Tangential CR complex, ∂̄_b, and the Lewy example

Intuition Beginner

A holomorphic function of several variables lives in an open region of complex space. But often the interesting data sits on the edge of such a region — a curved wall, one real dimension thinner than the space around it. The question this unit answers is: what does it mean for a function defined only on that wall to "remember" that it came from a holomorphic function inside?

In one complex variable a wall is just a curve, and there is no room on it for any complex structure to survive. In two or more variables the wall is thick enough to keep a partial complex structure. At each point of the wall there is a family of directions along which the wall still looks complex — directions you can rotate by a quarter turn and stay on the wall. Those directions form what people call the complex-tangential part of the wall.

A function on the wall is called a CR function when it is, in effect, holomorphic along exactly those surviving complex directions. The boundary operator that measures the failure of this — call it the tangential boundary operator — is the central character. It is the wall's own version of the rule "no dependence on the conjugate variable."

Visual Beginner

Picture a smooth curved surface sitting inside a higher-dimensional space, drawn by analogy with an ordinary curved sheet in three dimensions. At a chosen point on the surface, attach the flat tangent plane. Inside that tangent plane, shade a smaller sub-plane: these are the directions that still carry a quarter-turn rotation keeping you on the surface — the complex-tangential directions. The remaining one direction, sticking partly out of the surface, is where the complex structure is lost.

A CR function is one whose rate of change is consistent with holomorphy along the shaded sub-plane only. A second panel shows two nearby complex-tangential arrows whose bracket — the failure of the surface to close up under these motions — tilts into the lost direction; the size of that tilt is the curvature quantity that decides whether the wall's equations can be solved.

Worked example Beginner

Take the simplest interesting wall: in two complex variables write the first coordinate as $x+iy$ and look at the surface where the imaginary part of the second coordinate equals the squared size of the first. This is a curved real wall of real dimension three, sitting in a real space of dimension four. It is the model wall that every nice curved boundary looks like up close.

Along this wall there is exactly one complex-tangential direction, packaged into a single boundary operator built from the wall's coordinates. A function on the wall is CR when this one operator sends it to zero. The boundary values of any holomorphic function from the region inside automatically pass this test: holomorphy in the region forces the tangential equation on the wall.

What this tells us: the wall inherits a stripped-down piece of the ambient complex world. The full set of holomorphic equations becomes a single equation along the surviving direction, and that one equation is the gatekeeper deciding which surface functions could possibly be boundary values of something holomorphic inside.

Check your understanding Beginner

Formal definition Intermediate+

Let $M\subset {\mathbb{C}}^n$ be a smooth real hypersurface, locally $M=\{\rho =0\}$ with $d\rho \ne 0$. Write $T^{1,0}{\mathbb{C}}^n$ for the holomorphic tangent bundle, spanned by $\partial /\partial z_j$, and let $T_{\mathbb{C}}M=TM{\otimes }_{\mathbb{R}}\mathbb{C}$ be the complexified tangent bundle of $M$.

Definition (CR bundle). The CR bundle, or holomorphic tangent bundle of $M$, is $$ T^{1,0}M ;=; T_{\mathbb{C}}M ,\cap, T^{1,0}\mathbb{C}^n . $$ Equivalently, with $J$ the ambient complex structure on $T{\mathbb{C}}^n\cong {\mathbb{R}}^{2n}$, the real part is $HM=TM\cap J(TM)$, the maximal $J$-invariant subspace of the real tangent space, and $T^{1,0}M$ is its $(1,0)$-piece. For a smooth hypersurface $T^{1,0}M$ has constant rank $n-1$ over $\mathbb{C}$; a section is a combination ${\sum }_j{a}_j\partial /\partial z_j$ that is tangent to $M$, that is, annihilates $\rho$. The pair $(M,T^{1,0}M)$ is the induced CR structure. It satisfies the integrability condition $[\Gamma (T^{1,0}M),\Gamma (T^{1,0}M)]\subseteq \Gamma (T^{1,0}M)$, inherited from the ambient flatness, so the structure is an embedded CR structure of hypersurface type.

Definition (tangential Cauchy-Riemann operator). Let ${\bar{L}}_1,\dots ,{\bar{L}}_{n-1}$ be a local frame for $\overline{T^{1,0}M}=T^{0,1}M$. A smooth function $u$ on $M$ is a CR function when ${\bar{L}}_{k}u=0$ for every $k$. More invariantly, extend $u$ to a smooth $\tilde{u}$ on a neighbourhood in ${\mathbb{C}}^n$ and set $$ \bar\partial_b u ;=; \big(\bar\partial \tilde u\big)\big|_{T^{0,1}M} , $$ the restriction of the ambient anti-holomorphic differential to the conjugate CR directions. The result is independent of the extension because two extensions differ by a multiple of $\rho$, whose anti-holomorphic differential is a multiple of the boundary-defining differential and dies on $T^{0,1}M$. This defines the tangential Cauchy-Riemann operator ${\bar{\partial }}_b$ on functions, with $u$ a CR function exactly when ${\bar{\partial }}_{b}u=0$.

Definition (tangential CR complex). Let ${\Lambda }^{0,q}M$ be the bundle of $(0,q)$-forms on $M$, the $q$-th exterior power of the dual of $T^{1,0}M$, obtained as the quotient of the ambient $(0,q)$-forms by those involving the boundary differential. The operator ${\bar{\partial }}_b:\Gamma ({\Lambda }^{0,q}M)\to \Gamma ({\Lambda }^{0,q+1}M)$ is the induced differential, and ${\bar{\partial }}_b^2=0$. The resulting complex $$ 0 \to \Gamma(\Lambda^{0,0}M) \xrightarrow{\bar\partial_b} \Gamma(\Lambda^{0,1}M) \xrightarrow{\bar\partial_b} \cdots \xrightarrow{\bar\partial_b} \Gamma(\Lambda^{0,n-1}M) \to 0 $$ is the tangential Cauchy-Riemann complex, or ${\bar{\partial }}_b$-complex, with cohomology $H_b^q(M)$.

Counterexamples to common slips

• The CR structure is not the restriction of the ambient complex structure to $M$. There is no complex structure on $M$ itself for $n\ge 2$: the real dimension $2n-1$ is odd. Only the rank-$(n-1)$ subbundle $HM$ is $J$-invariant; the leftover real direction (the characteristic, or Reeb, direction) carries no complex partner on $M$.
• ${\bar{\partial }}_b^2=0$ does not make the complex elliptic. The operator ${\bar{\partial }}_b$ degenerates in the characteristic direction, so the box operator ${\square }_b={\bar{\partial }}_b{\bar{\partial }}_b^{\ast }+{\bar{\partial }}_b^{\ast }{\bar{\partial }}_b$ is not elliptic — it is only subelliptic, and only when the Levi form is non-degenerate. This single fact is the source of every difficulty below, including the Lewy phenomenon.
• A CR function need not extend holomorphically to either side. Whether ${\bar{\partial }}_{b}u=0$ forces extension depends on the sign of the Levi form; the equation alone is necessary, not sufficient, for being a boundary value.

Key theorem with proof Intermediate+

Theorem (Lewy 1957; local non-solvability of ${\bar{\partial }}_b$). Let $M=\{\mathrm{Im}w=|z{|}^2\}\subset {\mathbb{C}}^2$ be the model hypersurface, with the tangential operator written in real coordinates $(x,y,t)$ on $M$ as $$ \bar L ;=; \frac{\partial}{\partial x} - i\frac{\partial}{\partial y} - 2 i,(x + i y),\frac{\partial}{\partial t} . $$ There exists a smooth complex-valued function $f\in C^{\infty }(M)$ such that the equation $\bar{L}u=f$ has no solution $u$ of class $C^1$ in any neighbourhood of any point. Equivalently, the inhomogeneous tangential Cauchy-Riemann equation ${\bar{\partial }}_{b}u=f$ is not locally solvable.

Proof. The argument follows Lewy ^{[Lewy 1957]} and the streamlined account in Hörmander ^{[Hörmander §2.6]}.

Step 1 — reduction to an ordinary differential equation in $t$. Suppose $\bar{L}u=f$ held on a neighbourhood, with $f=f(t)$ depending only on the characteristic variable $t$ (independent of $x,y$). For each fixed small radius $r$ define the circle average $$ U(t, r) ;=; \int_0^{2\pi} u\big(r\cos\theta,, r\sin\theta,, t\big), d\theta . $$ Integrating the equation $\bar{L}u=f(t)$ around the circle of radius $r$ in the $(x,y)$-plane and using that ${\partial }_x-i{\partial }_y$ in polar form is $e^{i\theta }({\partial }_r+\frac{i}{r}{\partial }_{\theta })$, the angular derivative integrates to zero and the radial-plus-characteristic part collects into an exact relation between $\partial U/\partial r$ and $\partial U/\partial t$.

Step 2 — the holomorphy obstruction. Setting $s=r^2$ and $V(t,s)=U(t,r)$, the averaged equation becomes the single Cauchy-Riemann-type equation $$ \frac{\partial V}{\partial s} + i,\frac{\partial V}{\partial t} ;=; g(t), $$ where $g$ is a fixed multiple of $f$, so $V$ is a holomorphic function of the complex variable $s+it$ up to the explicit inhomogeneous term, on the half-disc $\{s>0\}$, continuous up to $\{s=0\}$. The boundary value of $V$ at $s=0$ is $2\pi u(0,0,t)$, a $C^1$ function of $t$ alone.

Step 3 — choose $f$ to violate the boundary regularity. A function holomorphic on a half-disc and continuous to the edge has boundary values whose Cauchy transform is constrained: the data $g(t)$ must be the boundary value of a holomorphic function. Choose $f(t)$ — equivalently $g(t)$ — to be a smooth function whose associated holomorphic extension to $\{s>0\}$ does not exist as a continuous function up to $s=0$; concretely, take $g$ smooth but not real-analytic, with a Fourier transform supported so that the one-sided extension blows up. Then no $C^1$ boundary value $u(0,0,t)$ can exist, contradicting the assumed solvability. Hence $\bar{L}u=f$ has no $C^1$ solution near the origin. Because the construction is translation-stable in $t$ and the origin was arbitrary up to the CR automorphisms of $M$, non-solvability holds at every point. $\square$

Bridge. This non-solvability builds toward the entire subelliptic theory of ${\bar{\partial }}_b$, and the foundational reason is already visible here: the operator $\bar{L}$ has no characteristic direction along which it is elliptic, so its solvability is governed not by the principal symbol but by the Levi form, the bracket $[\bar{L},L]=\frac{1}{i}[{\partial }_x-i{\partial }_y,{\partial }_x+i{\partial }_y]+\cdots$ landing in the characteristic direction ${\partial }_t$. This is exactly the mechanism that the $\bar{\partial }$-Neumann problem 06.10.10 tames from the interior: the same Levi form that here obstructs tangential solvability is the form whose positivity yields the subelliptic estimate for the box operator. The Lewy example generalises the elementary fact that not every smooth right-hand side is a boundary value of a holomorphic function, and putting these together, the central insight is that the tangential complex is a boundary trace of the interior $\bar{\partial }$-theory, solvable precisely where the interior problem is well-posed. The Levi form is dual, on the boundary, to the curvature that controls $\bar{\partial }$ inside; this is the bridge from the Lewy obstruction to Kohn's Hodge theory for ${\square }_b$.

Exercises Intermediate+

Advanced results Master

The tangential CR complex is the boundary trace of the interior Dolbeault theory, and its analysis organises around the Levi form and the subelliptic estimates it permits.

Kohn's ${\square }_b$ and the subelliptic estimate. Kohn introduced the boundary box operator ${\square }_b={\bar{\partial }}_b{\bar{\partial }}_b^{\ast }+{\bar{\partial }}_b^{\ast }{\bar{\partial }}_b$ on $(0,q)$-forms. Because ${\bar{\partial }}_b$ degenerates in the characteristic direction, ${\square }_b$ is not elliptic; instead, when the Levi form has at least $\max (q+1,n-q)$ eigenvalues of the same sign (Kohn's condition $Y(q)$), ${\square }_b$ satisfies a subelliptic estimate $\Vert u{\Vert }_{1/2}^2\lesssim ({\square }_{b}u,u)+\Vert u{\Vert }_0^2$ with a gain of $1/2$ derivative. The estimate yields a Hodge decomposition $L_{(0,q)}^2(M)=\mathrm{im}{\bar{\partial }}_b\oplus \mathrm{im}{\bar{\partial }}_b^{\ast }\oplus {\mathcal{H}}_b^q$ with finite-dimensional harmonic space ${\mathcal{H}}_b^q\cong H_b^q(M)$, and local solvability of ${\bar{\partial }}_b$ in that degree. The Lewy example is exactly the failure of $Y(0)$ on a strongly pseudoconvex three-dimensional CR manifold: $n=2$, $q=0$, only one Levi eigenvalue, $\max (1,2)=2>1$.

CR extension and the Kohn-Rossi theorem. On the boundary $M=\partial \Omega$ of a smoothly bounded strongly pseudoconvex domain in ${\mathbb{C}}^n$, $n\ge 2$, every CR function (and more generally every ${\bar{\partial }}_b$-closed $(0,q)$-form for $q<n-1$) extends. Kohn-Rossi ^{[Kohn-Rossi 1965]} proved that a CR function on $M$ is the boundary value of a function holomorphic in $\Omega$, and that the tangential cohomology $H_b^q(M)$ vanishes for $0<q<n-1$ and is infinite-dimensional at the top degree $q=n-1$. This is the one-sided extension that the Lewy obstruction shows is not free: extension holds toward the pseudoconvex side, fails toward the other, and the difference is the sign of the Levi form. The Hans Lewy extension theorem is the local version: a CR function near a point where the Levi form has a positive eigenvalue extends holomorphically to the corresponding side.

Embeddability: Boutet de Monvel and the Rossi counterexample. An abstract CR manifold need not embed in any ${\mathbb{C}}^N$ as a real hypersurface. Boutet de Monvel ^{[Boutet de Monvel 1975]} proved that a compact strongly pseudoconvex CR manifold of dimension $2n-1\ge 5$ is globally embeddable, by showing ${\square }_b$ has closed range and constructing enough CR functions to separate points. The dimension hypothesis is sharp: in dimension $3$ ($n=2$) there are compact strongly pseudoconvex CR structures, due to Rossi and to Andreotti-Siu, that admit no non-constant global CR functions and hence no embedding. The three-dimensional case — the dimension of the Lewy hypersurface — is exactly where both global embeddability and bottom-degree local solvability break down, a structural coincidence governed by the deficit in condition $Y(0)$.

The Levi form as the universal control. Across all these results one Hermitian form decides everything: the Levi form ${\mathcal{L}}_p(L,\bar{L})=⟨[L,\bar{L}],\xi ⟩$, where $\xi$ is the characteristic conormal. Its definiteness is strong pseudoconvexity; its signature governs which cohomology groups vanish, which equations are solvable, whether the manifold embeds, and on which side CR functions extend. The Lewy operator is the minimal model in which this form is non-degenerate but the same-sign-eigenvalue count is too small at the bottom degree, isolating the obstruction in its purest form.

Synthesis. The tangential CR complex is the foundational reason the boundary of a domain in ${\mathbb{C}}^n$ carries its own function theory, and putting the pieces together shows that one Hermitian invariant — the Levi form — governs the whole apparatus. This is dual to the interior $\bar{\partial }$-Neumann theory: the subelliptic gain that the Levi form grants ${\square }_b$ on the boundary is exactly the boundary shadow of the gain it grants the interior box operator, and the central insight is that local solvability of ${\bar{\partial }}_b$ at degree $q$, vanishing of $H_b^q$, global embeddability, and one-sided CR extension are four faces of Kohn's condition $Y(q)$. The Lewy example generalises beyond a single curious equation: it is the precise point where $Y(0)$ fails in the lowest dimension, and this is exactly why dimension three resists both embedding and bottom-degree solvability. The bridge from the obstruction to the positive theory is that the very Levi form that blocks solvability at degree $0$ on a strongly pseudoconvex hypersurface is the one that forces solvability and finite cohomology at the intermediate degrees, so the Lewy phenomenon and the Kohn-Rossi extension theorem are a single statement read at two ends of the same complex.

Full proof set Master

Proposition (independence of ${\bar{\partial }}_b$ from the extension). Let $M=\{\rho =0\}$ be a smooth real hypersurface and $u\in C^{\infty }(M)$. If ${\tilde{u}}_1,{\tilde{u}}_2$ are two smooth extensions of $u$ to a neighbourhood in ${\mathbb{C}}^n$, then $(\bar{\partial }{\tilde{u}}_1){|}_{T^{0,1}M}=(\bar{\partial }{\tilde{u}}_2){|}_{T^{0,1}M}$.

Proof. The difference $h={\tilde{u}}_1-{\tilde{u}}_2$ vanishes on $M=\{\rho =0\}$, so by the smooth division lemma $h=a\rho$ for some smooth $a$. Then $\bar{\partial }h=(\bar{\partial }a)\rho +a\bar{\partial }\rho$. Restricting to a point $p\in M$, the first term carries the factor $\rho (p)=0$. For the second, evaluate on a vector $\bar{L}\in T_p^{0,1}M$: by definition $T_p^{0,1}M\subseteq T_{\mathbb{C}}M$ consists of anti-holomorphic vectors tangent to $M$, and any tangent vector annihilates $\rho$, so $⟨\bar{\partial }\rho ,\bar{L}⟩=\bar{L}\rho =0$. Hence $⟨\bar{\partial }h,\bar{L}⟩=0$ for every $\bar{L}\in T_p^{0,1}M$, giving $(\bar{\partial }{\tilde{u}}_1-\bar{\partial }{\tilde{u}}_2){|}_{T^{0,1}M}=0$. $\square$

Proposition (${\bar{\partial }}_b^2=0$). On a smooth embedded CR hypersurface the induced operator satisfies ${\bar{\partial }}_b\circ {\bar{\partial }}_b=0$.

Proof. Work with the ambient $\bar{\partial }$ on ${\mathbb{C}}^n$, which satisfies ${\bar{\partial }}^2=0$. The bundle ${\Lambda }^{0,q}M$ is the quotient of the ambient $(0,q)$-forms restricted to $M$ by the ideal generated by $\bar{\partial }\rho$ (the forms with a normal anti-holomorphic factor), and ${\bar{\partial }}_b$ is the induced map on the quotient. The ideal $(\bar{\partial }\rho )$ is preserved by $\bar{\partial }$ modulo $\rho$: $\bar{\partial }(\bar{\partial }\rho \wedge \beta )=-\bar{\partial }\rho \wedge \bar{\partial }\beta$, again in the ideal. Therefore $\bar{\partial }$ descends to the quotient complex, and the descended operator inherits ${\bar{\partial }}^2=0$. Concretely, for the frame $\{{\bar{L}}_k\}$ of $T^{0,1}M$, integrability $[{\bar{L}}_j,{\bar{L}}_k]={\sum }_{\ell }{c}_{jk}^{\ell }{\bar{L}}_{\ell }$ (no characteristic component, because the CR structure is integrable) makes the symbol-level Koszul differential close, so ${\bar{\partial }}_b^2=0$ holds on every form degree. $\square$

Proposition (the Lewy field has non-degenerate Levi form). For $\bar{L}={\partial }_{\bar{z}}-iz{\partial }_t$ on the model $M\subset {\mathbb{C}}^2$, the Levi form is non-degenerate; the model boundary is strongly pseudoconvex.

Proof. Set $L=\overline{\bar{L}}={\partial }_z+i\bar{z}{\partial }_t$. Since the coordinate fields commute, the only surviving terms in $[\bar{L},L]$ are those where a derivative hits the variable coefficient of the other field: $$ [\bar L, L] = \big(\partial_{\bar z}(i\bar z)\big)\partial_t - \big(\partial_z(-iz)\big)\partial_t = i\partial_t + i\partial_t = 2i,\partial_t . $$ The characteristic conormal is $\xi =dt$ (the conormal to the contact distribution $\mathrm{span}\{L,\bar{L}\}$). The Levi form value is $\mathcal{L}(L,\bar{L})=\frac{1}{2i}⟨[\bar{L},L],\xi ⟩=\frac{1}{2i}\cdot 2i=1\ne 0$. A single non-zero eigenvalue is a non-degenerate (indeed definite) Levi form, so $M$ is strongly pseudoconvex. $\square$

Theorem (local non-solvability, restated). The equation $\bar{L}u=f$ on the model $M$ has, for suitable smooth $f$, no $C^1$ solution in any neighbourhood of any point.

Proof. The Key-theorem argument applies on the packaged inputs: rotational averaging (Step 1) reduces the equation to the inhomogeneous Cauchy-Riemann equation ${\partial }_{s}V+i{\partial }_{t}V=g$ on a half-disc (Step 2), whose continuous boundary values are constrained to be boundary traces of a one-sided holomorphic function. Selecting $g$ smooth but lacking a continuous one-sided holomorphic extension (Step 3) produces an $f$ for which no $C^1$ solution exists. Translation invariance in $t$ and the transitivity of the CR automorphism group of the model on $M$ propagate the obstruction to every point. The strong pseudoconvexity established in the previous Proposition is what makes the averaged equation one-sided rather than two-sided, which is the operative feature; an indefinite Levi form would average to a two-sided problem and solvability could be restored. $\square$

Connections Master

• Pseudoconvexity and the Levi form 06.10.03. The Levi form defined there as the boundary obstruction to holomorphic convexity is exactly the Hermitian form $\mathcal{L}(L,\bar{L})=⟨[L,\bar{L}],\xi ⟩$ that governs the tangential complex here. Strong pseudoconvexity (definite Levi form) is simultaneously the condition for the $\bar{\partial }$-Neumann subelliptic estimate inside and the condition that makes the Lewy operator one-sided and hence non-solvable at the bottom degree. The same invariant reads as convexity inside and as solvability obstruction on the boundary.

• Bochner-Martinelli kernel and formula 06.10.06. The Bochner-Martinelli formula represents a $C^1$ function by a boundary integral plus an interior $\bar{\partial }$-defect; restricting its boundary term to $\partial \Omega$ produces the integral characterisation of CR boundary values (the Bochner-Severi moment conditions), which are the weak form of ${\bar{\partial }}_{b}f=0$. The kernel-theoretic proof of CR extension toward the pseudoconvex side is the explicit-integral counterpart of the Kohn-Rossi Hilbert-space proof referenced in this unit.

• The $\bar{\partial }$-Neumann problem 06.10.10. Kohn's interior $\bar{\partial }$-Neumann operator $N$ and the boundary box operator ${\square }_b$ are two faces of one elliptic-boundary-value problem: the subelliptic $1/2$-estimate for $N$ on strongly pseudoconvex domains restricts on the boundary to the subelliptic estimate for ${\square }_b$ under condition $Y(q)$. The non-solvability of ${\bar{\partial }}_b$ at degree $0$ here is precisely the boundary signature of where the interior problem's estimate degenerates, making this unit the boundary companion to that interior theory.

• Szegő kernel and Fefferman boundary asymptotics 06.10.09. The Szegő projection onto boundary CR functions is the orthogonal projection killed by ${\bar{\partial }}_b$, and its singularity along the boundary diagonal is the Heisenberg-type singularity dictated by the non-degenerate Levi form computed in this unit. The Boutet de Monvel-Sjöstrand parametrix for the Szegő kernel is built from the same ${\bar{\partial }}_b$-microlocal analysis whose failure at degree $0$ is the Lewy phenomenon.

Historical & philosophical context Master

Hans Lewy's 1957 note An example of a smooth linear partial differential equation without solution ^{[Lewy 1957]} (Ann. of Math. (2) 66, 155-158) is among the most consequential three pages in twentieth-century analysis. Before it, the prevailing expectation — encouraged by the Cauchy-Kovalevskaya theorem and by Malgrange-Ehrenpreis solvability for constant-coefficient operators — was that a smooth linear partial differential equation with smooth right-hand side should be locally solvable. Lewy exhibited a first-order operator with smooth (even polynomial) coefficients and a smooth right-hand side for which no solution exists in any neighbourhood of any point. The operator was not contrived: it is the tangential Cauchy-Riemann operator on the boundary of a strongly pseudoconvex domain in ${\mathbb{C}}^2$, so the obstruction is geometric, not pathological. The discovery launched the general theory of local solvability, culminating in the Nirenberg-Trèves condition $(\Psi )$ characterising exactly which principal-type operators are locally solvable.

The geometric reading was supplied immediately by the several-complex-variables community. J. J. Kohn and Hugo Rossi in 1965 ^{[Kohn-Rossi 1965]} (Ann. of Math. (2) 81, 451-472) recast the operator as the bottom of the tangential CR complex and proved the boundary extension theorem, showing that the Lewy obstruction is the degree-zero failure of a complex whose higher cohomology vanishes — the same Levi positivity blocks solvability at the bottom and enforces it in the middle. Kohn's subsequent Hodge theory for ${\square }_b$ and the embeddability theorem of Louis Boutet de Monvel in 1975 ^{[Boutet de Monvel 1975]} (Séminaire Goulaouic-Lions-Schwartz) placed the Lewy example as the sharp boundary case of a structural dichotomy: dimension three, where condition $Y(0)$ fails and Rossi's non-embeddable CR structures live, is exactly where Lewy's equation refuses to be solved. The example thereby sits at the confluence of partial differential equations, CR geometry, and the boundary behaviour of holomorphic functions, and its descendants run from microlocal analysis to the Fefferman program in CR geometry.

Bibliography Master

@article{Lewy1957,
  author  = {Lewy, Hans},
  title   = {An example of a smooth linear partial differential equation without solution},
  journal = {Ann. of Math. (2)},
  volume  = {66},
  year    = {1957},
  pages   = {155--158}
}

@article{KohnRossi1965,
  author  = {Kohn, J. J. and Rossi, Hugo},
  title   = {On the extension of holomorphic functions from the boundary of a complex manifold},
  journal = {Ann. of Math. (2)},
  volume  = {81},
  year    = {1965},
  pages   = {451--472}
}

@incollection{BoutetdeMonvel1975,
  author    = {Boutet de Monvel, Louis},
  title     = {Int{\'e}gration des {\'e}quations de {C}auchy-{R}iemann induites formelles},
  booktitle = {S{\'e}minaire Goulaouic-Lions-Schwartz 1974--1975},
  publisher = {{\'E}cole Polytechnique},
  year      = {1975},
  note      = {Exp. No. 9}
}

@book{Boggess1991,
  author    = {Boggess, Albert},
  title     = {CR Manifolds and the Tangential Cauchy-Riemann Complex},
  series    = {Studies in Advanced Mathematics},
  publisher = {CRC Press},
  year      = {1991}
}

@book{ChenShaw2001,
  author    = {Chen, So-Chin and Shaw, Mei-Chi},
  title     = {Partial Differential Equations in Several Complex Variables},
  series    = {AMS/IP Studies in Advanced Mathematics},
  volume    = {19},
  publisher = {American Mathematical Society and International Press},
  year      = {2001}
}

@book{HormanderSCV1990,
  author    = {H{\"o}rmander, Lars},
  title     = {An Introduction to Complex Analysis in Several Variables},
  edition   = {3rd},
  publisher = {North-Holland},
  year      = {1990}
}

@article{KohnSubelliptic1965,
  author  = {Kohn, J. J.},
  title   = {Boundaries of complex manifolds},
  journal = {Proc. Conf. Complex Analysis (Minneapolis, 1964), Springer},
  year    = {1965},
  pages   = {81--94}
}

@article{NirenbergTreves1970,
  author  = {Nirenberg, Louis and Tr{\`e}ves, Fran{\c{c}}ois},
  title   = {On local solvability of linear partial differential equations},
  journal = {Comm. Pure Appl. Math.},
  volume  = {23},
  year    = {1970},
  pages   = {1--38, 459--509}
}