Define the linear operator
\(P: X\rightarrow X\) by
$$ (Pu) (t)=\frac{1}{g(1)}g(B)u(1)t. $$
Then we have
$$\begin{aligned} \bigl(P^{2}u\bigr) (t)&=\frac{1}{g(1)}g(B) \biggl( \frac{1}{g(1)}g(B)u(1) \biggr)t \\ &= \frac{1}{g^{2}(1)}g(B)g(B)u(1)t=\frac{1}{g(1)}g(B)u(1)t=(Pu) (t). \end{aligned}$$
This shows that
P is a continuous projection operator. In the following, we will assert that
\(\operatorname{Im} P= \operatorname{Ker} L\). In fact, if
\(v\in \operatorname{Im} P\), there is
\(u\in X\), such that
$$ v(t)=Pu(t)=\frac{1}{g(1)}g(B)u(1)t. $$
Thus, it follows from (
2.6) that
\(v\in \operatorname{Ker} L\). Conversely, if
\(v\in \operatorname{Ker} L\), we have
$$ v(t)=\psi t,\quad \psi \in \operatorname{Ker}(I-B). $$
On account of the second identity in (
2.6), there exists
\(\psi _{1} \in \mathbb{R}^{n}\), such that
\(\psi =g(B)\psi _{1}\). Taking
\(u(t)=g(1)\psi _{1}\), we then have
$$ Pu(t)=\frac{1}{g(1)}g(B)u(1)t=g(B)\psi _{1}t=\psi t, $$
which implies that
\(v\in \operatorname{Im} P\). Thus, we conclude that
\(\operatorname{Im} P=\operatorname{Ker} L\), and consequently,
$$ X=\operatorname{Ker} L\oplus \operatorname{Ker} P. $$
Therefore, the generalized inverse
\(K_{P}: \operatorname{Im} L\rightarrow \operatorname{dom}L\cap \operatorname{Ker} P\) can be given by
$$ (K_{P}v) (t)= \int _{0}^{1}G(t,s)v(s)\,ds+l\biggl(I- \frac{1}{g(1)}g(B)\biggr)\varphi (v)t, $$
(2.9)
where the constant
l is given in (H4). Note that, since
\(G(1,s)=0\) for all
\(s\in [0,1]\) and from (
2.9), we have
$$ P(K_{P}v) (t)=\frac{l}{g(1)}g(B) \biggl(I-\frac{1}{g(1)}g(B) \biggr)\varphi (v)t=0. $$
Hence,
\(K_{P}v\in \operatorname{Ker} P\). For
\(v\in \operatorname{Im} L\), we know that
$$ (K_{P}v) (1)=l\biggl(I-\frac{1}{g(1)}g(B)\biggr)\varphi (v) $$
and
$$\begin{aligned} & A \int _{0}^{1}(K_{P}v) (t)\,d\alpha (t) \\ &\quad= A \int _{0}^{1} \biggl( \int _{0}^{1}G(t,s)v(s)\,ds+l\biggl(I- \frac{1}{g(1)}g(B)\biggr) \varphi (v)t \biggr) \,d\alpha (t) \\ &\quad= \varphi (v)+lA\biggl(I-\frac{1}{g(1)}g(B)\biggr)\varphi (v) \int _{0}^{1} t\,d\alpha (t) \\ &\quad= \varphi (v)+lA\biggl(I-\frac{1}{g(1)}g(B)\biggr)\varphi (v)h =\varphi (v)+lB\biggl(I- \frac{1}{g(1)}g(B)\biggr)\varphi (v) \\ &\quad= \varphi (v)+l(B-I+I) \biggl(I-\frac{1}{g(1)}g(B)\biggr)\varphi (v) \\ &\quad= \varphi (v)-l(I-B) \biggl(I-\frac{1}{g(1)}g(B)\biggr)\varphi (v) +l \biggl(I- \frac{1}{g(1)}g(B)\biggr)\varphi (v) \\ &\quad= \varphi (v)-l(I-B)\varphi (v) +l\biggl(I-\frac{1}{g(1)}g(B)\biggr) \varphi (v) =l\biggl(I- \frac{1}{g(1)}g(B)\biggr)\varphi (v). \end{aligned}$$
Therefore,
\((K_{P}v)(1)=A\int _{0}^{1}(K_{P}v)(t)\,d\alpha (t)\), and consequently,
\(K_{P}\) is well defined. Furthermore, if
\(u\in \operatorname{dom}L \cap \operatorname{Ker} P\), then, using (
2.7) and Lemma
2.3, we have
$$\begin{aligned} &(K_{P}Lu) (t)\\ &\quad = - \int _{0}^{1}G(t,s)u''(s) \,ds+l\biggl(I-\frac{1}{g(1)}g(B)\biggr)\varphi (Lu)t \\ &\quad= u(t)-u(1)t+ l\biggl(I-\frac{1}{g(1)}g(B)\biggr)A \biggl( \int _{0}^{1} \bigl(u(t)-u(1)t\bigr)\,d \alpha (t) \biggr)t \\ &\quad= u(t)-u(1)t+ l\biggl(I-\frac{1}{g(1)}g(B)\biggr) \biggl(A \int _{0}^{1}u(t)\,d\alpha (t)-Ahu(1) \biggr)t \\ &\quad= u(t)-u(1)t+ l\biggl(I-\frac{1}{g(1)}g(B)\biggr) (I-B)u(1)t \\ &\quad= u(t)-u(1)t+ l(I-B)u(1)t \\ &\quad= u(t). \end{aligned}$$
This shows that
\(K_{P}=(L|_{\operatorname{dom}L\cap \operatorname{Ker} P})^{-1}\) and that
\(LK_{P}v(t)=v(t)\),
\(v\in \operatorname{Im} L\). For
\(v\in \operatorname{dom}L\), by (
2.8) we obtain
$$ (K_{P}v)'(t)= \int _{t}^{1}(1-s)v(s)\,ds- \int _{0}^{t}sv(s)\,ds+l\biggl(I- \frac{1}{g(1)}g(B)\biggr)A \int _{0}^{1} \int _{0}^{1}G(t,s)v(s)\,ds\,d\alpha (t). $$
Notice that
$$\begin{aligned} &\max_{t\in [0,1]} \biggl\vert \int _{0}^{1}G(t,s)v(s)\,ds \biggr\vert \leq \max _{t\in [0,1]} \int _{0}^{1}t(1-t) \bigl\vert v(s) \bigr\vert \,ds \leq \frac{ \Vert v \Vert _{1}}{4}, \\ &\max_{t\in [0,1]} \biggl\vert \int _{0}^{1}G'_{t}(t,s)v(s) \,ds \biggr\vert = \max_{t\in [0,1]} \biggl\vert - \int _{0}^{t}sv(s)\,ds + \int _{t}^{1}(1-s)v(s)\,ds \biggr\vert \\ &\phantom{\max_{t\in [0,1]} \biggl\vert \int _{0}^{1}G'_{t}(t,s)v(s) \,ds \biggr\vert }\leq \max_{t\in [0,1]} \biggl(t \int _{0}^{t} \bigl\vert v(s) \bigr\vert \,ds+(1-t) \int _{t} ^{1} \bigl\vert v(s) \bigr\vert \,ds \biggr) \\ &\phantom{\max_{t\in [0,1]} \biggl\vert \int _{0}^{1}G'_{t}(t,s)v(s) \,ds \biggr\vert }\leq \max_{t\in [0,1]} \biggl(t \int _{0}^{1} \bigl\vert v(s) \bigr\vert \,ds+(1-t) \int _{0} ^{1} \bigl\vert v(s) \bigr\vert \,ds \biggr)= \Vert v \Vert _{1}, \end{aligned}$$
and
$$\begin{aligned} \biggl\vert \int _{0}^{1} \int _{0}^{1}G(t,s)v(s)\,ds\,d\alpha (t) \biggr\vert &\leq \int _{0}^{1} \int _{0}^{1}t(1-t) \bigl\vert v(s) \bigr\vert \,ds\,d \Biggl(\bigvee_{0}^{t}(\alpha ) \Biggr)\\ & = \int _{0}^{1}t(1-t)\,d \Biggl(\bigvee _{0}^{t}(\alpha ) \Biggr) \cdot \Vert v \Vert _{1}, \end{aligned}$$
where
\(\bigvee_{0}^{t}(\alpha )\) denotes the total variation of
α on
\([0,t]\) defined by
$$ \bigvee_{0}^{t}(\alpha )=\sup _{\mathcal{P}}\sum_{i=1}^{n} \bigl\vert \alpha (t_{i})-\alpha (t_{i-1}) \bigr\vert , $$
where the supremum runs over the set of all partitions
$$ \mathcal{P}=\bigl\{ P=\{t_{0}, \ldots, t_{n}\} | P \text{ is a partition of } [0,t] \bigr\} . $$
Let
\(\|\cdot \|_{*}\) be the max-norm of matrices defined by
$$ \Vert A \Vert _{*}=\max_{1\leq i\leq n, 1\leq j\leq m} \vert a_{ij} \vert ,\quad \text{for } A=(a_{ij})_{n\times m}, $$
and
\(\|\cdot \|_{\mathbb{R}^{n}}\) be the maximum norm in
\(\mathbb{R} ^{n}\). Then we have
$$ \Vert Av \Vert _{\mathbb{R}^{n}}\leq m \Vert A \Vert _{*} \Vert v \Vert _{\mathbb{R}^{m}},\quad \text{for } A=(a_{ij})_{n\times m} \text{ and } v\in \mathbb{R} ^{m}. $$
Thus,
$$ \Vert K_{P}v \Vert _{\infty }\leq \Biggl( \frac{1}{4}+n \vert l \vert \cdot \biggl\Vert \biggl(I- \frac{1}{g(1)}g(B)\biggr)A \biggr\Vert _{*} \int _{0} ^{1}t(1-t)\,d \Biggl(\bigvee _{0}^{t}(\alpha ) \Biggr) \Biggr) \Vert v \Vert _{1} $$
and
$$ \bigl\Vert K'_{P}v \bigr\Vert _{\infty }\leq \Biggl(1+n \vert l \vert \cdot \biggl\Vert \biggl(I-\frac{1}{g(1)}g(B) \biggr)A \biggr\Vert _{*} \int _{0}^{1}t(1-t) \,d \Biggl(\bigvee _{0}^{t}(\alpha ) \Biggr) \Biggr) \Vert v \Vert _{1}. $$
Consequently, we conclude that
$$ \Vert K_{P}v \Vert _{X}\leq M \Vert v \Vert _{1}, $$
where
\(M=1+n|l| \|(I-\frac{1}{g(1)}g(B))A\|_{*}\int _{0}^{1}t(1-t)\,d (\bigvee_{0}^{t}(\alpha ) )\). It is easy to see that
$$ (QNu) (t)=kg(B)A \int _{0}^{1} \int _{0}^{1}G(t,s)f\bigl(s,u(s), u'(s) \bigr)\,ds\,d \alpha (t)\in \mathbb{R}^{n} $$
and
$$\begin{aligned} & K_{P}(I-Q)Nu(t)\\ &\quad=K_{P}Nu(t)-K_{P}QNu(t) \\ &\quad= \int _{0}^{1}G(t,s)Nu(s)\,ds+l\biggl(I- \frac{1}{g(1)}g(B)\biggr)\varphi (Nu)t \\ &\qquad{} + \int _{0}^{1}G(t,s)\,ds\cdot QNu(t)+l\biggl(I- \frac{1}{g(1)}g(B)\biggr)\varphi (QNu)t \\ &\quad= \int _{0}^{1}G(t,s)Nu(s)\,ds+\frac{t(1-t)}{2} QNu(t) \\ &\qquad{} +l\biggl(I-\frac{1}{g(1)}g(B)\biggr) \bigl(\varphi (Nu)+\varphi (QNu) \bigr)t. \end{aligned}$$
By using the standard argument, we can show that
\(QN((\overline{ \varOmega }))\) is bounded and
\(K_{P}(I-Q)N(\overline{\varOmega })\) is compact. Thus,
N is
L-compact on
Ω̅. □