1: \section{Regularity and Estimates for \boxb}\setS{BB}
2:
3: We shall now restrict our attention to the class of domains described in \rfE[MC]{Model}, i.e. $\Omega = D \times N$ in the strictly pseudoconvex manifold $M=\hy{2} \times N$ . For the analysis of the preceding sections, this demands that we restrict our attention to precompact sets $D$. We shall permit $\nu$ to be any non-negative constant and it will be suppressed from any notation.
4:
5: We need weighted versions of the Folland-Stein spaces to deal with the weighted estimates of the previous section. Accordingly we define
6: \bgD{Weights}
7: We define the spaces $\mathscr{V}^{k}\SjO{m}$ inductively by $\mathscr{V}^{0}\SjO{m} = \SjO{m}$ and
8: \[ \mathscr{V}^{k+1} \SjO{m} =\left \{ \varphi \in \mathscr{V}^{k}\SjO{m}: \rho \NabH \varphi, \NabH[\top]\varphi \in \mathscr{V}^{k}\SjO{m} \right\}\] with corresponding inductively defined norms.
9: \enD
10: Here $\NabH[\top]$ is the tangential component of the connection given explicitly by
11: \[ \NabH = \NabH[\top] + \nabla_\bt{Y} \otimes \ut^\bt{0} + \nabla_Y \otimes \ut^0.\]
12: Since we are assuming that $D$ is precompact, we may take $\rho$ to be any smooth defining function for $D$ and identify it naturally with a function on $M$.
13:
14: There are currently no sharp trace theorems for the Folland-Stein spaces on strictly pseudoconvex CR manifolds. However we can employ the partial Fourier decomposition to develop a useful notion of vanishing at the boundary.
15: \bgD{Vanishing}
16: We shall say a $(0,q)$-form $\varphi \in \SjO{j}$ vanishes on the boundary $\dO$ and write $\varphi \in \So[j]$ if each function $\varphi^\top_\s, \varphi^\bot_\s$ in the partial Fourier decomposition of $\varphi$ is contained in $\Wo[j]$. \enD
17: It is not difficult to show that $\So[1]$ is equal to the closure of $\Cic{\Omega}$ in $\SjO{1}$. Also $\So[j] =\SjO{j} \cap \So[1]$. We can now describe the domain spaces for our sharp estimate for \boxb.
18: \bgD{FinalSpace}
19: \[ \HV{k,j} := \left\{ \varphi \in \mathscr{V}^{j}\SjO{k}: \varphi^\bot \in \So[k+j], \nabla_{\bt{Y}} \varphi^\top \in \So[k+j-1] \right\}\]
20: with norm
21: \[ \norm[2]{\varphi}{\HV{k,j}} := \norm[2]{\varphi}{\mathscr{V}^{j}\SjO{k}} + \norm[2]{\varphi^\bot}{\SjO{k+j}} + \norm[2]{\nabla_{\bt{Y}} \varphi^\top}{\SjO{k+j-1}}.\]
22: \enD
23: In particular, we note that $\SjO{k+j} \subset \HV{k,j} \subset \SjO{k}$ with the inclusion maps continuous.
24:
25: Now set $\Kr:= \Ker {\boxb}$ with $\boxb$ regarded as an unbounded operator on $L^2(\Lambda_\ut^{0,q}\Omega)$. Then since \[\ip{\boxb \varphi}{\varphi}{L^2(\Omega)} = \ \norm[2]{\db \varphi}{L^2(\Omega)} + \norm[2]{\dbs \varphi}{L^2(\Omega)}\] it follows immediately that $\Kr = \Ker{\db} \cap \Ker{\dbs}$.
26:
27: \bgT{Main}
28: Suppose $\Omega$ is as in \rfE[MC]{Model}. Fix $1 \leq q \leq n-2$. Then for each $k \geq0$, the operators
29: \begin{align*}
30: \boxb&: \HV{k,2} \cap (\Kr)^\bot \to \SjO{k} \cap (\Kr)^\bot\\
31: 1&+ \boxb: \HV{k,2} \to \SjO{k}
32: \end{align*}
33: are both isomorphisms on $(0,q)$-forms.
34: \enT
35:
36: \pf Expand out the $(0,q)$-form $\varphi$ in terms of the partial Fourier decomposition and decompose the relevant norms using \rfL[MC]{FourierFS} . Apply \rfT[TV]{Transverse} to each of the tranverse components and \rfT[WR]{Tangential} to each of the tangential components of $\varphi$. The constants are uniform across all choices of $\s$ so a simple comparison theorem argument completes the proof.
37: \epf
38:
39: \bgC{NonHomogeneous}
40: Suppose the $(0,q)$-form $\vs \in \SjO{k}$ satisfies $\db \vs =0$ and $\vs \bot \Kr$. Then there exists a unique $\varphi \in \HV{k,1}$ such that $\varphi \bot \Kr$ and $\db \varphi = \vs$. Furthermore there is a constant $C$ independent of $\vs$ such that \[\norm{\varphi}{\HV{k,1}} \leq C\norm{\vs}{\SjO{k}}.\]
41: \enC
42:
43: \pf
44: By \rfT{Main} there exists a unique $\ua \in \HV{k,2} \bot \Kr$ such that $\boxb \ua = (\db \dbs + \dbs \db) \ua =\vs$ and $\norm{\ua}{\HV{k,2}} \leq C\upp \norm{\vs}{\SjO{k}}$ for some $C\upp$ independent of $\vs$. From $\db \vs =0$ it follows that $\vs \bot \rng{\dbs}$ and so $\vs = \db \dbs \ua$. Take $\varphi = \dbs \ua$. From the proof of \rfL[MC]{KohnLaplacian} we know that $\dbs[\top]$ and $\nabla_\bt{Y}$ commute. Tangential derivatives map $\SjO{k+1}$ to $\SjO{k}$. Thus from \rfL[MC]{Facts} it follows that $\dbs$ is a bounded linear operator from $\HV{k,2}$ to $\HV{k,1}$.
45:
46: \epf
47:
48: \bgL{Cohomology}
49: Suppose $1\leq q \leq n-2$. If $\hKR{0,q}{N}=0$ then $\Kr = 0$ otherwise $\dim \Kr = \infty$.
50: \enL
51:
52: \pf
53: A $(0,q)$-form $\varphi$ is in $\Kr$ if and only if $\varphi^\bot=0$ and $\varphi^\top$ expands as a sum of forms $\varphi^\top_\s \s$ such that $\Gm=0$ and $\nWs \varphi^\top_\s =0$. Now $\{\s \in \Vv: \Gm =0\} \cong \hKR{0,q}{N}$. If this set is non-empty then we can construct elements of $\Kr$ from any holomorphic function multiplied by an appropriate power of $s$.
54: \epf
55:
56: \bgC{Cohomology}
57: If $\hKR{0,q}{N}=0$ then $\hKR{0,q}{\Omega} = 0$ otherwise $\dim \hKR{0,q}{\Omega} = \infty$.
58:
59: \enC
60:
61: \pf
62: This now follows immediately from the observation that $\Kr \bot\rng{\db}$.
63: \epf
64:
65: \bgL{Hypoelliptic}
66: The operator $\boxb$ is hypoelliptic up to the boundary if and only if $\hKR{0,q}{N}=0$.
67: \enL
68:
69: \pf
70: If $\hKR{0,q}{N}=0$ then $\boxb$ is injective by \rfL{Cohomology}. Thus by \rfT{Main} if $\boxb \varphi \in \Ci{\Omega}$ then $\varphi = \bigcup\limits_{k \geq 0} \SjO{k}$. Since there is a continuous inclusion $\SjO{2k} \subset \HjO{k}$, hypoellipticity follows from the Sobolev Lemma.
71:
72: If $\hKR{0,q}{N} \ne 0$ then we can choose some $\s \in \Vv$ such that $\Gm =0$, then following \rfX[TG]{NegRegularity} we can easily construct $\varphi \in \Kr$ such that $\varphi \notin \SjO{1}$.
73:
74: \epf
75:
76: It is worth pointing out here that regardless of the cohomology of the foliating manifold $N$, the operator $1+\boxb$ is always hypoelliptic up to the boundary of $\Omega$. Likewise $\boxb$ is always hypoelliptic in the interior.
77:
78: \bgL{NonCompact}
79: For any $1\leq q \leq n-2$ The bounded operator $(1+\boxb)^{-1}: L^2(\Omega) \to L^2(\Omega)$ is not compact on $(0,q)$-forms.
80: \enL
81:
82: \pf If $\hKR{0,q}{N}\ne 0$ then this can be seen from \rfL{Cohomology}. However in general
83: this is really a statement about the operators $\Pt$. Fix $w_0 \in D$ and $\e>0$ such that the disc $|w-w_0|<\e$ is contained in $D$. Choose any $\s \in \Vv$ and let $f$ be a function on $D$ that can be expressed as the product of $s^{(\uls+\nu)/2}$ and a holomorphic function. Then $(1+\Pt) f = (1+\Gm) f$. Now consider the sequence $f_k$ where $f_k= s^{(\uls+\nu)/2} c_k w^k$ where $c_k$ is chosen to make $\norm{f_k}{L^2(D)}=1$. If some subsequence $f_j$ converges in $L^2(D)$ then the same subsequence must converge in $L^2(|w-w_0|<\e)$. However the functions $(w-w_0)^j$ are mutually orthogonal on $|w-w_0|<\e$ in the Euclidean $L^2$ metric. Since $s$ is bounded above and below on this disc, it follows that the subsequence $f_j$ cannot converge on $|w-w_0|$ in the hyperbolic $L^2$ metric. Thus the bounded sequence $f_k$ has no convergent subsequences and so $(1+\Pt)^{-1}$ is not compact as an operator on $L^2(D)$.
84:
85: \epf
86:
87: As an immediate consequence of this, we see
88: \bgC{NonSub}
89: The operator $\boxb$ is not globally subelliptic on $\Omega$, i.e. there is no $\e >0$ such that
90: \[ \norm{\varphi}{\HjO{\e}} \leq C \norm{\boxb \varphi}{L^2(\Omega)} + \norm{\varphi}{L^2(\Omega)}.\]
91: \enC
92:
93: For completeness, in addition to the global theory, we can combine the results of \rfT[TV]{Local} and \rfC[TG]{Local} to establish the following local estimates for $\boxb$.
94:
95: \bgT{Local}
96: Under the same conditions as \rfT{Main}, then for all $k \geq 0$ and any pair of smooth cut-off functions $\xi \subset \zeta$ on $M$ that depend solely on $w$ there is a constant $C>0$ such that
97: \[ \norm{ \xi \varphi}{\HV{k,2}} \leq C \norm{\zeta}{\SjO{k}} + \norm{\zeta \varphi}{\lO} + \norm{ \nabla_\bt{Y} \varphi^\top}{\lO}\] whenever $\varphi \in \HV{k,2}$.
98: \enT
99:
100: