pi(\cdot)$.
%Next we show that if $\partial f$ includes no jumps or only positive jumps, the condition (\ref{clmont}) is globally satisfied. For negative jumps, (\ref{clmont}) holds only locally.
%We finish this section with a concrete example for a locally Lipschitz function, for which the assumption (\ref{clmont}) holds.
\begin{figure}[t]
\centering
\includegraphics[trim=0cm 0cm 0cm 0cm,clip,width=0.6\textwidth]{fig1} %posjumps
\caption{An exemplary function $b$ with one positive jump and its anti-derivative $f(x)= \int_0^x b(t)\,dt$ }
\label{nonnegative_jump}
\end{figure}
\begin{example} \label{ex1}
Let $f :\R \to \R$ be a continuous, piecewise $C^{0,1}$ function such that its first derivative $b(x):=f'(x)$ has finite non-negative jumps at the points $x_i^J$ of discontinuity (see Fig. \ref{nonnegative_jump}). This means that %$f(x)= \int_0^x b(t)\,dt$ and moreover,
$\underline{b}(x_i^J)\leq \overline{b}(x_i^J)$, where
$$
\underline{b}(x_i^J):=f'(x_i^J-0)= \lim _{h\to 0^-}\frac{f(x_i^J+h)-f(x_i^J)}{h}
$$
and
$$
\overline{b}(x_i^J):= f'(x_i^J+0)=\lim _{h\to 0^+}\frac{f(x_i^J+h)-f(x_i^J)}{h}.
$$
Hence,
$$
\partial f(x_i^J) =\left[\underline{b}(x_i^J), \, \overline{b}(x_i^J) \right].
$$
%The graph of such a $\partial f$ is plotted in Figure %\ref{positivejumps}.
%\ref{nonnegative_jump}.
%Let $f :\R \to \R$ be a locally Lipschitz function such that the graph of $\partial f$ consists of several decreasing straight segments and nonnegative jumps (see Figure \ref{positivejumps}). In this case
%$$
%\partial f(x) =\left[\underline{f}(x), \overline{f}(x) \right], \quad \underline{f}(x)\leq \overline{f}(x)
%$$
%with
%$$
%\underline{f}(x):=f(x-0):= \lim _{h\to 0^-}\frac{f(x+h)-f(x)}{h},
%$$
%$$
%\overline{f}(x):= f(x+0):=\lim _{h\to 0^+}\frac{f(x+h)-f(x)}{h}
%$$
%and $\underline{f}(x)\leq \overline{f}(x)$.
\\
%{\color{blue}Bild Fehlt}
Setting $\mathcal{E}=\cup\, \mathcal{E}_i$ the finite set of open intervals $\mathcal{E}_i$, where the function $b|_{\mathcal{E}_i}$ is smooth with the Lipschitz constant $c_{L_i}$, we define
%\alpha_0:=\max_{\mathcal{E}_i} \max_{x\in \mathcal{E}_i}|f'(x)|
\begin{equation} \label{eq1_ex1}
\alpha_0=\displaystyle \max_{i} \, c_{L_i} . %|\min \{0, {\color{red}f''}(x)\} |.
\end{equation}
Let $\{x_i^J\}_{i=1}^k$ be the set of jags between $x_1$ and $x_2$ with $ x_2 < x_1^J < \cdots < x_k^J<x_1 $, where $\underline{b}(x_i^J)\leq \overline{b}(x_i^J)$. Then, for any $x_1>x_2$ we have the following
\begin{eqnarray*}
\frac{b(x_1)-b(x_2)}{x_1-x_2} & = & \frac{b(x_1)-b(x_2)+ \displaystyle \sum _{i=1}^k\overline{b}(x^J_i)-\displaystyle \sum _{i=1}^k\overline{b}(x^J_i)}{x_1-x_2} \\
&\geq& \frac{b(x_1)-b(x_2)+ \displaystyle \sum _{i=1}^k\underline{b}(x^J_i)-\displaystyle \sum _{i=1}^k\overline{b}(x^J_i)}{x_1-x_2}\\
&= &\frac{b(x_1)-\overline{b}(x^J_k) -b(x_2)+\underline{b}(x^J_1)
+ \displaystyle \sum _{i=2}^k \big(\underline{b}(x^J_i)-\overline{b}(x^J_{i-1})\big)}{x_1-x_2} \\
&\geq& \frac{-\alpha_0(x_1-x^J_k)-\alpha_0(x^J_1-x_2)-\alpha_0 \displaystyle \sum _{i=2}^k (x^J_i -x^J_{i-1})}{x_1-x_2}\\
&=& \frac{-\alpha_0(x_1-x_2)}{x_1-x_2}=-\alpha_0,
\end{eqnarray*}
from which the assumption (\ref{clmont}) follows immediately.
%Let $\{x_i^J\}_{i=1}^k$ be the set of jags between $x_1$ and $x_2$. Then,
%for any $x_1>x_2$ we can estimate as follows:
%\begin{eqnarray*}
%\frac{\underline{f}(x_1)-\overline{f}(x_2)}{x_1-x_2} & = & \frac{\underline{f}(x_1)-\overline{f}(x_2)+ \displaystyle \sum _{i=1}^k\overline{f}(x^J_i)-\displaystyle \sum _{i=1}^k\overline{f}(x^J_i)}{x_1-x_2} \\
%&\geq& \frac{\underline{f}(x_1)-\overline{f}(x_2)+ \displaystyle \sum _{i=1}^k\underline{f}(x^J_i)-\displaystyle \sum _{i=1}^k\overline{f}(x^J_i)}{x_1-x_2}\\
%&= &\frac{\underline{f}(x_1)-\overline{f}(x^J_k) -\overline{f}(x_2)+\underline{f}(x^J_1)
%+ \displaystyle \sum _{i=2}^k \big(\underline{f}(x^J_i)-\overline{f}(x^J_{i-1})\big)}{x_1-x_2} \\
%&\geq& \frac{-\alpha_0(x_1-x^J_k)-\alpha_0(x^J_1-x_2)-\alpha_0 \displaystyle \sum _{i=2}^k (x^J_i -x^J_{i-1})}{x_1-x_2}\\ &=&
%\frac{-\alpha_0(x_1-x_2)}{x_1-x_2}=-\alpha_0.
%\end{eqnarray*}
%Hence, for any $x_1^* \in \partial f(x_1)$ and $x_2^* \in \partial f(x_2)$ with $x_1>x_2$, we have
%\begin{eqnarray*}
%x_1^*-x_2^* &= &\lambda \underline{f} (x_1) + (1-\lambda) \overline{f} (x_1) - \beta \underline{f} (x_2) + (1-\beta) \overline{f} (x_2)\\
%&\geq& \lambda \underline{f} (x_1) + (1-\lambda) \underline{f} (x_1)- \beta \overline{f} (x_2) - (1-\beta) \overline{f} (x_2) = \underline{f} (x_1) - \overline{f} (x_2) \geq -\alpha_0 (x_1-x_2)
%\end{eqnarray*}
%for some $\lambda, \beta \in [0,1]$, from which the assumption (\ref{clmont}) follows immediately. %Here, $k$ stands for the number of the jags between $x_1$ and $x_2$, i.e. the set $\{x_i^J\}$ is ordered by $i=1,\ldots,k$.
\qed
\end{example}
\begin{rem}
%\begin{enumerate}
%\item
If the graph of $\partial f$ consists of several decreasing straight line segments and non-negative jumps, %as shown in Fig. \ref{positivejumps},
then the value $-\alpha_0$ in (\ref{clmont}) is the steepest decreasing slope.
%\item {\color{red} If $b|_{\mathcal{E}_i}$ is monotonically increasing and if the jumps are non-negative, then $\alpha_0=0$.}
%\end{enumerate}
\end{rem}
%\begin{figure}[b]
%\centering
%\includegraphics[trim=0cm 17cm 0cm 6cm,clip,width=0.8\textwidth]{posjumps} %posjumps
%\caption{An example of generalized gradient with non-negative jumps}
% \label{positivejumps}
%\end{figure}
The next example treats the case of negative jumps.
\begin{example} \label{ex2}
Let $f :\R \to \R$ be a continuous, piecewise $C^{0,1}$ function such that its first derivative has finite negative jumps at the points $x_i^J$ of discontinuity (see Fig. \ref{negative_jump}).
%Let $f :\R \to \R$ be a locally Lipschitz function with $\partial f$ as depicted on Fig.~\ref{02} containing only negative jumps.
In this case,
$$
\partial f(x_i^J) =\left[\overline{b}(x_i^J), \underline{b}(x_i^J) \right].
$$
%Set
%\[
%\alpha_0=\displaystyle \max_{E_i} \max _{x\in E_i} |\min \{0, f'(x)\} |.
%\]
Then, for any $x_1>x_2$, we have
\begin{equation} \label{ex2_0}
\frac{b(x_1)-b(x_2)}{x_1-x_2} \geq -\alpha_0 + \frac{\displaystyle \sum _{i=1}^k c^J_i}{x_1-x_2},
\end{equation}
with $\alpha_0\geq 0$ as defined in (\ref{eq1_ex1}) %$\alpha > 0$ stands for the lower bound of all slopes of increasing segments of $\partial f$ and
and $c^J_i=\overline{b}(x^J_i)-\underline{b}(x^J_i)<0$ standing for the value of the jump in the point $x^J_i$. \\
\begin{figure}[t]
\centering
\includegraphics[trim=0cm 0cm 0cm 0cm,clip,width=0.6\textwidth]{fig2} %posjumps
\caption{An exemplary function $b$ with one negative jump and its anti-derivative $f(x)=\int_0^x b(t) \, dt$}
\label{negative_jump}
\end{figure}
Further, for the sake of simplicity, we consider a function with one negative jump $c_J$ and estimate as follows:
\begin{equation} \label{ex2_1}
c_J(x_1-x_2) = c_J(x_1-x_J) + c_J(x_J-x_2) = c_J(x_1-x_J)_{+} + c_J(x_J-x_2)_{+}.
\end{equation}
%\begin{figure}[h]
%\begin{center}
%\includegraphics[trim=0cm 3cm 0cm 5cm,clip,height=13cm, width=13cm]{ particularn01}
%\end{center}
%\vspace{-9cm}
%\caption{\small An example of generalized gradient with negative jumps }\label{negativejumps}
%\end{figure}
Using $x^2\geq (x-a)_{+}$ for any $a\geq \frac{1}{4}$, we get
$$
(x_1-x_2)^2\geq (x_1-x_2-\frac{1}{4})_{+}\geq (x_1-x_2-(x_J-x_2))_{+}=(x_1-x_J)_{+} \quad \forall \, x_2\leq x_J -\frac{1}{4}
$$
as well as
$$
(x_1-x_2)^2\geq (x_1-x_2-\frac{1}{4})_{+}\geq (x_1-x_2-(x_1-x_J))_{+}=(x_J-x_2)_{+}
\quad \forall \, x_1\geq x_J +\frac{1}{4}.
$$
%Combining one of the last two inequalities with (\ref{ex2_1}) and (\ref{ex2_0}) yields
%Hence,
%$$
%\frac{\underline{f}(x_1)-\overline{f}(x_2)}{x_1-x_2} \geq 2c_J\quad \forall \, x_1\geq x_J +\frac{1}{4} \; \; \mbox{and} \; \; \forall \, x_2\leq x_J -\frac{1}{4}.
%$$
Hence, the condition (\ref{clmont}) is fulfilled with $\alpha_0 -2c_J>0$, but only for those $x_1, x_2$ satisfying $x_1\geq x_J + \frac{1}{4}$ and $x_2 \leq x_J - \frac{1}{4}$.
%In conclusion, in case of negaive jump $c^J$, uniqueness can only be validated a posteriori (\see also \cite{}) when the solution is known and we need additional local condition for uniqueness, namely,
%\begin{equation}
%u_n(x)\leq x^J- \frac{1}{4} \; \; \mbox{or} \; \; v_n(x)\geq x^J+ \frac{1}{4} \quad \mbox{on} \; \Gamma_C.
%\end{equation}
%In case of $k$ jags between $x_1$ and $x_2$, the assumption (\ref{clmont}) holds again, but now for any $x_1$ and $x_2$ such that $x_1 \geq x^J_k + \frac{1}{4}$ and
%$x_2\leq x^J_1 - \frac{1}{4}$, and vice versa.
%In this case the constant $\alpha_0$ is defined by means of $\alpha_0:=-2\min\{c^J_i : i=\overline{1,k}\}$.}
Let $h(t)$ be a Lipschitz continuous function with Lipschitz constant $c_L$. We include the negative jump $c^J$ at the point $t_J$ through the Heaviside function $\Theta$ by
\[
b(t)=h(t)+c^J\Theta(t-t_J)%, \; a>0, \; c^J<0.
\]
%Further, we consider $ f(t):=\int_0^t b(s) \, ds$
and compute
\[
f^0(t;d)=\left \{ \begin{array} {ll}
h(t)d, & \, \mbox{if} \; t<t_J\\[0.1cm]
h(t)d+c_Jd, & \, \mbox{if} \; t> t_J,
\end{array}
\right.
\quad f^0(t_J;d)=\left \{ \begin{array} {ll}
h(t_J)d, & \, \mbox{if} \; d>0\\[0.1cm]
h(t_J)d+c_Jd, & \, \mbox{if} \; d\leq 0.
\end{array}
\right.
\]
Hence, for any $t_1, t_2 <t_J$, we have
\begin{align}
f^0(t_1;t_2-t_1)&+ f^0(t_2;t_1-t_2)= h(t_1)(t_2-t_1)+h(t_2)(t_1-t_2) \nonumber \\[0.1cm]
=&(h(t_2)-h(t_1))(t_1-t_2)\leq c_L(t_1-t_2)^2. \label{est1_ex2}
\end{align}
Analogously, for any $t_1, t_2 > t_J$, we get
\begin{align*}
f^0(t_1;t_2-t_1)+f^0(t_2;t_1-t_2)=& h(t_1)(t_2-t_1)+c_J(t_2-t_1)+h(t_2)(t_1-t_2)\\[0.1cm]
&+c_J(t_1-t_2)\leq c_L(t_1-t_2)^2.
\end{align*}
In both cases, the condition (\ref{unique_f}) is satisfied.
%Then,
%\[
%\varphi(u,\tilde{u})+ \varphi(\tilde{u},u) = \int_{\Gamma_C} f^0(u_n;\tilde{u}_n-u_n)\,ds + \int_{\Gamma_C} f^0(\tilde{u}_n;u_n-\tilde{u}_n)\,ds
%\]
\\
Let now $t_1< t_J \leq t_2$, then we have
\begin{align} \label{est2_ex2}
f^0(t_1;t_2-t_1)&+f^0(t_2;t_1-t_2)= h(t_1)(t_2-t_1)+h(t_2)(t_1-t_2) + c_J(t_1-t_2) \nonumber \\[0.1cm]
& \quad \leq c_L(t_1-t_2)^2 + c_J(t_1-t_2) \leq \alpha_0 (t_1-t_2)^2
\end{align}
for some $\alpha_0$ if and only if
\begin{align} \label{est3_ex2}
\frac{c_J}{t_1-t_2}+c_L \leq \alpha_0.
\end{align}
Thus, we assume
\begin{align} \label{est4_ex2}
c_L<\alpha_0 \quad \mbox{and} \quad t_1 \leq t_J- \frac{c_J}{c_L-\alpha_0},
\end{align}
which, since $t_J \leq t_2$, implies (\ref{est3_ex2}) immediately. \\
%{\color{green} Inequality (\ref{est3_ex2}) implies, that as long as not both values $t_1, t_2$ approach the jump at $t_J$, (\ref{unique_f}) can still be shown. Hat diese Ungleichung eine geometrische Bedeutung, ala wenn man nur $t_1,t_2$ hat sieht man den Sprung eigentlich nicht? }{\color{red} muss nachdenken..., $t_J$ kommt nicht vor, aber $c_J$ schon; {\bf Uniqueness result for an adhesive law with one negative jump, compare to Nesemann?} }\\
%{\color{magenta} ABER von oben f\"{u}r alle $t_1\leq t_J-\frac{1}{4}$ und $t_2\geq t_J+\frac{1}{4}$ gilt
% \[
% -c_J(t_2-t_1)=-c_J(t_2-t_J)_+ -c_J(t_J-t_1)_+ \leq -2c_j (t_2-t_1)^2
% \]
% und somit
% \[
% f^0(t_1;t_2-t_1)+f^0(t_2;t_1-t_2) \leq (c_L-2c_J)(t_1-t_2)^2.
% \]
% }
Based on the results presented in the Examples~\ref{ex1} and \ref{ex2} we derive the following uniqueness result.
\begin{theorem} \label{theo2}
Let $\mathbf{u}$ be a solution of the BIHVI problem $(\mathcal{P})$. Then $\mathbf{u}$ is unique if one of the following conditions holds:
\begin{itemize}
\item [(a)] The jumps $c^i_J$ are non-negative and
%the constant $\alpha_0$ defined as in (\ref{eq1_ex1})
the Lipschitz constant $c_L$ satisfies $c_L < c_P$.
%\item [(b)] {\color{red} The jumps $c^i_J$ are non-negative and $b|_{\mathcal{E}_i}$ is monotonically increasing.}
\item [(b)] The jump $c_J$ is negative, $c_L<c_P$
and ${u}_n(x)\leq t_J-\frac{1}{\tilde{c}}$ on $\Gamma_C$
for some positive $\tilde{c}$ such that $\tilde{c}<-\frac{c_P-c_L}{c_J}$.
\end{itemize}
\end{theorem}
{\bf Proof} (b)\, Let $\mathbf{u} \in \mathcal{V}$ be such that ${u}_n(x)\leq t_J-\frac{1}{\tilde{c}}$ on $\Gamma_C$ and $\mathbf{v} \in \mathcal{V}$ arbitrary. With $\alpha_0:= c_L-\tilde{c}\,c_J$, using (\ref{est1_ex2}) and (\ref{est2_ex2}), we obtain
\begin{align}
\varphi(\mathbf{u}, \mathbf{v}) & + \varphi(\mathbf{v}, \mathbf{u}) = \int_{\Gamma_C } f^0(u_n;v_n-u_n) \, ds + \int_{ \Gamma_C } f^0(v_n;u_n-v_n) \, ds \nonumber \\
=& \int_{ \{ s\in \Gamma_C \,:\, v_n(s)<t_J \}} f^0(u_n;v_n-u_n) \, ds + \int_{ \{ s\in \Gamma_C \,:\, v_n(s)\geq t_J \}} f^0(u_n;v_n-u_n) \, ds \nonumber \\
& + \int_{ \{ s\in \Gamma_C \,:\, v_n(s)<t_J \}} f^0(v_n;u_n-v_n) \, ds + \int_{ \{ s\in \Gamma_C \,:\, v_n(s)\geq t_J \}} f^0(v_n;u_n-v_n) \, ds \nonumber \\
\leq & \int_{ \{ s\in \Gamma_C \,:\, v_n(s)<t_J \}} c_L (u_n-v_n)^2 \, ds + \int_{ \{ s\in \Gamma_C \,:\, v_n(s) \geq t_J \}} \alpha_0 (u_n-v_n)^2 \, ds \nonumber \\
\leq & \max \{c_L, \alpha_0\} \left\|u_n-v_n \right\|^2_{L^2(\Gamma_C)} \leq \alpha_0 \left\|\mathbf{u}-\mathbf{v} \right\|^2_{\mathcal{V}}. \label{est5_ex2}
\end{align}
Further, we assume that $\mathbf{\tilde{u}} \in \mathcal{V}$ is another solution of $(\mathcal{P})$. Following the proof of Theorem \ref{theo0}, in virtue of $\alpha_0 = c_L-\tilde{c}\,c_J < c_p$ and (\ref{est5_ex2}) with $\mathbf{v}:=\mathbf{\tilde{u}}$, we obtain a contradiction. \qed
\begin{rem}
We note that our uniqueness result is more general then the result in \cite[Theorem 7]{Nesemann}, which demands $\tilde{c}=4$.
\end{rem}
%\begin{enumerate}
%\item
% either $\mathbf{u}_n(x)<t_J$ or $\mathbf{u}_n(x)\geq t_J$ on $\Gamma_C$. Then, $\mathbf{u}$ is a unique solution of $(\mathcal{P})$. %Under this assumption, $\mathbf{u}$ is the unique solution of $(\mathcal{P})$.
% \end{theorem}
%{\bf Proof} Indeed, assume that
%$\mathbf{u}, \tilde{\mathbf{u}}$ are two solutions of $(\mathcal{P})$ such that $\mathbf{u}_n(x)<t_J$ and $\tilde{\mathbf{u}}_n(x)<t_J$ on $\Gamma_C$. Following the proof of Theorem \ref{theo0}, we obtain
%\[
%c_p\|\mathbf{u}-\tilde{\mathbf{u}}\|^2_{\mathcal{V}}\leq 0,
%\]
% which, if $\mathbf{u}\neq \tilde{\mathbf{u}}$, leads to a contradiction. \qed \\
% We emphasize that for negative jumps, uniqueness can be only established a posteriori, if the solution is known.
\end{example}
\subsection{Uniqueness of the regularized problem}
\begin{assumption}
Assume that the assumption (\ref{clmont}) is satisfied for the regularization function, i.e.~there exists a constant $\alpha_0 \geq 0$ (in general depending on $\varepsilon>0$) such that
\begin{equation} \label{uniqreg}
\left ( S_x(x_1, \varepsilon)- S_x(x_2, \varepsilon) \right ) (x_1-x_2) \geq - \alpha_0|x_1-x_2|^2 \quad \forall \, x_1, x_2 \in \R.
\end{equation}
\end{assumption}
Hence, for any $\mathbf{u}, \mathbf{v} \in \mathcal{V}$, we have
\begin{align}
\langle D J_{\varepsilon}(\mathbf{u}) - D J_{\varepsilon}(\mathbf{v}),\mathbf{v}-\mathbf{u}\rangle_{\Gamma_C}
&= \int_{\Gamma_C}\left ( S_x( {u}_n(s),\varepsilon) - S_x(
{v}_n(s),\varepsilon)\right )(v_n(s)- {u}_n(s))\,ds \nonumber \\
&\leq \alpha_0 \| {u}_n- {v}_n\|^2_{L^2(\Gamma_C)} \leq \alpha_0 \|\mathbf{u}-\mathbf{v}\|^2_{\mathcal{V}}. \label{ass1_th5}
\end{align}
%By similar arguments as in the proof of
Due to Theorem~\ref{theo0}, we have the following uniqueness result for the regularized problem.
\begin{theorem}\label{theouniq}
Under the assumption \eqref{uniqreg} with $\alpha_0 < c_p$, there exists a unique solution to the regularized problem $(\mathcal{P}_{\varepsilon})$, which depends Lipschitz continuously on the right hand side $\mathbf{F} \in \mathcal{V}^*$.
\end{theorem}
%{\bf Proof}\,
%For any $\mathbf{u}, \mathbf{v} \in \mathcal{V}$, we have
%\begin{equation} \label{ass1_th5}
% \langle D J_{\varepsilon}(\mathbf{u}) - D J_{\varepsilon}(\mathbf{v}),\mathbf{v}-\mathbf{u}\rangle_{\Gamma_C}
%= \int_{\Gamma_C}\left ( S_x( \mathbf{u}_n(s),\varepsilon) - S_x(
%\mathbf{v}_n(s),\varepsilon)\right )(\mathbf{v}_n(s)- \mathbf{u}_n(s))\,ds \leq \alpha_0 \| \mathbf{u}_n- \mathbf{v}_n\|^2_{L^2(\Gamma_C)} \leq \alpha_0 \|\mathbf{u}-\mathbf{v}\|^2_{\mathcal{V}}.
%\end{equation}
%Hence, since $\alpha_0 < c_P$, the condition (\ref{unique}) is satisfied.
%The statement of this theorem is now a direct consequence of Theorem~\ref{theo0}. \qed \\
One of the most interesting question is, which classes of smoothing functions preserve the property of unique solvability of the original problem as $\varepsilon \to 0^+$. As the next example shows, this is the case for the smoothing function (\ref{Zang}) relevant to our computation. In particular, we show that the constant $\alpha_0$ in (\ref{uniqreg}) does not depend on $\varepsilon$.
\begin{example} \label{ex3} In view of our application to contact problems with nonmonotone laws in the boundary conditions, we consider $f(x)=\max\{g_1(x), g_2(x)\}$ with %the Clarke's subdifferential $\partial f$ depicted in Fig.~\ref{positivejumps}. Let
$
g_i(x)=\frac{a_i}{2}x^2 +c_ix +d_i. %, \quad \alpha_i, c_i \geq 0, \quad i=1,2.
$
%For simplicity, we use an exemplary function with only one jag.
% belongs to the special class of pseudomonotone functions (see Example \ref{exuniq1}) for which (\ref{clmont}) is satisfied. In this case $- \alpha_0$ is the steepest slope of $\partial f$, i.e., $-\alpha_0=\min\{-\alpha_1,\alpha_2\}$.
%\begin{figure}[h]
%\begin{center}
%\includegraphics[trim= 4cm 10cm 0cm 4cm,clip, height=9cm, width=9cm]{fig23onejag}
%\end{center}
%\caption{An example of Clarke's subdifferential with one jag}\label{onejag}
%\end{figure}
We use the smoothing function (\ref{Zang}) as a regularization of $f$. The corresponding approximation of $\partial f$ is given by
\[
S_x(x, \varepsilon) = \left \{ \begin{array} {ll}
g_1'(x) & \mbox{if} \, \; g_2(x) -g_1(x) \leq - \frac{\varepsilon}{2} \\[0.1cm]
\lambda_1 g_1'(x) +\lambda_2 g_2'(x)
& \mbox{if} \, \; \left |g_1(x) -g_2(x) \right | < \frac{\varepsilon}{2} \\[0.1cm]
g_2'(x) & \mbox{if} \, \; g_2(x) -g_1(x) \geq \frac{\varepsilon}{2},
\end{array}
\right.
\]
where
$$
\lambda_2= \frac{1}{\varepsilon} (g_2(x) -g_1(x))+ \frac{1}{2},\quad \lambda_1 =
1 - \lambda_2.
$$
We need to consider only the region $I_{\varepsilon}:=\{x\in \R \, :\, |g_1(x)-g_2(x)|<\frac{\varepsilon}{2}\}$, where $f$ is actually approximated. Otherwise, the condition (\ref{uniqreg}) is automatically satisfied. \\ % with $\alpha_0=\max\{\alpha_1,\alpha_2\}>0$. \\
Let $x_i \in I_{\varepsilon}, \; i=1,2$. Setting $g(x):=g_1(x)-g_2(x)$, we introduce
$$
\lambda_2^i= \frac{1}{\varepsilon}(g_2(x_i)-g_1(x_i))=-\frac{1}{\varepsilon}g(x_i) \; %\forall \, x_i \in I_{\varepsilon},\; %i=1,2 \;
\mbox{and} \; \lambda_1^i=1-\lambda_2^i,\; i=1,2.
$$
Hence,
\begin{equation} \label {defg}
\lambda_1^1-\lambda_1^2 = \lambda_2^2-\lambda_2^1
= \frac{1}{\varepsilon} (g(x_1)-g(x_2))
\end{equation}
and thus,
\begin{align}
S_x(x_1, \varepsilon) - S_x(x_2, \varepsilon) &= \lambda_1^1(a_1 x_1+c_1)+ \lambda_2^1(a_2 x_1 +c_2)
-\lambda_1^2(a_1 x_2 +c_1) \nonumber\\
& \quad - \lambda_2^2(a_2 x_2 +c_2) \nonumber \\
& = a_1 \lambda_1^1 (x_1-x_2)+a_2 \lambda_2^1 (x_1-x_2) %\nonumber \\[0.1cm]
+ (\lambda_1^1-\lambda_1^2)(a_1x_2+c_1) \nonumber\\
& \quad + (\lambda_2^1-\lambda_2^2)(a_2 x_2+c_2)\nonumber\\
& = \left ( a_1 \lambda_1^1 + a_2 \lambda_2^1 \right ) (x_1-x_2)
+(\lambda_1^1-\lambda_1^2) (g_1^{\prime}(x_2)- g_2^{\prime}(x_2))\nonumber\\
&= \left (a_1 \lambda_1^1+a_2 \lambda_2^1 \right )(x_1-x_2)+ (\lambda_1^1-\lambda_1^2) g^{\prime}(x_2). \label{onejag1}
\end{align}
%for any $x_1 , x_2\in I_\varepsilon$. \\
Similarly to (\ref{onejag1}), we find that
\begin{equation} \label{onejag2}
\hspace{-1cm}S_x(x_1, \varepsilon) - S_x(x_2, \varepsilon)= \left( a_1 \lambda_1^2 + a_2 \lambda_2^2\right) (x_1-x_2)
+ (\lambda_1^1-\lambda_1^2) g^{\prime}(x_1).
\end{equation}
Summing now (\ref{onejag1}) and (\ref{onejag2}), we get
\begin{align} \label{onejag3}
2\left (S_x(x_1, \varepsilon) - S_x(x_2, \varepsilon)\right ) &= (a_1 \lambda_1^1+a_2 \lambda_2^1+ a_1 \lambda_1^2 + a_2 \lambda_2^2) (x_1-x_2) \nonumber\\
& \quad
+ (\lambda_1^1-\lambda_1^2) ( g^{\prime}(x_1) +g^{\prime}(x_2) ).
\end{align}
From (\ref{defg}), using the structure of $g_1$ and $g_2$, we find that
\begin{align}
\lambda_1^1-\lambda_1^2 &= \frac{1}{\varepsilon} (g(x_1)-g(x_2)) = \frac{1}{\varepsilon} \left( \frac{a_1-a_2}{2}(x_1^2-x_2^2) +(c_1-c_2)(x_1-x_2)\right )\nonumber \\
&= \frac{1}{\varepsilon} (x_1-x_2) \left( \frac{a_1-a_2}{2} x_1 + \frac{c_1-c_2}{2}+ \frac{a_1-a_2}{2} x_2 + \frac{c_1-c_2}{2} \right )\nonumber \\
&= \frac{1}{2\varepsilon} (x_1-x_2) ( g^{\prime}(x_1) +g^{\prime}(x_2) ). \label{onejag4}
\end{align}
%Finally, with $\alpha_0:=\frac{1}{2}\min\{-\alpha_1,-\alpha_2\}$,
Hence,
\begin{align*}
& 2\left (S_x(x_1, \varepsilon) - S_x(x_2, \varepsilon)\right ) (x_1-x_2)
%&=(a_1 \lambda_1^1+a_2 \lambda_2^1+ a_1 \lambda_1^2 + a_2 \lambda_2^2) (x_1-x_2)^2 + \frac{1}{2\varepsilon} (x_1-x_2)^2 ( g^{\prime}(x_1) +g^{\prime}(x_2) )^2 \\
\geq (a_1 \lambda_1^1+a_2 \lambda_2^1+ a_1 \lambda_1^2 + a_2 \lambda_2^2) (x_1-x_2)^2 \\
& \quad =(a_1 (1-\lambda_2^1)+a_2 \lambda_2^1+ a_1 (1-\lambda_2^2) + a_2 \lambda_2^2) (x_1-x_2)^2\\
%&=(a_1 +(a_2-a_1) \lambda_2^1+ a_1 + (a_2-a_1) \lambda_2^2) (x_1-x_2)^2\\
%&=(2a_1 +(a_2-a_1) \lambda_2^1+ a_1 + (a_2-a_1) \lambda_2^2) (x_1-x_2)^2\\
%& = (a_1 |(1-\lambda_2^1)|+a_2 \lambda_2^1+ a_1 |(1-\lambda_2^2)| + a_2 \lambda_2^2) (x_1-x_2)^2\\
& \quad \geq (-|a_1| |(1-\lambda_2^1)| - |a_2| |\lambda_2^1|- |a_1| |(1-\lambda_2^2)| - |a_2| |\lambda_2^2|) (x_1-x_2)^2\\
& \quad \geq -\max\{|a_1| , |a_2|\} ( |(1-\lambda_2^1)| + |\lambda_2^1|+ |(1-\lambda_2^2)| + |\lambda_2^2|) (x_1-x_2)^2\\
& \quad \geq -\max\{|a_1| , |a_2|\} ( 3/2 + 1/2+ 3/2 + 1/2) (x_1-x_2)^2\\
& \quad =-4\max\{|a_1| , |a_2|\} (x_1-x_2)^2
\end{align*}
and consequently, (\ref{uniqreg}) is satisfied with $\alpha_0 = 2\max\{|a_1| , |a_2|\}$.
\end{example}
\begin{rem}
If we use (\ref{Zang}) as a smoothing approximation for the maximum (resp. minimum) function of several quadratic functions $a_i x^2 +c_i x +d_i$, then (\ref{uniqreg}) is satisfied and, therefore, the regularized problem is always uniquely solvable if all $|a_i|$ are sufficiently small. Since the constant $\alpha_0$ does not depend on $\varepsilon$, the property of uniqueness is preserved as $\varepsilon \to 0^+$. %But, in general, e.g.~in the case of negative jumps, we do not have uniqueness of the original problem.
\end{rem}
\section{Discretization with boundary elements}
To avoid domain approximation, let $\Omega\subset \R^{d}$, $d=2,3$, also be a polygonal domain. Let $\mathcal{T}_h$ be a sufficiently fine finite element mesh of the boundary~$\Gamma$ respecting the decomposition of $\Gamma$ into $\Gamma_D$, $\Gamma_N$ and $\Gamma_C$, %$p: \mathcal{T}_h \rightarrow \mathbb{N}$,
$p=(p_T)_{T\in \mathcal{T}_h}$ a polynomial degree distribution over $\mathcal{T}_h$, $\mathbb{P}_{p_{T}}(\hat{T})$ the space of polynomials of order $p_{T}$ on the reference element $\hat{T}$, and $\Psi_T: \hat{T} \rightarrow T \in \mathcal{T}_h$ a bijective, (bi)-linear transformation. In 2D, $\hat{T}$ is the interval $[-1,1]$, whereas in 3D it is the reference square $[-1,1]^2$. Let $\Sigma_{T, hp}$ be the set of all affinely transformed (tensor product based) Gauss-Lobatto nodes on the element $T$ of the partition $\mathcal{T}_h$ of $\Gamma$, corresponding to the polynomial degree $p_T$, and set $\displaystyle \Sigma_{hp}:=\displaystyle \bigcup_{T\in \mathcal{T}_h|_{\Gamma_C}}\Sigma_{T, hp}$, see \cite{krebs2007p,MS-1,gwinner2013hp}. Furthermore, we assume in this section that $g \in C^0(\overline{\Gamma_C})$ to allow point evaluation.\\
For the discretization of the displacement $\mathbf{u}$ we use
\begin{align*}
\mathcal{V}_{hp}&=\{\mathbf{v_{hp}} \in \mathbf{C}^0(\Gamma)\, :\, \mathbf{v_{hp}}|_{T} \circ \Psi_T \in [\mathbb{P}_{p_{T}
}(\hat{T})]^{d} \quad \forall \, T \in \mathcal{T}_h, \; \mathbf{v_{hp}} =0 \; \mbox{on} \; \overline{\Gamma_D} \} , \\
\mathcal{K}_{hp}^{\Gamma} &=\{ \mathbf{v_{hp}} \in \mathcal{V}_{hp} \, :\, (\mathbf{v_{hp}}\cdot \mathbf{n})(P_i) \leq g(P_i) \quad \forall \, P_i \in \Sigma_{hp} \}.
\end{align*}
In general $\mathcal{K}_{hp}^{\Gamma} \not\subseteq \mathcal{K} ^{\Gamma}$. For the approximation of the Poincar\'e-Steklov operator, namely $V^{-1}$, we need the space
\[
\mathcal{W}_{hp}= \{ \psi_{hp} \in \mathbf{L}^2(\Gamma) \, : \, \psi_{hp}|_{T} \circ \Psi_T \in [\mathbb{P}_{p_T-1}(\hat{T})]^{d} \quad \forall \, T \in \mathcal{T}_h\} \subset \mathbf{H}^{-1/2}(\Gamma).
\]
For more details on the approximation techniques based on boundary element method see e.g. \cite{ccjg,Co-1,CoSt,Gu,GS-1,Ha,MS-1,MS-2,SlSp}.
Let $\{\phi_i\}_{i=1}^{N_D}$ and $\{\psi_j\}_{j=1}^{N_N}$ be a bases of $\mathcal{V}_{hp}$ and $\mathcal{W}_{hp}$, respectively. Then, the boundary matrices are given by
\begin{align*}
(V_{hp})_{j,i} &= \langle V \psi_i, \psi_j\rangle, \quad (K_{hp})_{j,i} = \langle K \phi_i, \psi_j\rangle, \\
(W_{hp})_{j,i} &= \langle W \phi_i, \phi_j\rangle, \quad
(I_{hp})_{j,i} = \langle \phi_i, \psi_j\rangle
\end{align*}
and, therewith, the approximation of the Galerkin matrix is
\[
{P}_{hp}=W_{hp}+\left( K_{hp}+ \frac{1}{2}I_{hp}\right)^\top V^{-1}_{hp} \left( K_{hp}+ \frac{1}{2}I_{hp}\right).
\]
With the canonical embeddings
\begin{gather*}
i_{hp} : \mathcal{W}_{hp} \hookrightarrow \mathbf{H}^{-1/2}(\Gamma) \, , \qquad
j_{hp} : \mathcal{V}_{hp} \hookrightarrow \mathbf{H}^{1/2}(\Gamma)
\end{gather*}
and their duals $i_{hp}^*$ and $j_{hp}^*$,
the discrete Poincar\'{e}-Steklov operator $P_{hp} : \mathcal{V}_{hp} \to \mathcal{V}_{hp}^*$ can be also represented by
\[
P_{hp}= j_{hp}^*W j_{hp}+ j_{hp}^*\left(K'+\frac{1}{2}I\right )i_{hp} (i_{hp}^* V i_{hp})^{-1} i_{hp}^*\left(K+\frac{1}{2}I\right)j_{hp}.
\]
According to \cite{cc}, see e.g.~\cite{Banz2015Stab} for the $hp$-version, this functional is well-defined and satisfies
\begin{equation} \label{coerc}
\exists\, c >0: \quad \langle P_{hp} \mathbf{v}_{hp}, \mathbf{v}_{hp} \rangle_{\Gamma_0} \geq c \| \mathbf{v}_{hp}\|_{\tilde{\mathbf{H}}^{1/2}(\Gamma_0)} \qquad \forall\, \mathbf{v}_{hp} \in \mathcal{V}_{hp}
\end{equation}
if $\min\{h,p^{-1}\}$ is sufficiently small.
Further, we consider the operator
$E_{hp}: \mathbf{H}^{1/2}(\Gamma)\to \mathbf{H}^{-1/2}(\Gamma)$, reflecting the consistency error in the discretization of the Poincar\'{e}-Steklov operator $P$, defined by
\[
E_{hp}:=P-P_{hp}=\left(\frac{1}{2}I+K' \right)\left(V^{-1}-i_{hp}(i_{hp}^*Vi_{hp})^{-1}i_{hp}^*\right)\left(\frac{1}{2}I+K\right). \]
From \cite{MS-1} (see also \cite{cc}), the operator $E_{hp}$ is bounded and there exists a constant $c>0$ such that
\[
\|E_{hp}\mathbf{v}\|_{\mathbf{H}^{-1/2}(\Gamma)}\leq c \, \displaystyle \inf_{{\psi_{hp}}\in \mathcal{W}_{hp}}\|V^{-1}(I+K)\mathbf{v} - {\psi_{hp}}\|_{\mathbf{H}^{-1/2}(\Gamma)} \quad \forall \, \mathbf{v} \in \mathbf{H}^{1/2}(\Gamma).
\]
Now, we turn to the discretization of the regularized problem $(\mathcal{P}_\varepsilon)$. %To this end, we recall $\Pi : \mathbf{H}^{1/2} (\Gamma) \to L^2(\Gamma_C)$ with $ \Pi \mathbf{u}= \mathbf{u} \cdot \mathbf{n} \mbox{ on } \Gamma_C$.
%The mapping $\Pi$ is linear and continuous from $\mathbf{H}^{1/2} (\Gamma)$ into $L^2(\Gamma_C)$, i.e.
%\begin{equation} \label{contPi}
%\exists c>0 \, :\, \| \Pi \mathbf{v} \|_{L^2(\Gamma_C)} \leq c \| \mathbf{v}\|_{\mathbf{H}^{1/2} (\Gamma)} \quad \forall \, v \in \mathbf{H}^{1/2} (\Gamma) \quad \mbox{with} \quad c\leq 1.
%\end{equation}
%Further, we denote by $\tilde{\mathcal{V}}_h$ the image of $\mathcal{V}_h$ with respect to $\Pi$ defined by (\ref{Pi}), i.e.
%\[
%\tilde{\mathcal{V}}_h= \{ w_h \in C(\bar{\Gamma}_C) \, :\, w_h |_{E} \in P_1(E) \quad \forall \, E\in \mathcal{T}_h|_{\Gamma_C}, \; w_h=0 \; \mbox{on} \; \bar{\Gamma}_D\},
%\]
%where the symbol $\mathcal{T}_h|_{\Gamma_C}$ denotes the partition of $\Gamma_C$ induced by $\mathcal{T}_h$.
%\\
%We employ Gauss-Lobatto quadrature in the discretization procedure. To this end, for any quadrature order $p_T$ on $T\in \mathcal{T}_h$ we introduce on the reference element $\hat{T}$ the $p_T+1$ Gauss-Lobatto points $\xi_j^{p_T+1}$
%and approximate the nonlinear non-smooth functional $DJ_\varepsilon$ by $DJ_{\varepsilon,hp}$ defined by
%\begin{equation}\label{discret}
%\langle DJ_{\varepsilon, hp} \mathbf{u}_{hp},\mathbf{v}_{hp}\rangle_{\Gamma_C} = \sum_{T \in \mathcal{T}_h} \sum_{i=0}^{p_T} \omega_{i}^{p_T+1} S_x( \mathbf{u}_{hp,n}^\varepsilon \circ \Psi_T (\xi_{i}^{p_T+1}) , \varepsilon)\, (\mathbf{v}_{hp,n} \circ \Psi_T )(\xi_{i}^{p_T+1}).
%\end{equation}
The discretized regularized problem ($\mathcal{P}_{\varepsilon, hp}$) is: Find
$\mathbf{u}^{\varepsilon}_{hp} \in \mathcal{K}^{\Gamma}_{hp}$ such that for all $\forall \, \mathbf{v}_{hp} \in \mathcal{K}_{hp}^{\Gamma}$
\begin{equation} \label{reg_dis_problem}
\langle {P}_{hp} \mathbf{u}^{\varepsilon}_{hp}, \mathbf{v}_{hp}-\mathbf{u}^{\varepsilon}_{hp} \rangle_{\Gamma_0} +
\langle D J_{\varepsilon}(\mathbf{u}^{\varepsilon}_{hp}), \mathbf{v}_{hp} -\mathbf{u}^{\varepsilon}_{hp} \rangle_{\Gamma_C}
\geq \langle \mathbf{F}, \mathbf{v}_{hp}-\mathbf{u}^{\varepsilon}_{hp} \rangle_{\Gamma_0}. % \not\subseteq \mathcal{K} ^{\Gamma}
\end{equation}
The convergence result for the $hp$-solution $\mathbf{u}^\varepsilon_{hp}$ of ($\mathcal{P}_{\varepsilon, hp}$) is due to the following abstract approximation result from \cite{Gw_Ov}.
Let $\mathcal{K}$ be a closed, convex nonempty subset of a reflexive Banach space $X$ and $VI(\psi,f,\mathcal{K})$ be the pseudo-monotone variational inequality:~Find $u \in \mathcal{K}$ such that
$
\psi(u,v) \geq \langle f, v-u\rangle \, \forall \, v \in \mathcal{K}.
$
%and include from \cite{Gw_Ov} the following abstract approximation result: \\
%where $\mathcal{K}$ is a closed, convex nonempty subset of a reflexive Banach space $X$. \\
%We consider the pseudo-monotone variational inequality $VI(\psi,f,\mathcal{K})$ : Find $u \in \mathcal{K}$ such that
%\[
%\psi(u,v) \geq \langle f, v-u\rangle \quad \forall \, v \in \mathcal{K}.
%\]
Let $T$ be a directed set. We introduce the family $\{\mathcal{K}_t\}_{t\in T}$ of nonempty, closed and convex sets and admit the following hypotheses:
\begin{description}
\item (H1) If $\{v_{t'}\}_{t'\in T'}$ weakly converges to $v$ in $X$, $v_{t'} \in
\mathcal{K}_{t'} ~(t' \in T') $ for a subnet $\{\mathcal{K}_{t'}\}_{t'\in T'}$ of the net $\{\mathcal{K}_t\}_{t\in T}$, then $v\in \mathcal{K}$.
\item (H2) For any $v\in \mathcal{K}$ and any $t \in T$ there exists $v_{t} \in \mathcal{K}_{t}$
such that $v_{t} \to v$ in $X$.
\item (H3) $\psi_t$ is pseudomonotone for any $t \in T$.
\item (H4) $f_t \to f$ in $V^\ast$.
\item (H5) For any nets $\{u_t\}$ and $\{v_t\}$
such that $u_t \in \mathcal{K}_t$, $v_t\in \mathcal{K}_t$,
$u_t \rightharpoonup u$, and $v_t \to v$ in $X$ it follows that
$$
\displaystyle {\liminf_{t \in T}}\,~ \psi_t(u_t,v_t) \leq \psi(u,v) \,.
$$
\item (H6)
There exist constants $c>0$, $d$, $d_0 \in \R$ and $\alpha >1$ (independent of $t \in T$) such that for some $w_t \in \mathcal{K}_t$
with $w_t \to w$ there holds
$$
-\psi_t(u_t,w_t)\geq c \|u_t\|^\alpha_V +d \|u_t\|_V +d_0, \quad \forall \, u_t \in \mathcal{K}_t, \; \forall \, t \in T \,.
$$
\end{description}
\begin{theorem} \label{conv} Under the hypotheses $(H1)$-$(H6)$, there exists a solution to the problem $VI(\psi_t,f_t,\mathcal{K}_t)$. Moreover, the family $\{u_t\}$ of solutions to the problem $VI(\psi_t,f_t,\mathcal{K}_t)$ is bounded in $X$ and there exists a subnet of $\{u_t\}$ that converges weakly in $X$ to a solution of the problem $VI(\psi,f,\mathcal{K})$. %Furthermore, any weak accumulation point of $\{u_t\}$ is a solution to the problem $VI(\psi,f,\mathcal{K})$.
\end{theorem}
The hypotheses (H1) and (H2) are due to Glowinski \cite{Gl} and describe the Mosco convergence \cite{At} of the family $\mathcal{K}_t$ to $\mathcal{K}$, whereas $f_t$ in (H4) is a standard approximation of the
linear functional $f$, for example, by numerical integration. The hypotheses (H5)-(H6) have been verified in %the same way as is done in
\cite{Ovcharova2015}.
We apply Theorem \ref{conv} with $t=(\varepsilon, h, p)$ and show convergence for $\varepsilon \to 0^+$, and either $p\to \infty$ or $h\to 0$. We emphasize that in fact the weak convergence of $\{\mathbf{u}^\varepsilon_{hp}\}$ to $\mathbf{u}$ in $\mathbf{H}^{1/2}(\Gamma)$ can be replaced by the strong one. The complete proof of this result for the $h$-version of BEM can be found in \cite{Ovcharova2015}.
\section{A-priori error estimate for the regularized problem} \label{sec:apriori}
In this section we provide an a-priori error estimate for the $hp$-approximate solution $\mathbf{u}^{\varepsilon}_{hp}$ of the regularized boundary variational problem $(\mathcal{P}_\varepsilon)$ under the regularity assumptions of \cite{MS-1} for Signorini contact, i.e.~$\mathbf{u}_\varepsilon \in \mathbf{H}^{3/2}(\Gamma)$ and $g\in H^{3/2}(\Gamma_C)$. For our more general variational problem $(\mathcal{P}_\varepsilon)$ we arrive at the same convergence order of $ \mathcal{O}(h^{1/4}p^{-1/4})$ as Maischak and Stephan in \cite{MS-1}.
To this end, we present the following C\'{e}a-Falk approximation lemma for the regularized problem $(\mathcal{P}_\varepsilon)$. %Since the proof follows exactly the arguments of \cite{MS-1}, its proof is omitted.
\begin{lemma} Let $\mathbf{u}_\varepsilon \in \mathcal{K}^\Gamma$ be the solution of the problem $(\mathcal{P}_\varepsilon)$ and let $\mathbf{u}^{\varepsilon}_{hp} \in \mathcal{K}_{hp}^\Gamma$ be the solution of the problem $(\mathcal{P}_{\varepsilon, hp})$. Assume that $\alpha_0<c_p$, $\mathbf{u}_\varepsilon$ $\in \mathbf{H}^{3/2}(\Gamma)$, $g\in H^{3/2}(\Gamma_C)$ and $P \mathbf{u}_\varepsilon -\mathbf{F} \in \mathbf{L}^2(\Gamma)$. Then there exists a constant $c=c(\mathbf{u}_\varepsilon, g, \mathbf{f}, \mathbf{t})>0$, but independent of $h$ and $p$ such that
\begin{align*}
c&\|\mathbf{u}_\varepsilon-\mathbf{u}^\varepsilon_{ hp}\|^2_{\mathbf{H}^{1/2}(\Gamma)} \leq \|E_{hp}(\mathbf{u}_\varepsilon)\|^2_{\mathbf{H}^{-1/2}(\Gamma)} \\
& \quad +\inf _{v\in \mathcal{K}^\Gamma}
\{ \|P \mathbf{u}_\varepsilon-\mathbf{F}\|_{\mathbf{L}^2(\Gamma)} \|\mathbf{u}^{\varepsilon}_{ hp}-\mathbf{v}\|_{\mathbf{L}^2(\Gamma)} +
\langle D J_\varepsilon(\mathbf{u}_\varepsilon),\mathbf{v}-\mathbf{u}^\varepsilon_{hp} \rangle_{\Gamma_C} \} \\
&\quad + \inf _{v_{hp}\in \mathcal{K}_{hp}^\Gamma} \left\{ \|\mathbf{u}_\varepsilon- \mathbf{v}_{ hp}\|_{\mathbf{H}^{1/2}(\Gamma)}^2 + \|P \mathbf{u}_\varepsilon-\mathbf{F}\|_{\mathbf{L}^2(\Gamma)} \|\mathbf{u}_\varepsilon-\mathbf{v}_{ hp}\|_{\mathbf{L}^2(\Gamma)} \right.\\
&\quad \left.+
\langle D J_\varepsilon(\mathbf{u}^{\varepsilon}_{ hp}),\mathbf{v}_{hp}-\mathbf{u}_{\varepsilon} \rangle_{\Gamma_C}
\right\}.
\end{align*}
\end{lemma}
{\bf Proof} Using the definitions of $(\mathcal{P}_\varepsilon)$ and $(\mathcal{P}_{\varepsilon,hp})$, and estimates similar to \cite[Theorem 3]{MS-1}, we obtain for all $v\in \mathcal{K}^\Gamma$, $v_{hp}\in \mathcal{K}_{hp}^\Gamma$
\begin{align*}
c_P &\|\mathbf{u}_\varepsilon-\mathbf{u}^\varepsilon_{ hp}\|^2_{\mathbf{H}^{1/2}(\Gamma)} \leq \|E_{hp}(\mathbf{u}_\varepsilon)\|^2_{\mathbf{H}^{-1/2}(\Gamma)} +
\|\mathbf{u}_\varepsilon- \mathbf{v}_{ hp}\|_{\mathbf{H}^{1/2}(\Gamma)}^2 \\
& + \|P \mathbf{u}_\varepsilon-\mathbf{F}\|_{\mathbf{L}^2(\Gamma)} \left( \|\mathbf{u}_\varepsilon-\mathbf{v}_{ hp}\|_{\mathbf{L}^2(\Gamma)}+ \|\mathbf{u}^{\varepsilon}_{ hp}-\mathbf{v}\|_{\mathbf{L}^2(\Gamma)} \right ) + {\rm D},
\end{align*}
where we abbreviate
\[
{\rm D}= \langle D J_\varepsilon(\mathbf{u}_\varepsilon),\mathbf{v}-\mathbf{u}_\varepsilon \rangle_{\Gamma_C} + \langle D J_\varepsilon(\mathbf{u}^{\varepsilon}_{ hp}),\mathbf{v}_{hp}-\mathbf{u}^\varepsilon_{hp} \rangle_{\Gamma_C}.
\]
To bound the term ${\rm D}$, we use (\ref{ass1_th5}) and estimate as follows:
\begin{align*}
{\rm D} & = \langle D J_\varepsilon(\mathbf{u}_\varepsilon),\mathbf{v}-\mathbf{u}^\varepsilon_{hp} \rangle_{\Gamma_C} + \langle D J_\varepsilon(\mathbf{u}^{\varepsilon}_{ hp}), \mathbf{v}_{hp}-\mathbf{u}_{\varepsilon} \rangle_{\Gamma_C} \\
& \quad +
\langle D J_\varepsilon(\mathbf{u}_\varepsilon)- D J_\varepsilon(\mathbf{u}^{\varepsilon}_{ hp}) ,\mathbf{u}^\varepsilon_{hp} - \mathbf{u}_\varepsilon \rangle_{\Gamma_C} \\
& \leq \langle D J_\varepsilon(\mathbf{u}_\varepsilon),\mathbf{v}-\mathbf{u}^\varepsilon_{hp} \rangle_{\Gamma_C} + \langle D J_\varepsilon(\mathbf{u}^{\varepsilon}_{ hp}), \mathbf{v}_{hp}-\mathbf{u}_{\varepsilon} \rangle_{\Gamma_C} + \alpha_0 \| \mathbf{u}_{\varepsilon} - \mathbf{u}^\varepsilon_{ hp}\|^2_{\mathcal{V}}.
\end{align*}
Therefore, since $\alpha_0 <c_P$, see Theorem \ref{theouniq}, we obtain the assertion. \qed
\begin{theorem}
Let $\mathbf{u}_\varepsilon \in \mathcal{K}^\Gamma$ be the solution of the problem $(\mathcal{P}_\varepsilon)$ and let $\mathbf{u}^{\varepsilon}_ {hp} \in \mathcal{K}_{hp}^\Gamma$ be the solution of the problem $(\mathcal{P}_{\varepsilon, hp})$. Assume that $\alpha_0<c_p$, $\mathbf{u}_\varepsilon$ $\in \mathbf{H}^{3/2}(\Gamma)$, $g\in H^{3/2}(\Gamma_C)$ and $P \mathbf{u}_\varepsilon -\mathbf{F} \in \mathbf{L}^2(\Gamma)$. Then there exists a constant $c=c(\mathbf{u}_\varepsilon, g, \mathbf{f}, \mathbf{t})>0$, but independent of $h$ and $p$ such that
\begin{equation} \label{a-priori_est}
\|\mathbf{u}_\varepsilon-\mathbf{u}^{\varepsilon}_{hp}\|_{\mathbf{H}^{1/2}(\Gamma)} \leq c h^{1/4}p^{-1/4}.
\end{equation}
\end{theorem} %\cite{Gwinner_2013, MS-1}
{\bf Proof} %Since (\ref{a-priori_est}) is
Taking into account the estimates obtained by Maischak and Stephan in \cite[Theorem 3]{MS-1} for the consistency error, the approximation error, and for $\|E_{hp}{\mathbf{u}}\|_{\mathbf{H}^{-1/2}(\Gamma)}$, %without $DJ_\varepsilon$, %errors, without consideration of $DJ_\varepsilon$ is obtained in \cite{},
we only need to estimate
\begin{equation}\label{bound1}
\inf\{\langle D J_\varepsilon(\mathbf{u}_\varepsilon),\mathbf{v}-\mathbf{u}^\varepsilon_{hp} \rangle_{\Gamma_C} \, : \, \mathbf{v} \in \mathcal{K}^\Gamma \}
\end{equation}
and
\begin{equation}\label{bound2}
\inf \{\langle D J_\varepsilon(\mathbf{u}^{\varepsilon}_{ hp}),\mathbf{v}_{hp}-\mathbf{u}_{\varepsilon} \rangle_{\Gamma_C} \, : \, \mathbf{v}_{hp}\in \mathcal{K}_{hp}^\Gamma\}.
\end{equation}
%\[
%{\rm D}^*: =
%\langle D J_\varepsilon(\mathbf{u}_\varepsilon),\mathbf{v}-\mathbf{u}^\varepsilon_{hp} \rangle_{\Gamma_C} + \langle D J_\varepsilon(\mathbf{u}^{\varepsilon}_{ hp}),\mathbf{v}_{hp}-\mathbf{u}_{\varepsilon} \rangle_{\Gamma_C}
%\]
%for all $\mathbf{v} \in \mathcal{K}^\Gamma$ and for all $\mathbf{v}_{hp} \in \mathcal{K}^\Gamma_{hp}$.
To estimate (\ref{bound2}) we take $\mathbf{v}_{hp}:= I_{hp} \mathbf{u}_\varepsilon \in \mathcal{K}_{hp}^{\Gamma}$, the interpolate of $\mathbf{u}_\varepsilon\in \mathbf{H}^{3/2}(\Gamma)\subset C^0(\Gamma)$.
%Analogously to (\ref{bound3}), we have
From (\ref{ass_3}),
\begin{align}
\langle D J_\varepsilon(\mathbf{u}^{\varepsilon}_{ hp}),\mathbf{v}_{hp}-\mathbf{u}_{\varepsilon} \rangle_{\Gamma_C}
& = \int_{\Gamma_C} S_x(u^\varepsilon_{hp, n}, \varepsilon)(v_{hp, n}- u_{\varepsilon, n}) \, ds \nonumber\\
& \leq c \left(1+\|u^\varepsilon_{ hp, n}\|_{L^2(\Gamma_C)}\right) \| u_{\varepsilon, n}- v_{hp, n}\|_{L^2(\Gamma_C)}. \label{bound3}
\end{align}
By \cite[Theorems 4.2 and 4.5] {Bernardi} and by the real interpolation between $H^1(\Gamma)$ and $L^2(\Gamma)$ there exists a constant $C_1>0$ such that
\begin{equation} \label{bound4}
\|u_{\varepsilon, n} - v_{hp, n}\|_{H^{1/2}(\Gamma)} \leq C_1 h^1 p^{-1} \, \|\mathbf{u}_{\varepsilon}\|_{H^{3/2}(\Gamma)}.
\end{equation}
To estimate (\ref{bound1}) similarly as \cite{MS-1} we define
%$\mathbf{v}^*_{hp}:= I_{hp} \mathbf{u}_\varepsilon \in \mathcal{K}_{hp}^{\Gamma}$, the interpolate of $\mathbf{u}_\varepsilon\in \mathbf{H}^{3/2}(\Gamma)\subset C^0(\Gamma)$ and
$\mathbf{v}^* \in \mathcal{K}^\Gamma \cap \mathbf{H}^1(\Gamma)$ by
\[
\mathbf{v}^* :=\left\{ \begin{array}{lc}
\mathbf{u}^\varepsilon_{hp,t}+[g+ \inf \{u^\varepsilon_{hp,n}-g_{hp},0\}] \mathbf{n} \;& \mbox{on}\; \Gamma_C \\
0 \;& \mbox{on}\; \Gamma_D \\
\gamma_N \mathbf{u}^\varepsilon_{hp} \; & \mbox{on}\; \Gamma_N,
\end{array}
\right.
\]
%$\mathbf{v}^*=\mathbf{u}^\varepsilon_{hp,t}+\inf \{\mathbf{u}^\varepsilon_{hp,n}-g_{hp},0\}\mathbf{n} $ on $\Gamma_C$, $\mathbf{v}^*= 0$ on $\Gamma_D$ and $\mathbf{v}^*=\mathbf{u}^\varepsilon_{hp}$ on $\Gamma_N$,
where $g_{hp}:=I_{hp} g$ is the interpolate of the gap function $g$, and $\gamma _N$ is the trace map onto $\Gamma_N$.
%Since the bound (\ref{a-priori_est}) is obtained in \cite{MS-1} for both the consistency and the approximation error, %without consideration of $DJ_\varepsilon$, %errors, without consideration of $DJ_\varepsilon$ is obtained in \cite{},
%we only need to estimate
%\[
%{\rm D}^*: =\langle D J_\varepsilon(\mathbf{u}_\varepsilon),\mathbf{v}^*-\mathbf{u}^\varepsilon_{hp} \rangle_{\Gamma_C} + \langle D J_\varepsilon(\mathbf{u}^{\varepsilon}_{ hp}),\mathbf{v}^*_{hp}-\mathbf{u}_{\varepsilon} \rangle_{\Gamma_C}.
%\]
\\
Analogously to (\ref{bound3}), we have
\begin{align}
\langle D J_\varepsilon(\mathbf{u}_\varepsilon),\mathbf{v}^*-\mathbf{u}^\varepsilon_{hp} \rangle_{\Gamma_C} & = \int_{\Gamma_C} S_x(u_{\varepsilon, n}, \varepsilon)(v^*_n- u^\varepsilon_{hp, n}) \, ds\nonumber\\
& \leq c \left(1+\|u_{\varepsilon, n}\|_{L^2(\Gamma_C)}\right) \|v^*_n- u^\varepsilon_{hp, n}\|_{L^2(\Gamma_C)}. \label{bound5}
\end{align}
The elaborate analysis in \cite{MS-1}, see the proof of Theorem 3, estimates (31) - (33), gives
\begin{equation}\label{bound6}
\|v^*_n-u^\varepsilon_{hp,n}\|_{L^2(\Gamma_C)} \leq C_2 h^{1/2}p^{-1/2}\left( \|g\|_{H^{1/2}(\Gamma_C)}+ \|\mathbf{u}^{\varepsilon}_{hp}\|_{H^{1/2}(\Gamma_C)} \right).
\end{equation}
\\
%Due to $\alpha_0 <c_P$ (see Theorem \ref{theouniq}) and
Finally, combining the error estimates for the interpolation (\ref{bound4}) and the consistency (\ref{bound6}) with (\ref{bound3}) and (\ref{bound5}), respectively, and taking into account the boundedness of $\|\mathbf{u}^{\varepsilon}_{hp}\|$ in $H^{1/2}(\Gamma_C)$ (see Theorem \ref{conv}), we prove the wished bound for (\ref{bound1}) and (\ref{bound2}). \qed
\begin{rem} Let $\mathbf{u} \in \mathcal{K}^\Gamma$ be the solution of the problem $(\mathcal{P})$ and $\mathbf{u}_\varepsilon \in \mathcal{K}^\Gamma$ be the solution of the problem $(\mathcal{P}_\varepsilon)$.
Taking $\mathbf{v}=\mathbf{u}_\varepsilon$ in (\ref{BdHemVar}) and $\mathbf{v}=\mathbf{u}$ in (\ref{reg}), and adding the two resulting inequalities yields
\[
\langle P \mathbf{u}-P \mathbf{u}_\varepsilon, \mathbf{u}_\varepsilon- \mathbf{u} \rangle _{\Gamma_0}+ \varphi(\mathbf{u}, \mathbf{u}_\varepsilon) + \varphi_\varepsilon( \mathbf{u}_\varepsilon, \mathbf{u}) \geq 0.
\]
Hence,
\[
c_P\|\mathbf{u}_\varepsilon-\mathbf{u}\|^2_{\mathcal {V}}\leq \langle P \mathbf{u}_\varepsilon-P \mathbf{u}, \mathbf{u}_\varepsilon- \mathbf{u} \rangle _{\Gamma_0} \leq \varphi(\mathbf{u}, \mathbf{u}_\varepsilon) + \varphi_\varepsilon( \mathbf{u}_\varepsilon, \mathbf{u}).
\]
We emphasize that, in general, the functional $\varphi_\varepsilon$ does not approximate the functional $\varphi$ arbitrarily close as $\varepsilon \to 0^+$. Nevertheless, we have convergence of the sequence $\{\mathbf{u}_\varepsilon\}$ of the solutions of the regularized problem ($\mathcal{P}_\varepsilon$) to the solution $\mathbf{u}$ of the boundary hemivariational problem ($\mathcal{P}$). In particular, for any $\mathbf{v}_\varepsilon \rightharpoonup \mathbf{v}$ it holds
%the functionals $\varphi_\varepsilon$ and $\varphi$ satisfy the condition
\[
\limsup_{\varepsilon\to 0} \, \varphi_\varepsilon (\mathbf{v}_\varepsilon, \mathbf{v}) \leq \varphi(\mathbf{v}, \mathbf{v}) \quad \forall \mathbf{v} \in \mathcal{K}^\Gamma,
\]
which according to Theorem \ref{conv} is a sufficient condition for convergence. \\
However, for convergence rates in $\varepsilon$,
\[
\varphi(\mathbf{u}, \mathbf{u}_\varepsilon) + \varphi_\varepsilon( \mathbf{u}_\varepsilon, \mathbf{u}) \leq c \, \varepsilon^\beta \|\mathbf{u}_\varepsilon-\mathbf{u}\|_{\mathcal{V}}
\]
for some positive constants $c$ and $\beta$ is needed. An estimate of this form can be found in \cite{Ovcharova2012} for the approximation of the absolute value function.
\end{rem}
\section{A-posteriori error estimate for the regularized problem}
\label{sec:aposteriori}
To be able to split the approximation error into the discretization error of a simpler variational equation and contributions arising from the constraints on $\Gamma_C$ we introduce the regularized mixed formulation \eqref{eq:MixedVarEq}-\eqref{eq:mixedprobConstraint}, which is equivalent to the regularized problem $(\mathcal{P}_\varepsilon)$. \\
Find $(\mathbf{u}^\varepsilon,\lambda^\varepsilon) \in \mathcal{V} \times M(\mathbf{u}^\varepsilon)$ such that
\begin{subequations} \label{eq:mixedprob}
\begin{alignat}{2}
\left\langle P \mathbf{u}^\varepsilon, \mathbf{v} \right\rangle_{\Gamma_0}+ \left\langle \lambda^\varepsilon, v_n \right\rangle_{\Gamma_C} & = \langle \mathbf{F}, \mathbf{v} \rangle_{\Gamma_0} & \quad & \forall \, \mathbf{v} \in \mathcal{V} \label{eq:MixedVarEq}\\
\left\langle \mu -\lambda^\varepsilon, u^\varepsilon_n - g \right\rangle_{\Gamma_C} & \leq 0 & \quad & \forall \, \mu \in M(\mathbf{u}^\varepsilon) \label{eq:mixedprobConstraint}
\end{alignat}
\end{subequations}
with the set of admissible Lagrange multipliers
\begin{align*}
M(\mathbf{u}^\varepsilon):=\left\{ \mu \in X^*:\ \left\langle \mu,\eta \right\rangle_{\Gamma_C} \geq \left\langle DJ_\varepsilon(\mathbf{u}^\varepsilon),\eta \right\rangle_{\Gamma_C} \ \forall \, \eta \in X,\ \eta \geq 0 \text{ a.e.~on } \Gamma_C \right\}
\end{align*}
where $X=\{w \mid \exists\, \mathbf{v} \in \mathcal{V},\ v_n|_{\Gamma_C} = w\}\subset H^{1/2}(\Gamma_C)$ and $X^*$ its dual space.
\begin{lemma}
\begin{enumerate}
\item Let $\mathbf{u}^\varepsilon$ solve the regularized problem $(\mathcal{P}_\varepsilon)$, then there exists a $\lambda^\varepsilon \in M(\mathbf{u}^\varepsilon)$ such that $(\mathbf{u}^\varepsilon,\lambda^\varepsilon)$ solves \eqref{eq:mixedprob}. \\[0.01cm]
\item Let $(\mathbf{u}^\varepsilon,\lambda^\varepsilon)$ solve \eqref{eq:mixedprob}, then $\mathbf{u}^\varepsilon$ solves $(\mathcal{P}_\varepsilon)$.
\end{enumerate}
\end{lemma}
{\bf Proof}\,
\begin{enumerate}
\item Define $\lambda^\varepsilon \in X^*$ by
\begin{align} \label{eq1_mixed}
\left\langle \lambda^\varepsilon, v_n \right\rangle_{\Gamma_C} = \langle \mathbf{F}, \mathbf{v} \rangle_{\Gamma_0} - \left\langle P \mathbf{u}^\varepsilon, \mathbf{v} \right\rangle_{\Gamma_0} \quad \forall \, v \in \mathcal{V}.
\end{align}
Then, by \eqref{reg}, we obtain for all $\mathbf{v} \in \mathcal{K}^\Gamma$ with $v_n|_{\Gamma_C}=u^\varepsilon_n|_{\Gamma_C} -\eta$ for $0 \leq \eta \in X$
\begin{gather*}
\left\langle \lambda^\varepsilon, v_n -u^\varepsilon_n \right\rangle_{\Gamma_C} = \langle \mathbf{F}, \mathbf{v}-\mathbf{u}^\varepsilon \rangle_{\Gamma_0} - \left\langle P \mathbf{u}^\varepsilon, \mathbf{v} -\mathbf{u}^\varepsilon \right\rangle_{\Gamma_0} \leq \left\langle DJ_\varepsilon(\mathbf{u}^\varepsilon),v_n -u^\varepsilon_n \right\rangle_{\Gamma_C} \\ \Rightarrow \quad
\left\langle \lambda^\varepsilon, -\eta \right\rangle_{\Gamma_C} \leq \left\langle DJ_\varepsilon(\mathbf{u}^\varepsilon),-\eta \right\rangle_{\Gamma_C},
\end{gather*}
i.e.~$\lambda^\varepsilon \in M(\mathbf{u}^\varepsilon)$. Analogously, we obtain for $v_n|_{\Gamma_C}=g$, $v_n|_{\Gamma_C}=2u^\varepsilon_n|_{\Gamma_C} -g$
\begin{gather*}
\left\langle \lambda^\varepsilon, g -u^\varepsilon_n \right\rangle_{\Gamma_C} \leq \left\langle DJ_\varepsilon(\mathbf{u}^\varepsilon),g -u^\varepsilon_n \right\rangle_{\Gamma_C} \ \text{and} \
\left\langle \lambda^\varepsilon, u^\varepsilon_n -g \right\rangle_{\Gamma_C} \leq \left\langle DJ_\varepsilon(\mathbf{u}^\varepsilon),u^\varepsilon_n -g \right\rangle_{\Gamma_C},
\end{gather*}
which implies
\[
% \Rightarrow \quad
\left\langle \lambda^\varepsilon, u^\varepsilon_n -g \right\rangle_{\Gamma_C} = \left\langle DJ_\varepsilon(\mathbf{u}^\varepsilon),u^\varepsilon_n -g \right\rangle_{\Gamma_C}.
\]
It remains to show \eqref{eq:mixedprobConstraint}. From the previous equation we obtain for $\mu \in M(\mathbf{u}^\varepsilon)$
\begin{align*}
\left\langle \mu -\lambda^\varepsilon, u^\varepsilon_n - g \right\rangle_{\Gamma_C} &=
\left\langle \mu , u^\varepsilon_n - g \right\rangle_{\Gamma_C} - \left\langle DJ_\varepsilon(\mathbf{u}^\varepsilon),u^\varepsilon_n -g \right\rangle_{\Gamma_C} \\
&\leq \left\langle DJ_\varepsilon(\mathbf{u}^\varepsilon),u^\varepsilon_n -g \right\rangle_{\Gamma_C} - \left\langle DJ_\varepsilon(\mathbf{u}^\varepsilon),u^\varepsilon_n -g \right\rangle_{\Gamma_C} =0.
\end{align*}
\item There holds trivially $ DJ_\varepsilon(\mathbf{u}^\varepsilon)$, $2\lambda - DJ_\varepsilon(\mathbf{u}^\varepsilon) \in M(\mathbf{u}^\varepsilon)$, and thus \eqref{eq:mixedprobConstraint} yields
\begin{gather*}
\left\langle -\lambda^\varepsilon, u^\varepsilon_n -g \right\rangle_{\Gamma_C} \leq \left\langle -DJ_\varepsilon(\mathbf{u}^\varepsilon),u^\varepsilon_n -g \right\rangle_{\Gamma_C},
\ \left\langle \lambda^\varepsilon, g -u^\varepsilon_n \right\rangle_{\Gamma_C} \leq \left\langle DJ_\varepsilon(\mathbf{u}^\varepsilon),g -u^\varepsilon_n \right\rangle_{\Gamma_C}
\\ \Rightarrow \quad
\left\langle \lambda^\varepsilon, u^\varepsilon_n -g \right\rangle_{\Gamma_C} = \left\langle DJ_\varepsilon(\mathbf{u}^\varepsilon),u^\varepsilon_n -g \right\rangle_{\Gamma_C}.
\end{gather*}
For $\mathbf{v} \in \mathcal{K}^\Gamma$, i.e.~$v_n|_{\Gamma_C}=g -\eta$ with $0 \leq \eta \in X$, we therefore obtain
\begin{align*}
\left\langle \lambda^\varepsilon, v_n -u^\varepsilon_n \right\rangle_{\Gamma_C} &= \left\langle \lambda^\varepsilon, g -u^\varepsilon_n \right\rangle_{\Gamma_C} +
\left\langle \lambda^\varepsilon, -\eta \right\rangle_{\Gamma_C} = \left\langle DJ_\varepsilon(\mathbf{u}^\varepsilon), g -u^\varepsilon_n \right\rangle_{\Gamma_C} + \left\langle \lambda^\varepsilon, -\eta \right\rangle_{\Gamma_C} \\
&\leq \left\langle DJ_\varepsilon(\mathbf{u}^\varepsilon), g -u^\varepsilon_n \right\rangle_{\Gamma_C} + \left\langle DJ_\varepsilon(\mathbf{u}^\varepsilon), -\eta \right\rangle_{\Gamma_C} = \left\langle DJ_\varepsilon(\mathbf{u}^\varepsilon), v_n -u^\varepsilon_n \right\rangle_{\Gamma_C}.
\end{align*}
It remains to show that $u^\varepsilon_n -g \leq 0$. Assume there exists a measurable set in which $u^\varepsilon_n - g>0$. Then, from \eqref{eq:mixedprobConstraint} we obtain for $\mu=DJ_\varepsilon(\mathbf{u}^\varepsilon) + \chi_{u^\varepsilon_n - g >0} \in M(\mathbf{u}^\varepsilon)$ with indicator function $\chi$
\begin{align*}
0 \geq \left\langle \mu -\lambda^\varepsilon, u^\varepsilon_n - g \right\rangle_{\Gamma_C}
=\left\langle \mu -DJ_\varepsilon(\mathbf{u}^\varepsilon), u^\varepsilon_n - g \right\rangle_{\Gamma_C}
=\left\langle \chi_{u^\varepsilon_n - g >0}, u^\varepsilon_n - g \right\rangle_{\Gamma_C} >0,
\end{align*}
i.e.~$\mathbf{u}^\varepsilon \in \mathcal{K}^\Gamma$.
\qed
\end{enumerate}
Given the discrete solution $\mathbf{u}_{hp}^\varepsilon \in \mathcal{K}_{hp}^\Gamma$ to $(\mathcal{P}_{\varepsilon,hp})$, we reconstruct $\lambda_{hp}^\varepsilon \in \operatorname{span}\left\{ \psi_i \right\}_{i=1}^M $ such that
\begin{align} \label{eq:reconstruct_lambda_h}
\left\langle \lambda_{hp}^\varepsilon, v_n \right\rangle_{\Gamma_C} = \langle \mathbf{F}, \mathbf{v} \rangle_{\Gamma_0} - \left\langle P_{hp} \mathbf{u}_{hp}^\varepsilon, \mathbf{v} \right\rangle_{\Gamma_0} \quad \forall \, v \in \mathcal{V}_{hp}
\end{align}
by solving a potentially over-constrained system of linear equations for an arbitrary choice of basis $\left\{ \psi \right\}$.
Following the Braess trick \cite{braess2005posteriori} as e.g.~in \cite{Banz2015}, we define the auxiliary problem
\begin{align} \label{eq:auxProb}
\mathbf{z} \in \mathcal{V} : \quad \left\langle P \mathbf{z}, \mathbf{v} \right\rangle_{\Gamma_0} = \langle \mathbf{F}, \mathbf{v} \rangle_{\Gamma_0} - \left\langle \lambda_{hp}^\varepsilon, v_n \right\rangle_{\Gamma_C} \quad \forall \, \mathbf{v} \in \mathcal{V}.
\end{align}
Subtracting \eqref{eq:MixedVarEq} and \eqref{eq:auxProb} yields
\begin{align} \label{eq:exchange_u_to_z}
\left\langle P(\mathbf{u}^\varepsilon -\mathbf{z}), \mathbf{v} \right\rangle_{\Gamma_0} = \left\langle \lambda_{hp}^\varepsilon - \lambda^\varepsilon, v_n \right\rangle_{\Gamma_C} \quad \forall \, \mathbf{v} \in \mathcal{V}
\end{align}
and additionally with the continuous inf-sup condition \cite[Theorem 3.2.1]{Chernov} this yields (see \cite{Banz2015})
\begin{align} \label{eq:apostLambda}
\left\| \lambda_{hp}^\varepsilon - \lambda^\varepsilon \right\|_{X^*} \leq \frac{C}{\beta} \left\| \mathbf{u}^\varepsilon - \mathbf{z} \right\|_{\mathcal{V}} \leq \frac{C}{\beta} \left\| \mathbf{u}^\varepsilon - \mathbf{u}_{hp}^\varepsilon \right\|_{\mathcal{V}} + \frac{C}{\beta} \left\| \mathbf{u}_{hp}^\varepsilon - \mathbf{z} \right\|_{\mathcal{V}}
\end{align}
with inf-sup constant $\beta>0$. See \cite[Theorem 3.2.1]{Chernov} for a proof of the inf-sup condition for the difficult case when $\bar{\Gamma}_C \cap \bar{\Gamma}_D = \emptyset$, i.e.~$X^*=\tilde{H}^{-1/2}(\Gamma_C)$.
% \begin{lemma} \label{lemma:APostDelamLaw}
% Let $S_x(\cdot, \varepsilon)$ be Lipschitz continuous, then there exists a constant $C>0
% $ independent of $h$ and $p$ such that
% $$ \left\langle DJ_{\varepsilon}(\mathbf{u}^\varepsilon_{hp})- DJ_\varepsilon(\mathbf{u}^\varepsilon), u^\varepsilon_n -u_{hp,n}^\varepsilon \right\rangle_{\Gamma_C} \leq C \|\mathbf{u}_{hp,n}^\varepsilon - \mathbf{u}_{n}^\varepsilon \|_{L^2(\Gamma_C)}^2
% $$
% {\color{red} Dies ist gezeigt schon in (\ref{ass1_th5})!} \\
% and
% $$\left\langle DJ_{\varepsilon}(\mathbf{u}_{hp}^\varepsilon) - DJ_\varepsilon(\mathbf{u}^\varepsilon) , (u_{hp,n}^\varepsilon -g)^+ \right\rangle_{\Gamma_C} \leq \frac{C}{2} \|\mathbf{u}_{hp,n}^\varepsilon - \mathbf{u}_{n}^\varepsilon \|_{L^2(\Gamma_C)}^2 + \frac{C}{2} \|(u_{hp,n}^\varepsilon -g)^+ \|_{L^2(\Gamma_C)}^2.
% $$
% {\bf Proof} \,
% From the Lipschitz continuity follows
% \begin{align*}
% \left\langle DJ_\varepsilon(\mathbf{u}_{hp}^\varepsilon) - DJ_\varepsilon(\mathbf{u}^\varepsilon), v_n \right\rangle_{\Gamma_C} = \int_{\Gamma_C} \left( S_x(\mathbf{u}_{hp,n}^\varepsilon) - S_x(\mathbf{u}_{n}^\varepsilon) \right) v_n \, ds \leq Lip \|\mathbf{u}_{hp,n}^\varepsilon - \mathbf{u}_{n}^\varepsilon \|_{L^2(\Gamma_C)} \|\mathbf{v}_{n} \|_{L^2(\Gamma_C)}.
% \end{align*}
% The assertion follows from Young's inequality. \qed
% \end{lemma}
% Since the constant $C$ in the previous lemma maybe very large we need an $h/p$-scaling when going from the $L^2$-norm to the $\tilde{H}^{1/2}(\Gamma_0)$-norm.
%
% \begin{theorem} \label{thm:L2Aposteriori}
% There exists a constant $C$ independent of $h$ and $p$ such that
% \begin{align}
% \left\| \mathbf{u}^\varepsilon - \mathbf{u}^\varepsilon_{hp} \right\|_{L^2(\Gamma_0)} \leq C \left( \left(\frac{h}{p}\right)^{1/2} \left\| \mathbf{u}^\varepsilon - \mathbf{u}^\varepsilon_{hp} \right\|_{\tilde{H}^{1/2}(\Gamma_0)} + \beta^{-1} \left( \left\| \mathbf{z} - \mathbf{u}^\varepsilon_{hp} \right\|_{\tilde{H}^{1/2}(\Gamma_0)} + \left\| \mathbf{\Psi} - \mathbf{\Psi}_{hp} \right\|_{H^{-1/2}(\Gamma)} \right) \right)
% \end{align}
% with $\mathbf{\Psi}:= V^{-1}(K+1/2)\mathbf{z}$ and $\mathbf{\Psi}_{hp}:= i_{hp}V_{hp}^{-1}i_{hp}^* (K+1/2)\mathbf{u}_{hp}$.
% \\
% {\bf Proof} \,
% Recall
% \begin{alignat*}{2}
% &\mathbf{u}^\varepsilon \in \mathcal{V}: \qquad & \left\langle P \mathbf{u}^\varepsilon, \mathbf{v} \right\rangle_{\Gamma_0}+ \left\langle \lambda^\varepsilon, v_n \right\rangle_{\Gamma_C} & = \langle \mathbf{F}, \mathbf{v} \rangle_{\Gamma_0} \quad \forall \, \mathbf{v} \in \mathcal{V}\\
% &\mathbf{u}_{hp}^\varepsilon \in \mathcal{V}_{hp}: \qquad & \left\langle P_{hp} \mathbf{u}_{hp}^\varepsilon, \mathbf{v} \right\rangle_{\Gamma_0}+ \left\langle \lambda_{hp}^\varepsilon, v_n \right\rangle_{\Gamma_C} &= \langle \mathbf{F}, \mathbf{v} \rangle_{\Gamma_0} \quad \forall \, \mathbf{v} \in \mathcal{V}_{hp} \subset \mathcal{V}.
% \end{alignat*}
% Hence, with $E_{hp}=P-P_{hp}$ there holds
% $$
% \left\langle P \mathbf{u}^\varepsilon, \mathbf{v} \right\rangle_{\Gamma_0}-\left\langle P_{hp} \mathbf{u}_{hp}^\varepsilon, \mathbf{v} \right\rangle_{\Gamma_0} + \left\langle \lambda^\varepsilon - \lambda_{hp}^\varepsilon, v_n \right\rangle_{\Gamma_C} = \left\langle P ( \mathbf{u}^\varepsilon-\mathbf{u}_{hp}^\varepsilon), \mathbf{v} \right\rangle_{\Gamma_0}+\left\langle E_{hp} \mathbf{u}_{hp}^\varepsilon, \mathbf{v} \right\rangle_{\Gamma_0} + \left\langle \lambda^\varepsilon - \lambda_{hp}^\varepsilon, v_n \right\rangle_{\Gamma_C} = 0 \quad \forall \, \mathbf{v} \in \mathcal{V}_{hp}
% $$
% Following the Aubin-Nitsche trick, we set
% {\color{red}
% \begin{align} \label{eq:AubinNitscheProblem}
% \mathbf{\Phi} \in H^1(\Gamma) \cap \tilde{H}^{1/2}(\Gamma_0): \quad
% \left\langle P^\top \mathbf{\Phi}, \mathbf{v} \right\rangle_{\Gamma_0} = \left\langle \mathbf{u}^\varepsilon- \mathbf{u}_{hp}^\varepsilon, \mathbf{v} \right\rangle_{\Gamma_0} \quad \forall \, \mathbf{v} \in L^2(\Gamma_0).
% \end{align} }
% Then we obtain $(\mathbf{e}:=\mathbf{u}^\varepsilon- \mathbf{u}_{hp}^\varepsilon, \mathbf{\Phi}_{hp} := I_{hp}\mathbf{\Phi} \in \mathcal{V}_{hp})$
% \begin{align*}
% \left\|\mathbf{e}\right\|_{L^2(\Gamma_0)}^2 = \langle \mathbf{e},\mathbf{e}\rangle_{\Gamma_0}= \left\langle P^\top \mathbf{\Phi}, \mathbf{e} \right\rangle_{\Gamma_0} =\left\langle P \mathbf{e}, \mathbf{\Phi} \right\rangle_{\Gamma_0} = \left\langle P \mathbf{e}, \mathbf{\Phi} - \mathbf{\Phi}_{hp} \right\rangle_{\Gamma_0} -\left\langle E_{hp} \mathbf{u}_{hp}^\varepsilon, \mathbf{\Phi}_{hp} \right\rangle_{\Gamma_0} - \left\langle \lambda^\varepsilon - \lambda_{hp}^\varepsilon, \mathbf{\Phi}_{hp,n} \right\rangle_{\Gamma_C}.
% \end{align*}
% The three terms are estimated individually. \\
% {\color{red} From polynomial interpolation error estimate in \cite{Bernardi97}, and the continuity property of the boundary linear operators, it follows that
% \begin{align*}
% \left\langle P e, \phi - \phi_{hp} \right\rangle_{\Gamma_0} \leq C_P \|\mathbf{e}\|_{ \tilde{H}^{1/2}(\Gamma_0)} \|\phi - \phi_{hp}\|_{ \tilde{H}^{1/2}(\Gamma_0)} \leq C \left(\frac{h}{p}\right)^{1/2} \|\mathbf{e}\|_{ \tilde{H}^{1/2}(\Gamma_0)} \|\phi\|_{H^{1}(\Gamma)} = C \left(\frac{h}{p}\right)^{1/2} \|\mathbf{e}\|_{ \tilde{H}^{1/2}(\Gamma_0)} \|(P^\top)^{-1} \mathbf{e}\|_{H^{1}(\Gamma)} \\
% \leq C \left(\frac{h}{p}\right)^{1/2} \|\mathbf{e}\|_{ \tilde{H}^{1/2}(\Gamma_0)} \|\mathbf{e}\|_{L^2(\Gamma)}.
% \end{align*} }
% Define $\mathbf{\Psi}_{hp}^*=V^{-1}(K+1/2)\mathbf{u}_{hp}^\varepsilon$, then we obtain
% \begin{align*}
% -\left\langle E_{hp} \mathbf{u}_{hp}^\varepsilon, \mathbf{\Phi}_{hp} \right\rangle_{\Gamma_0} &= -\left\langle \left(V^{-1} - i_{hp}V_{hp}^{-1}i_{hp}^*\right) (K+1/2) \mathbf{u}_{hp}^\varepsilon, (K+1/2) \mathbf{\Phi}_{hp} \right\rangle_{\Gamma} = -\left\langle \mathbf{\Psi}^*_{hp}-\mathbf{\Psi}+\mathbf{\Psi} - \mathbf{\Psi}_{hp}, (K+1/2) \mathbf{\Phi}_{hp} \right\rangle_{\Gamma} \\
% &= \left\langle {\color{red}{(P-W)}}(\mathbf{z}-\mathbf{u}_{hp}^\varepsilon), \mathbf{\Phi}_{hp} \right\rangle_{\Gamma_0} -\left\langle \mathbf{\Psi} - \mathbf{\Psi}_{hp}, (K+1/2) \mathbf{\Phi}_{hp} \right\rangle_{\Gamma}\\ & \leq C \left( \|\mathbf{z}-\mathbf{u}_{hp}^\varepsilon \|_{\tilde{H}^{1/2}(\Gamma_0)} +\| \mathbf{\Psi}-\mathbf{\Psi}_{hp}\|_{H^{-1/2}(\Gamma)} \right) \|\mathbf{\Phi}_{hp} \|_{\tilde{H}^{1/2}(\Gamma_0)}.
% \end{align*}
% From \eqref{eq:apostLambda} we obtain
% \begin{align*}
% - \left\langle \lambda^\varepsilon - \lambda_{hp}^\varepsilon, \mathbf{\Phi}_{hp,n} \right\rangle_{\Gamma_C} \leq \|\lambda^\varepsilon - \lambda_{hp}^\varepsilon\|_{X^*} \|\mathbf{\Phi}_{hp,n}\|_X \leq C \beta^{-1} \left\| \mathbf{u}_{hp}^\varepsilon - \mathbf{z} \right\|_{\tilde{\mathbf{H}}^{1/2}(\Gamma_0)} \|\mathbf{\Phi}_{hp}\|_{ \tilde{H}^{1/2}(\Gamma_0)}.
% \end{align*}
% Since $\mathbf{\Phi} \in H^1(??) \subset L^2 (??)$ we obtain from \eqref{eq:AubinNitscheProblem}
% $$ c\| \mathbf{\Phi} \|^2_{\tilde{H}^{1/2}(\Gamma_0)} \leq \left\langle P^\top \mathbf{\Phi}, \mathbf{\Phi} \right\rangle_{\Gamma_0} = \left\langle \mathbf{e}, \mathbf{\Phi} \right\rangle_{\Gamma_0} \leq \|\mathbf{e}\|_{H^{-1/2}(\Gamma_0)} \| \mathbf{\Phi} \|_{\tilde{H}^{1/2}(\Gamma_0)} $$
% and with the $H^{1/2}$-stability of the interpolation
% $$ \|\mathbf{\Phi}_{hp}\|_{ \tilde{H}^{1/2}(\Gamma_0)} \leq C \|\mathbf{\Phi}\|_{ \tilde{H}^{1/2}(\Gamma_0)} \leq C \|\mathbf{e}\|_{H^{-1/2}(\Gamma_0)} \leq C \|\mathbf{e}\|_{L^2(\Gamma_0)}.$$
% Putting everything together and dividing by $\|\mathbf{e}\|_{L^2(\Gamma_0)}$ yields the assertion.
% \qed
% \end{theorem}
\begin{theorem} \label{thm:aposteriori}
Under the assumption \eqref{uniqreg} and if $S_x(\cdot, \varepsilon)$ is Lipschitz continuous, then there exists a constant $C$ independent of $h$ and $p$ such that for $\varsigma>0$ arbitrary
\begin{align*}
\left(c_P - \alpha_0 -4\varsigma \right)\|\mathbf{u}^\varepsilon - \mathbf{u}_{hp}^\varepsilon \|_{\mathcal{V}}^2 \leq & \left(\frac{C}{\varsigma} + 1 \right) \|\mathbf{z} - \mathbf{u}_{hp}^\varepsilon \|^2_{\mathcal{V}} + \frac{1}{4\varsigma}\left\| ( \lambda_{hp}^\varepsilon -DJ_{\varepsilon}(\mathbf{u}_{hp}^\varepsilon) )^- \right\|_{ X^*}^2
\\
&+ C\left(\frac{1}{\varsigma} + \frac{1}{\beta^2}+\frac{1}{\varsigma \beta^2}\right) \left\|(u_{hp,n}^\varepsilon -g)^+ \right\|^2_{ X} \\
& - \left\langle (\lambda_{hp}^\varepsilon - DJ_{\varepsilon}(\mathbf{u}_{hp}^\varepsilon))^+ , (u_{hp,n}^\varepsilon -g)^- \right\rangle_{\Gamma_C}
\end{align*}
with $\lambda_{hp}^\varepsilon$ satisfying \eqref{eq:reconstruct_lambda_h}.
\end{theorem}
{\bf Proof}\,
From the Lipschitz continuity it follows that
\begin{align*}
\left\langle DJ_\varepsilon(\mathbf{u}_{hp}^\varepsilon) - DJ_\varepsilon(\mathbf{u}^\varepsilon), v_n \right\rangle_{\Gamma_C} &= \int_{\Gamma_C} \left( S_x({u}_{hp,n}^\varepsilon) - S_x({u}_{n}^\varepsilon) \right) v_n \, ds\\
&\leq Lip \|{u}_{hp,n}^\varepsilon - {u}_{n}^\varepsilon \|_{L^2(\Gamma_C)} \|{v}_{n} \|_{L^2(\Gamma_C)}
\end{align*}
and thus, by Young's inequality ($\varsigma>0$ arbitrary),
\begin{align}
\left\langle DJ_{\varepsilon}(\mathbf{u}_{hp}^\varepsilon) - DJ_\varepsilon(\mathbf{u}^\varepsilon) , (u_{hp,n}^\varepsilon -g)^+ \right\rangle_{\Gamma_C} & \leq \varsigma \|\mathbf{u}_{hp}^\varepsilon - \mathbf{u}^\varepsilon \|_{L^2(\Gamma_C)}^2 \nonumber \\
& \quad + \frac{C^2}{4\varsigma} \|(u_{hp,n}^\varepsilon -g)^+ \|_{L^2(\Gamma_C)}^2. \label{eq:apostHelp}
\end{align}
From the conformity $\mathcal{V}_{hp} \subset \mathcal{V}$ we obtain with \eqref{eq:exchange_u_to_z}
\begin{align*}
c_P \|\mathbf{u}^\varepsilon - \mathbf{u}_{hp}^\varepsilon \|_{\mathcal{V}}^2 &\leq \left\langle P(\mathbf{u}^\varepsilon -\mathbf{u}_{hp}^\varepsilon), \mathbf{u}^\varepsilon -\mathbf{u}_{hp}^\varepsilon \right\rangle_{\Gamma_0}\\
&= \left\langle P(\mathbf{z}-\mathbf{u}_{hp}^\varepsilon), \mathbf{u}^\varepsilon -\mathbf{u}_{hp}^\varepsilon \right\rangle_{\Gamma_0} + \left\langle \lambda_{hp}^\varepsilon - \lambda^\varepsilon, u^\varepsilon_n -u_{hp,n}^\varepsilon \right\rangle_{\Gamma_C} \\
&\leq C_P \|\mathbf{z} - \mathbf{u}_{hp}^\varepsilon \|_{\mathcal{V}} \|\mathbf{u}^\varepsilon - \mathbf{u}_{hp}^\varepsilon \|_{\mathcal{V}} + \left\langle \lambda_{hp}^\varepsilon - \lambda^\varepsilon, u^\varepsilon_n -u_{hp,n}^\varepsilon \right\rangle_{\Gamma_C}.
\end{align*}
From complementarity $ \left\langle \lambda^\varepsilon, u^\varepsilon_n -g \right\rangle_{\Gamma_C} = \left\langle DJ_\varepsilon(\mathbf{u}^\varepsilon),u^\varepsilon_n -g \right\rangle_{\Gamma_C}$, $\lambda^\varepsilon \in M(\mathbf{u}^\varepsilon)$, $v=v^+ + v^- = \max\{0,v\} + \min\{0,v\}$, $u^\varepsilon_n - g \leq 0$ a.e.~on $\Gamma_C$ we obtain
\begin{align*}
& \left\langle \lambda_{hp}^\varepsilon - \lambda^\varepsilon, u^\varepsilon_n -u_{hp,n}^\varepsilon \right\rangle_{\Gamma_C} =
\left\langle \lambda_{hp}^\varepsilon - DJ_{\varepsilon}(\mathbf{u}^\varepsilon_{hp}), u^\varepsilon_n -u_{hp,n}^\varepsilon \right\rangle_{\Gamma_C}
\\ &\quad + \left\langle DJ_{\varepsilon}(\mathbf{u}^\varepsilon_{hp})- \lambda^\varepsilon, u^\varepsilon_n -g \right\rangle_{\Gamma_C}
+ \left\langle DJ_{\varepsilon}(\mathbf{u}^\varepsilon_{hp})- \lambda^\varepsilon, g -u_{hp,n}^\varepsilon \right\rangle_{\Gamma_C} \\
& =
\left\langle \lambda_{hp}^\varepsilon - DJ_{\varepsilon}(\mathbf{u}^\varepsilon_{hp}), u^\varepsilon_n -u_{hp,n}^\varepsilon \right\rangle_{\Gamma_C}
+ \left\langle DJ_{\varepsilon}(\mathbf{u}^\varepsilon_{hp})- DJ_\varepsilon(\mathbf{u}^\varepsilon), u^\varepsilon_n -g \right\rangle_{\Gamma_C}
\\ &\quad+ \left\langle DJ_{\varepsilon}(\mathbf{u}^\varepsilon_{hp})- \lambda^\varepsilon, g -u_{hp,n}^\varepsilon \right\rangle_{\Gamma_C}\\
& =
\left\langle \lambda_{hp}^\varepsilon - DJ_{\varepsilon}(\mathbf{u}^\varepsilon_{hp}), u^\varepsilon_n -u_{hp,n}^\varepsilon \right\rangle_{\Gamma_C}
+ \left\langle DJ_{\varepsilon}(\mathbf{u}^\varepsilon_{hp})- DJ_\varepsilon(\mathbf{u}^\varepsilon), u^\varepsilon_n -u_{hp,n}^\varepsilon \right\rangle_{\Gamma_C}
\\ &\quad+ \left\langle \lambda^\varepsilon - DJ_\varepsilon(\mathbf{u}^\varepsilon) , u_{hp,n}^\varepsilon -g \right\rangle_{\Gamma_C}\\
& \leq \left\langle (\lambda_{hp}^\varepsilon - DJ_{\varepsilon}(\mathbf{u}^\varepsilon_{hp}))^+, u^\varepsilon_n -g+g -u_{hp,n}^\varepsilon \right\rangle_{\Gamma_C} + \left\langle (\lambda_{hp}^\varepsilon - DJ_{\varepsilon}(\mathbf{u}^\varepsilon_{hp}))^-, u^\varepsilon_n -u_{hp,n}^\varepsilon \right\rangle_{\Gamma_C}\\
&\quad + \left\langle DJ_{\varepsilon}(\mathbf{u}^\varepsilon_{hp})- DJ_\varepsilon(\mathbf{u}^\varepsilon), u^\varepsilon_n -u_{hp,n}^\varepsilon \right\rangle_{\Gamma_C}
+ \left\langle \lambda^\varepsilon - DJ_\varepsilon(\mathbf{u}^\varepsilon) , (u_{hp,n}^\varepsilon -g)^+ \right\rangle_{\Gamma_C}
\\
& \leq \left\langle (\lambda_{hp}^\varepsilon - DJ_{\varepsilon}(\mathbf{u}^\varepsilon_{hp}))^+, g -u_{hp,n}^\varepsilon \right\rangle_{\Gamma_C} + \left\langle \lambda^\varepsilon - DJ_\varepsilon(\mathbf{u}^\varepsilon) , (u_{hp,n}^\varepsilon -g)^+ \right\rangle_{\Gamma_C}\\
&\quad + \left\langle DJ_{\varepsilon}(\mathbf{u}^\varepsilon_{hp})- DJ_\varepsilon(\mathbf{u}^\varepsilon), u^\varepsilon_n -u_{hp,n}^\varepsilon \right\rangle_{\Gamma_C} + \left\langle (\lambda_{hp}^\varepsilon - DJ_{\varepsilon}(\mathbf{u}^\varepsilon_{hp}))^-, u^\varepsilon_n -u_{hp,n}^\varepsilon \right\rangle_{\Gamma_C}.
\end{align*}
Since
\begin{align*}
&\left\langle (\lambda_{hp}^\varepsilon - DJ_{\varepsilon}(\mathbf{u}^\varepsilon_{hp}))^+, g -u_{hp,n}^\varepsilon \right\rangle_{\Gamma_C}+ \left\langle \lambda^\varepsilon - DJ_\varepsilon(\mathbf{u}^\varepsilon) , (u_{hp,n}^\varepsilon -g)^+ \right\rangle_{\Gamma_C} \\
&=
- \left\langle (\lambda_{hp}^\varepsilon - DJ_{\varepsilon}(\mathbf{u}^\varepsilon_{hp}))^+, (u_{hp,n}^\varepsilon -g)^+ +(u_{hp,n}^\varepsilon -g)^- \right\rangle_{\Gamma_C}
\\& \quad + \left\langle \lambda^\varepsilon - DJ_\varepsilon(\mathbf{u}^\varepsilon) - (\lambda_{hp}^\varepsilon - DJ_{\varepsilon}(\mathbf{u}_{hp}^\varepsilon)) + (\lambda_{hp}^\varepsilon - DJ_{\varepsilon}(\mathbf{u}_{hp}^\varepsilon))^+ , (u_{hp,n}^\varepsilon -g)^+ \right\rangle_{\Gamma_C} \\
& \quad + \left\langle (\lambda_{hp}^\varepsilon - DJ_{\varepsilon}(\mathbf{u}_{hp}^\varepsilon))^-, (u_{hp,n}^\varepsilon -g)^+ \right\rangle_{\Gamma_C} \\
&\leq \left\langle \lambda^\varepsilon - \lambda_{hp}^\varepsilon , (u_{hp,n}^\varepsilon -g)^+ \right\rangle_{\Gamma_C} + \left\langle DJ_{\varepsilon}(\mathbf{u}_{hp}^\varepsilon) - DJ_\varepsilon(\mathbf{u}^\varepsilon) , (u_{hp,n}^\varepsilon -g)^+ \right\rangle_{\Gamma_C} \\
& \quad - \left\langle (\lambda_{hp}^\varepsilon - DJ_{\varepsilon}(\mathbf{u}_{hp}^\varepsilon))^+ , (u_{hp,n}^\varepsilon -g)^- \right\rangle_{\Gamma_C},
\end{align*}
application of the Cauchy-Schwarz inequality, Young's inequality with $\varsigma >0$, \eqref{eq:apostLambda}, \eqref{eq:apostHelp} and \eqref{ass1_th5} yields
\begin{align*}
\left(c_P - \alpha_0 -4\varsigma \right)\|\mathbf{u}^\varepsilon - \mathbf{u}_{hp}^\varepsilon \|_{\mathcal{V}}^2 \leq & \left(\frac{C}{\varsigma} + 1 \right) \|\mathbf{z} - \mathbf{u}_{hp}^\varepsilon \|^2_{\mathcal{V}} + \frac{1}{4\varsigma}\left\| ( \lambda_{hp}^\varepsilon -DJ_{\varepsilon}(\mathbf{u}_{hp}^\varepsilon) )^- \right\|_{ X^*}^2
\\
&+ C\left(\frac{1}{\varsigma} + \frac{1}{\beta^2}+\frac{1}{\varsigma \beta^2}\right) \left\|(u_{hp,n}^\varepsilon -g)^+ \right\|^2_{ X} \\
& - \left\langle (\lambda_{hp}^\varepsilon - DJ_{\varepsilon}(\mathbf{u}_{hp}^\varepsilon))^+ , (u_{hp,n}^\varepsilon -g)^- \right\rangle_{\Gamma_C}. \qquad \qed
\end{align*}
The a-posteriori error estimate decomposes into the discretization error of a variational equality $\|\mathbf{z} - \mathbf{u}_{hp}^\varepsilon \|^2_{\mathcal{V}}$, which can be further estimated by e.g.~residual error estimates \cite{cc96} or bubble error estimates, e.g.~\cite{Banz2015}, and violation of the consistency condition $\left\| ( \lambda_{hp}^\varepsilon -DJ_{\varepsilon}(\mathbf{u}_{hp}^\varepsilon) )^- \right\|_{ X^*}^2$, violation of the non-penetration condition $ \left\|(u_{hp,n}^\varepsilon -g)^+ \right\|^2_{ X}$ and violation of the complementarity condition $- \left\langle (\lambda_{hp}^\varepsilon - DJ_{\varepsilon}(\mathbf{u}_{hp}^\varepsilon))^+ , (u_{hp,n}^\varepsilon -g)^- \right\rangle_{\Gamma_C}$. Localizing an approximation of the global a-posteriori error estimate gives rise to the following solve-mark-refine algorithm for $hp$-adaptivity.
\begin{alg} \label{alg:adaptive}
\vspace*{2mm} \hrule
\vspace*{2mm} (Solve-mark-refine algorithm for $hp$-adaptivity) \vspace*{2mm}
\hrule
\begin{enumerate}
\item Choose initial discretization $\mathcal{T}_{h}$ and $p$, steering parameters $\theta \in (0,1)$ and $\delta \in (0,1)$.
\item For $k=0,1,2,\ldots$ do
\begin{enumerate}
\item solve regularized discrete problem ($\mathcal{P}_{varepsilon, hp}$).
\item compute discrete Lagrange multiplier by \eqref{eq:reconstruct_lambda_h}
\item compute local error indicators $\Xi^2$ to current solution.
\item mark all elements $T\in \mathcal{N}$
$$ \mathcal{N}:=\operatorname{argmin}\{ | \{ \widehat{\mathcal{N}} \subset \mathcal{T}_{h}: \sum_{T\in \widehat{\mathcal{N}}} \Xi^2(T) \geq \theta \sum_{T\in \mathcal{T}_{h}} \Xi^2(T) \} | \}$$
\item estimate local analyticity, i.e.~compute Legendre coefficients of
$$v_{hp}|_T(\Psi_T(x))=\sum_{j=0}^{p_T} a_{i}L_i(x), \qquad a_{i}=\frac{2i+1}{2} \int_{-1}^1 v_{hp}|_T(\Psi_T(x)) L_i(x) \;dx$$
and use a least square approach to compute the slop $m$ of $|\log |a_{i}||=mi+b$, for each direction of $u^{hp}$ on $\Gamma_\Sigma$, of $\psi^{hp}$ on $\Gamma_D$, respectively. If $e^{-m}\leq \delta$ for all directions then $p$-refine, else $h$-refine marked element $E$. If $p_E=0$ always $p$-refine to have a decision basis next time.
\item refine marked elements based on the decision in 2(e).
\end{enumerate}
\end{enumerate}
\hrule
\end{alg}
\section{Numerical experiments}
\label{sec:NumExp}
For the adaptivity algorithm \ref{alg:adaptive} we use D\"orfler marking with marking parameter $\theta=0.3$, and the Legendre expansion strategy \cite{houston2005note} with $\delta=0.5$ for the decision between $h$- and $p$-refinement. The error of the auxiliary problem is estimated by a bubble error estimate as in e.g.~\cite{Banz2015} and the non-localized norms in the delamination related error contributions are approximated by scaled $L^2$-norms, namely $hp^{-1}\| (u^\varepsilon_{hp,n}-g)^+\|^2_{L^2(\Gamma_C)}$ and $h^{-1}p \|( \lambda_{hp}^\varepsilon - DJ_{\varepsilon}(\mathbf{u}^\varepsilon_{hp}))^- \|^2_{L^2(\Gamma_C)}$.
The integral $\langle D J_{\varepsilon}(\mathbf{u}^{\varepsilon}_{hp}), \mathbf{v}_{hp} -\mathbf{u}^{\varepsilon}_{hp} \rangle_{\Gamma_C}$ is computed by a composite Gauss-Quadrature with $\lceil p_T \rceil+17$ quadrature points per element.
If not mentioned otherwise the regularization parameter is $\varepsilon=10^{-4}$.
For the numerical experiments we choose $\Omega=(0,1/2)^2$, $\Gamma_D=\{0\} \times [0,1/2]$, $\Gamma_C=(0,1/2] \times \{0\} $, $\Gamma_N=\partial \Omega \setminus \left( \Gamma_D \cup \Gamma_C \right)$. The material parameters are $E=5$, $\nu=0.45$, $\mathbf{f} \equiv 0$, ${\mathbf t} = 0.25$ on $[1/4,1/2]\times\{1\}$ and zero elsewhere, $g=0$. The delamination law is given via
\begin{align*}
f(u_n(x))&=\min\{g_1(g(x)-u_n(x)),g_2(g(x)-u_n(x)),g_3(g(x)-u_n(x))\}\\
&= -\max\{-g_1(-u_n(x)),-g_2(-u_n(x)),-g_3(-u_n(x))\}
\end{align*}
with
\begin{gather*}
g_1(y)=\frac{A_1}{2t_1}y^2, \qquad g_2(y)=b_2(y^2-t_1^2) +d_2, \qquad g_3(y)=d_3
\end{gather*}
and parameters
\begin{gather*}
A_1=0.05, \quad
A_2=0.03,\quad
t_1=0.02, \quad t_2=0.04, \\
b_2=\frac{A_2}{2t_2}, \quad d_2=A_1\frac{t_1}{2}, \quad d_3=b_2(t_2^2-t_1^2)+d_2 .
\end{gather*}
The regularized delamination law $S_x$ with regularization parameter $\varepsilon=10^{-4}$ is plotted in Figure~\ref{fig:regDelamLaw}. The characteristic saw tooth shape is already present, but the absolute value in the tips and the slope approximating the jump are still noticeable coarse approximated.
\begin{figure}[tbp]
\centering
\includegraphics[trim = 0mm 1mm 14mm 9mm, clip,width=55.0mm, keepaspectratio]{reg_law.pdf}
\caption{Regularized delamination law $S_x$ for $\varepsilon=10^{-4}$}
\label{fig:regDelamLaw}
\end{figure}
The discrete Lagrange multiplier $\lambda^\varepsilon_{hp}$ is obtain by solving \eqref{eq:reconstruct_lambda_h} where $\psi_i$ are discontinuous, piecewise polynomials on $\Gamma_C$ on a one time coarsened mesh $(H=2h)$ with polynomial degree reduced by one $(q=p-1)$ compared to the mesh and polynomial degree distribution of $\mathbf{u}^\varepsilon_{hp}$. Figure~\ref{fig:discreteSol} displays the deformation of the rectangle and the normal stresses on $\Gamma_C$ obtained from the lowest order uniform $h$-method with 16384 elements and regularization parameter $\varepsilon=10^{-4}$. The normal stress on $\Gamma_C$, Figure~\ref{fig:discreteSol}~(b), reflects the delamination law from Figure~\ref{fig:regDelamLaw} well.
\begin{figure}[tbp]
\centering \mbox{
\subfigure[Deformation]{
\includegraphics[trim = 12mm 0mm 12mm 7mm, clip,width=51.0mm, keepaspectratio]{displacement_N4096.pdf}} \qquad
\subfigure[Contact stresses]{
\includegraphics[trim = 2mm 1mm 11mm 6mm, clip,width=55.0mm, keepaspectratio]{stress_N4096.pdf}} }
\caption{Uniform $h$-version, p=1, 16384 elements with 4096 elements on $\Gamma_C$}
\label{fig:discreteSol}
\end{figure}
Figure~\ref{fig:error} displays the reduction of the error in $(\mathbf{u},\lambda)$ and of the error estimate.
Since the exact solution is not known, we compute the error approximately by $\|\mathbf{u}_{fine}-\mathbf{u}_{hp}\|_P$ and $\|\lambda_{fine}-\lambda_{hp}\|_V$, with norms induced by the Poincar\'{e}-Steklov operator, single layer potential acting on $\Gamma_C$, respectively. The pair $(\mathbf{u}_{fine},\lambda_{fine})$ is a very fine (last) approximation for each sequence of discretization. These error expressions are only meaningful if the distance is sufficiently large, which at the end of the different discretization sequences is not true and, therefore, are then omitted. Hence, the error curves are "shorter" then the error estimate curves.\\
The uniform $h$-version with $p=1$ exhibits an experimental order of convergence (eoc) of 0.65 for the error and of 0.53 for the error estimate, all w.r.t.~degrees of freedom (dof). Hence, the error estimate is only reliable but not efficient. For the $h$-adaptive scheme with $p=1$, the eoc is 0.84 for the error and only 0.56 for the error estimate over the last 20, 10 iterations, respectively. In case of $hp$-adaptivity the eoc appears to be exponential for the error and 1.52 for the error estimate also over the last 10 iterations.
\begin{figure}[tbp]
\centering
\includegraphics[trim = 14mm 7mm 15mm 8mm, clip,width=70.0mm, keepaspectratio]{error_neu.pdf}
\caption{$\sqrt{\|\mathbf{u}_{fine}-\mathbf{u}_{hp}\|_P^2 + \|\lambda_{fine}-\lambda_{hp}\|_V^2}$ and error estimate for different families of discrete solutions, $\varepsilon=10^{-4}$}
\label{fig:error}
\end{figure}
Looking at the single contributions of the error estimate plotted in Figure~\ref{fig:errorContributions} we find that in all cases the complementarity contribution is the dominant error contribution with the lowest order of convergence and that the consistency contribution has always a significantly higher order of convergence. Given the nature of this numerical experiment, the violation of the non-penetration condition is always zero and, therefore, is omitted in these plots.
For the uniform $h$-version, Figure~\ref{fig:errorContributions}~(a), we have that the Neumann and $V^{-1}$ contributions converge at $0.62$ almost as fast as the error approximation. Given the nature, that $\lambda_{h,1}$ is piecewise constant with double mesh size, $-\mathbf{u}_{h,1} \cdot \mathbf{n}$ is linearly increasing and thus $-S_x(\mathbf{u}_{h,1} \cdot \mathbf{n})$ in the simplest case as well, the complementarity contribution is zero on the left element and non-zero on the right element. This oscillatory case occurs over a large fraction of $\Gamma_C$ and thus the $h$-adaptive scheme performs almost uniform mesh refinements on $\Gamma_C$, c.f.~Figure~\ref{fig:adaptiveMeshes}(a). Hence, the little gain in $h$-adaptivity, Figure~\ref{fig:errorContributions}~(b), for the error estimate. This oscillatory effect can be reduced by increasing the polynomial degree as in $hp$-adaptivity, which balances out the single error contributions, Figure~\ref{fig:errorContributions}~(c) and is thus capable of producing more localized refinements, c.f.~Figure~\ref{fig:adaptiveMeshes}(b). \\
\begin{figure}[tbp]
\centering \mbox{
\subfigure[uniform $h$-version with $p=1$]{
\includegraphics[trim = 14mm 7mm 15mm 8mm, clip, clip,width=58.0mm, keepaspectratio]{error_cont_uni.pdf}} \
\subfigure[$h$-adaptive with $p=1$ ]{
\includegraphics[trim = 14mm 7mm 15mm 8mm, clip,width=58.0mm, keepaspectratio]{error_cont_h.pdf}} } \\
\mbox{
\subfigure[$hp$-adaptive]{
\includegraphics[trim = 14mm 7mm 15mm 8mm, clip,width=58.0mm, keepaspectratio]{error_cont_hp.pdf}}
}
\caption{Error contributions of the error estimates}
\label{fig:errorContributions}
\end{figure}
\begin{figure}[tbp]
\centering \mbox{
\subfigure[$h$-adaptive, mesh nr.~10 (inner), nr.~20 (outer)]{
\begin{overpic} [trim = 28.5mm 10mm 22.5mm 5.5mm, clip, clip,width=55.0mm, keepaspectratio]{mesh_h_20.pdf}
\put(6.5,8){\includegraphics[trim = 29.5mm 7mm 23mm 3mm, clip, clip,width=47.0mm, keepaspectratio]{mesh_h_10.pdf}}
\end{overpic}
}
\ \
\subfigure[$hp$-adaptive, mesh nr.~15 (inner), nr.~25 (outer)]{
\begin{overpic} [trim = 28.5mm 10mm 22.5mm 5.5mm, clip, clip,width=55.0mm, keepaspectratio]{mesh_hp_25.pdf}
\put(6.7,8){\includegraphics[trim = 27.5mm 7mm 21.5mm 3mm, clip, clip,width=47.0mm, keepaspectratio]{mesh_hp_15.pdf}}
\end{overpic}
}
}
\caption{Adaptively generated meshes}
\label{fig:adaptiveMeshes}
\end{figure}
% \begin{figure}
% %\centering \mbox{
% \subfigure[$h$-adaptive, mesh nr.~10 (inner), nr.~20 (outer)]{
% %\begin{overpic}
% [trim = 137mm 18mm 120mm 10mm, clip,width=60.0mm, keepaspectratio]{mesh_h_20.pdf}
% % \put(6.6,8.5){ \includegraphics[trim = 137mm 18mm 120mm 10mm, clip, width=50.0mm, keepaspectratio]{mesh_h_10.pdf}}
% %\end{overpic} }
% % \hspace{+1.5cm}
% % \subfigure[$hp$-adaptive, mesh nr.~15 (inner), nr.~25 (outer)]{
% % \begin{overpic}[trim = 137mm 18mm 120mm 10mm, clip,width=60.0mm, keepaspectratio]{mesh_hp_25.pdf}
% % \put(6.6,8.5){ \includegraphics[trim = 137mm 18mm 120mm 10mm, clip, width=50.0mm, keepaspectratio]{mesh_hp_15.pdf}}
% % \end{overpic} }
% %}
% \caption{Adaptively generated meshes}
% \label{fig:adaptiveMeshes}
% \end{figure}
Finally, Figure~\ref{fig:errorInfluenceReg} shows the influence of the regularization parameter on the errors $\|\mathbf{u}_{fine}-\mathbf{u}_{h,1}^\varepsilon\|_P$ and $\|\lambda_{fine}-\lambda_{h,1}^\varepsilon\|_V$ where $(\mathbf{u}_{fine},\lambda_{fine})$ is the Galerkin solution to a uniform mesh with 16384 elements, $p=1$ and regularization parameter $\varepsilon=2.5 \cdot 10^{-6}$. Here, the pair $(\mathbf{u}_{h,1}^\varepsilon,\lambda_{h,1}^\varepsilon)$ is computed on a uniform mesh with 8192 elements and the regularization parameter is varied. We find that for very large $\varepsilon$, the regularized delamination law has lost its saw tooth characteristic and small variations of $\varepsilon$ have no noticeable influence. For very small $\varepsilon$ the discretization error is dominant which sets in for $\mathbf{u}$ much earlier than for $\lambda$. In the intermediate range the eoc w.r.t.~$\varepsilon$ is 0.75 for $\mathbf{u}$, whereas $\lambda$ displays an (asymptotic) eoc of 1.0 w.r.t.~$\varepsilon$. The error estimate for the regularized problem increases in $\varepsilon$ slightly.\\
{\bf Acknowledgment} The authors are grateful to J. Gwinner for the helpful discussions and comments.
\begin{figure}[tbp]
\centering
\includegraphics[trim = 14mm 7mm 15mm 8mm, clip,width=65.0mm, keepaspectratio]{sweep_error.pdf}
\caption{Error and error estimate for different regularization parameters $\varepsilon$, uniform $h$-version 8192 elements, $p=1$}
\label{fig:errorInfluenceReg}
\end{figure}
%\begin{acknowledgements}
%If you'd like to thank anyone, place your comments here
%and remove the percent signs.
%\end{acknowledgements}
% BibTeX users please use one of
%\bibliographystyle{spbasic} % basic style, author-year citations
\bibliographystyle{spmpsci} % mathematics and physical sciences
%\bibliographystyle{spphys} % APS-like style for physics
\bibliography{bib_HVI} % name your BibTeX data base
% % Non-BibTeX users please use
% \begin{thebibliography}{}
% %
% % and use \bibitem to create references. Consult the Instructions
% % for authors for reference list style.
% %
% \bibitem{RefJ}
% % Format for Journal Reference
% Author, Article title, Journal, Volume, page numbers (year)
% % Format for books
% \bibitem{RefB}
% Author, Book title, page numbers. Publisher, place (year)
% % etc
% \end{thebibliography}
\end{document}
% end of file template.tex
\begin{thebibliography}{10}
\providecommand{\url}[1]{{#1}}
\providecommand{\urlprefix}{URL }
\expandafter\ifx\csname urlstyle\endcsname\relax
\providecommand{\doi}[1]{DOI~\discretionary{}{}{}#1}\else
\providecommand{\doi}{DOI~\discretionary{}{}{}\begingroup
\urlstyle{rm}\Url}\fi
\bibitem{At}
Attouch, H.: Variational convergence for functions and operators.
\newblock Pitman, Boston (1984)
\bibitem{Banz2015Stab}
Banz, L., Gimperlein, H., Issaoui, A., Stephan, E.P.: {S}tabilized mixed
$hp$-{BEM} for frictional contact problems in linear elasticity.
\newblock Numer. Math. (2016).
\newblock DOI: 10.1007/s00211-016-0797-y
\bibitem{banz2013posteriori}
Banz, L., Stephan, E.P.: {A posteriori error estimates of $hp$-adaptive
IPDG-FEM for elliptic obstacle problems}.
\newblock Appl. Numer. Math. \textbf{76}, 76--92 (2014)
\bibitem{Banz2015}
Banz, L., Stephan, E.P.: {O}n hp-adaptive {BEM} for frictional contact problems
in linear elasticity.
\newblock Comput. Math. Appl. \textbf{69}(7), 559--581 (2015)
\bibitem{Bernardi}
Bernardi, C., Maday, Y.: {S}pectral methods.
\newblock In: Handb. Numer. Anal., vol.~V, pp. 209--485. North-Holland,
Amsterdam (1997)
\bibitem{braess2005posteriori}
Braess, D.: {A} posteriori error estimators for obstacle problems--another
look.
\newblock Numer. Math. \textbf{101}(3), 415--421 (2005)
\bibitem{cc}
Carstensen, C.: {I}nterface problem in holonomic elastoplasticity.
\newblock Math. Methods Appl. Sci. \textbf{16}(11), 819--835 (1993)
\bibitem{cc96}
Carstensen, C.: {A} posteriori error estimate for the symmetric coupling of
finite elements and boundary elements.
\newblock Computing \textbf{57}(4), 301--322 (1996)
\bibitem{cf}
Carstensen, C., Funken, S.A., Stephan, E.P.: {O}n the adaptive coupling of
{FEM} and {BEM} in 2--d--elasticity.
\newblock Numer. Math. \textbf{77}(2), 187--221 (1997)
\bibitem{ccjg}
Carstensen, C., Gwinner, J.: {FEM} and {BEM} coupling for a nonlinear
transmission problem with {S}ignorini contact.
\newblock SIAM J. Numer. Anal. \textbf{34}(5), 1845--1864 (1997)
\bibitem{Chernov}
Chernov, A.: {N}onconforming boundary elements and finite elements for
interface and contact problems with friction - hp-version for mortar, penalty
and {N}itsche's methods.
\newblock Ph.D. thesis, Universit\"{a}t Hannover (2006)
\bibitem{chernov2008hp}
Chernov, A., Maischak, M., Stephan, E.P.: hp-{M}ortar boundary element method
for two-body contact problems with friction.
\newblock Math. Methods Appl. Sci. \textbf{31}(17), 2029--2054 (2008)
\bibitem{Clarke}
Clarke, F.H.: {O}ptimization and nonsmooth analysis.
\newblock John Wiley and Sons (1983)
\bibitem{Co-1}
Costabel, M.: {Boundary integral operators on Lipschitz domains: Elementary
results}.
\newblock SIAM J. Math. Anal. \textbf{19}(3), 613--626 (1988)
\bibitem{CoSt}
Costabel, M., Stephan, E.P.: {C}oupling of finite and boundary element methods
for an elastoplastic interface problem.
\newblock SIAM J. Numer. Anal. \textbf{27}(5), 1212--1226 (1990)
\bibitem{Noll_Ovcharova}
Dao, M., Gwinner, J., Noll, D., Ovcharova, N.: {N}onconvex bundle method with
application to a delamination problem.
\newblock Computational Optimization and Applications (2016).
\newblock DOI: 10.1007/s10589-016-9834-0
\bibitem{Gl}
Glowinski, R.: {N}umerical methods for nonlinear variational problems.
\newblock Springer Series in Computational Physics (1984)
\bibitem{Gu}
Guediri, H.: {O}n a boundary variational inequality of the second kind
modelling a friction problem.
\newblock Math. Methods Appl. Sci. \textbf{25}(2), 93--114 (2002)
\bibitem{Gwinner_PhD}
Gwinner, J.: {N}ichtlineare {V}ariationsungleichungen mit {A}nwendungen.
\newblock Ph.D. thesis, Universit\"{a}t Mannheim (1978)
\bibitem{gwinner2013hp}
Gwinner, J.: hp-{FEM} convergence for unilateral contact problems with {T}resca
friction in plane linear elastostatics.
\newblock J. Comput. Appl. Math. \textbf{254}, 175--184 (2013)
\bibitem{Gw_Ov}
Gwinner, J., Ovcharova, N.: {F}rom solvability and approximation of variational
inequalities to solution of nondifferentiable optimization problems in
contact mechanics.
\newblock Optimization (ahead-of-print), 1--20 (2015)
\bibitem{GS-1}
Gwinner, J., Stephan, E.P.: {A} boundary element procedure for contact problems
in linear elastostatics.
\newblock ESAIM Math. Model. Numer. Anal. \textbf{27}(4), 457 -- 480 (1993)
\bibitem{Ha}
Han, H.D.: {A} direct boundary element method for {S}ignorini problems.
\newblock Math. Comp. \textbf{55}(191), 115--128 (1990)
\bibitem{houston2005note}
Houston, P., S{\"u}li, E.: {A note on the design of hp-adaptive finite element
methods for elliptic partial differential equations}.
\newblock Comput. Methods Appl. Mech. Engrg. \textbf{194}(2-5), 229--243 (2005)
\bibitem{HsWendl}
Hsiao, G.C., Wendland, W.L.: {B}oundary integral equations.
\newblock Springer (2008)
\bibitem{Kleiber}
Kleiber, M., Borkowski, A.: {H}andbook of computational solid mechanics: survey
and comparison of contemporary methods.
\newblock Springer (1998)
\bibitem{krebs2007p}
Krebs, A., Stephan, E.: {A p-version finite element method for nonlinear
elliptic variational inequalities in 2D}.
\newblock Numer. Math. \textbf{105}(3), 457--480 (2007)
\bibitem{MS-2}
Maischak, M., Stephan, E.P.: {A} {FEM}--{BEM} coupling method for a nonlinear
transmission problem modelling {C}oulomb friction contact.
\newblock Comput. Methods Appl. Mech. Engrg. \textbf{194}(2), 453--466 (2005)
\bibitem{MS-1}
Maischak, M., Stephan, E.P.: {A}daptive hp-versions of {BEM} for {S}ignorini
problems.
\newblock Appl. Numer. Math. \textbf{54}(3), 425--449 (2005)
\bibitem{maischak2007adaptive}
Maischak, M., Stephan, E.P.: {A}daptive hp-versions of boundary element methods
for elastic contact problems.
\newblock Comput. Mech. \textbf{39}(5), 597--607 (2007)
\bibitem{Nesemann}
Nesemann, L., Stephan, E.P.: {N}umerical solution of an adhesion problem with
{FEM} and {BEM}.
\newblock Appl. Numer. Math. \textbf{62}(5), 606--619 (2012)
\bibitem{Ovcharova2012}
Ovcharova, N.: {R}egularization {M}ethods and {F}inite {E}lement
{A}pproximation of {H}emivariational {I}nequalities with {A}pplications to
{N}onmonotone {C}ontact {P}roblems.
\newblock Ph.D. thesis, Universit\"{a}t der Bundeswehr M\"{u}nchen (2012)
\bibitem{Ovcharova2015}
Ovcharova, N.: On the coupling of regularization techniques and the boundary
element method for a hemivariational inequality modelling a delamination
problem.
\newblock Math. Methods Appl. Sci. (2015).
\newblock ArXiv 1603.05091
\bibitem{Ov_Gw}
Ovcharova, N., Gwinner, J.: {A} study of regularization techniques of
nondifferentiable optimization in view of application to hemivariational
inequalities.
\newblock J. Optim. Theory Appl. \textbf{162}(3), 754--778 (2014)
\bibitem{SlSp}
Sloan, I.H., Spence, A.: {T}he {G}alerkin method for integral equations of the
first kind with logarithmic kernel: {T}heory.
\newblock IMA J. Numer. Anal. \textbf{8}(1), 105--122 (1988)
\bibitem{Steinbach}
Steinbach, O.: {N}umerische {N}{\"a}herungsverfahren f{\"u}r elliptische
{R}andwertprobleme: {F}inite {E}lemente und {R}andelemente.
\newblock Vieweg + Teubner (2003)
\bibitem{stephan2009hp}
Stephan, E.P.: {T}he hp-version of {BEM}—fast convergence, adaptivity and
efficient preconditioning.
\newblock J. Comput. Math. \textbf{27}(2-3), 348--359 (2009)
\bibitem{Gudladt}
Wetzel, M., Holtmannsp{\"o}tter, J., Gudladt, H.J., Czarnecki, J.:
{S}ensitivity of double cantilever beam test to surface contamination and
surface pretreatment.
\newblock Int. J. Adhes. Adhes. \textbf{46}, 114--121 (2013)
\end{thebibliography}