06.10.05 · riemann-surfaces / several-variables

Solution of the Levi problem

shipped3 tiersLean: none

Anchor (Master): Oka 1953 / Bremermann 1954 / Norguet 1954 (originators); Krantz Ch. 4; Hörmander §4.2; Range Ch. IV

Intuition Beginner

A region in complex space is called a domain of holomorphy when it is the natural home of some holomorphic function: a function that lives on the region and refuses to extend past any part of its wall. The previous chapter showed that such regions are special. In one variable every open region qualifies, but in several variables most regions leak — a holomorphic function on them automatically spreads to a larger region, so they are not the true home of anything.

Eugenio Levi noticed, around 1910, that the wall of a leak-free region has to curve the right way. His condition is local: stand at any boundary point and check how the wall bends in the complex directions tangent to it. A region passing this test everywhere is called pseudoconvex. Levi proved that every domain of holomorphy passes his test. The reverse question stayed open for forty years: if a region passes the local curvature test at every wall point, must it be a genuine domain of holomorphy?

The answer is yes. That positive answer is the Levi problem, and its resolution is the structural centrepiece of several-variable theory.

Visual Beginner

Picture a region whose boundary wall, viewed near one point, bulges inward like the inside of a bowl when you look only along the complex directions. Pseudoconvexity is exactly this one-sided bulge, checked at every point of the wall and in every complex tangent direction at once. A flat wall, or one that dimples outward, fails the test and lets functions leak through.

The Levi problem asks you to manufacture, from this purely local curvature data, a single global holomorphic function that blows up at a chosen wall point and so cannot be continued past it. The picture below shows the strategy. Near the target point you write down an easy local function that already blows up there. It is not yet holomorphic on the whole region, so it has a small defect. A correction term, supplied by solving a standard auxiliary equation, cancels the defect everywhere while leaving the blow-up intact.

The barrier that makes the correction possible is the height $- lo g d$ , where $d$ is the distance to the wall. On a pseudoconvex region this height curves upward along every complex line, and that upward curve is precisely the budget that pays for the correction.

Worked example Beginner

Take the simplest several-variable region: the ball $B$ of radius $1$ in two complex variables, the set of points $(z_{1}, z_{2})$ with $∣ z_{1} ∣^{2} + ∣ z_{2} ∣^{2} < 1$ . Its wall is the sphere where $∣ z_{1} ∣^{2} + ∣ z_{2} ∣^{2} = 1$ . Is this region a domain of holomorphy?

First, the local test. The wall is the level set of $ρ (z) = ∣ z_{1} ∣^{2} + ∣ z_{2} ∣^{2} - 1$ , and this $ρ$ curves upward in every complex direction — it is the squared distance to the origin, shifted down by one. So the ball passes Levi's curvature test at every wall point. It is pseudoconvex.

Now build a function that blows up at a chosen wall point, say $p = (1, 0)$ . Try $f (z) = 1/ (1 - z_{1})$ . As the point slides toward $p$ , the denominator $1 - z_{1}$ goes to $0$ , so $f$ shoots to infinity exactly at $p$ . And $f$ is already holomorphic on the whole ball, because $1 - z_{1}$ never vanishes there: if $z_{1} = 1$ then $∣ z_{1} ∣^{2} = 1$ and the point is on the wall, not inside. No correction term is even needed here.

What this tells us: the ball is a domain of holomorphy, certified by an honest function blowing up at its boundary, and the only ingredient was the upward curvature of the wall. For more complicated pseudoconvex regions the blow-up function is harder to write by hand, and the correction term in the picture does the work that $1 - z_{1}$ did for free here.

Check your understanding Beginner

Formal definition Intermediate+

Throughout, $Ω \subseteq C^{n}$ is open and connected. Two predicates from the foundations of this chapter are in play.

From 06.10.01, $Ω$ is a domain of holomorphy if there is no pair of nonempty open sets $Ω_{1} \subseteq Ω_{2}$ with $Ω_{1} \subseteq Ω \cap Ω_{2}$ , $Ω_{2} \neq \subseteq Ω$ , and $Ω_{2}$ connected, such that every $f \in O (Ω)$ has a holomorphic continuation to $Ω_{2}$ agreeing with $f$ on $Ω_{1}$ . Equivalently (Cartan–Thullen), $Ω$ is holomorphically convex: for every compact $K ⋐ Ω$ the hull $$ \hat K_\Omega = { z \in \Omega : |f(z)| \le \sup_K |f| \ \text{for all } f \in \mathcal{O}(\Omega)} $$ is again relatively compact in $Ω$ .

From 06.10.02, a function $φ : Ω \to [- \infty, \infty)$ is plurisubharmonic (PSH) if it is upper semicontinuous and subharmonic on every complex line. A continuous plurisubharmonic exhaustion is a continuous PSH $φ : Ω \to R$ whose sublevel sets $Ω_{c} = {z : φ (z) < c}$ are relatively compact in $Ω$ for all $c$ .

Definition (pseudoconvex domain). $Ω \subseteq C^{n}$ is pseudoconvex if it admits a continuous plurisubharmonic exhaustion function. Equivalently $- lo g d (z, \partial Ω)$ is plurisubharmonic on $Ω$ (Hartogs pseudoconvexity), and — for $Ω$ with $C^{2}$ boundary defined by $ρ$ — the Levi form $$ L_\rho(p; w) = \sum_{j,k=1}^n \frac{\partial^2 \rho}{\partial z_j , \partial \bar z_k}(p), w_j \bar w_k \ \ge 0 \qquad \text{for all } w \text{ with } \sum_j \frac{\partial \rho}{\partial z_j}(p), w_j = 0, $$ the inequality holding on the complex tangent space at every boundary point $p$ . The equivalence of these three formulations is developed in 06.10.03.

The Levi problem. The implication domain of holomorphy $\Rightarrow$ pseudoconvex is the elementary direction: if $Ω$ is a domain of holomorphy then $- lo g d (\cdot, \partial Ω)$ is PSH, proved by a Hartogs-figure / Kontinuitätssatz argument. The Levi problem is the converse, $$ \Omega \ \text{pseudoconvex} \implies \Omega \ \text{is a domain of holomorphy}, $$ equivalently pseudoconvex $\Rightarrow$ holomorphically convex. Its resolution closes the equivalence chain $$ \text{domain of holomorphy} \iff \text{holomorphically convex} \iff \text{pseudoconvex}. $$

The strategy is separation by a single function: to a boundary point $p_{0}$ one attaches an $f \in O (Ω)$ unbounded along a sequence approaching $p_{0}$ . The existence of such $f$ for every $p_{0}$ forces $\hat{K}_{Ω} ⋐ Ω$ , hence holomorphic convexity. The obstruction to writing $f$ down directly is the overdetermined nature of the Cauchy–Riemann system; the tool that removes it is the solvability of $\overset{ˉ}{\partial} u = g$ with control, the several-variable Hörmander theory built in 06.10.04 over the dimension-one $\overset{ˉ}{\partial}$ of 06.04.05.

Key theorem with proof Intermediate+

Theorem (solution of the Levi problem). Let $Ω \subseteq C^{n}$ be pseudoconvex. Then $Ω$ is a domain of holomorphy; equivalently $Ω$ is holomorphically convex. ^{[Krantz Ch. 4]}

The proof takes for granted one analytic input, stated here as a hypothesis and proved in 06.10.04.

$L^{2}$ existence hypothesis (Hörmander). Let $Ω$ be pseudoconvex and $φ \in C^{2} (Ω)$ strictly plurisubharmonic with Levi form bounded below by $c (z) > 0$ . For every $(0, 1)$ -form $g$ with $\overset{ˉ}{\partial} g = 0$ and $\int_{Ω} ∣ g ∣^{2} e^{- φ} d V < \infty$ there is $u \in L_{loc}^{2} (Ω)$ with $\overset{ˉ}{\partial} u = g$ and $$ \int_\Omega |u|^2 e^{-\varphi}, dV \ \le \ \int_\Omega \frac{|g|^2}{c}, e^{-\varphi}, dV . $$ ^{[Hörmander 1965]}

Proof of Theorem. It suffices to prove holomorphic convexity, since by Cartan–Thullen 06.10.01 that is equivalent to being a domain of holomorphy. Fix a boundary point $p_{0} \in \partial Ω$ (or the point at infinity if $Ω$ is unbounded) and a sequence $z_{ν} \in Ω$ with $z_{ν} \to p_{0}$ . It is enough to produce $f \in O (Ω)$ with $∣ f (z_{ν}) ∣ \to \infty$ : then $p_{0} \in / \hat{K}_{Ω}$ for any compact $K$ with $z_{ν} \in / K$ eventually, so $\hat{K}_{Ω}$ has no limit point on $\partial Ω$ and is relatively compact.

Step 1: a local peak. Choose a holomorphic local separating function near $p_{0}$ . By pseudoconvexity one may take a strictly PSH exhaustion $ψ$ and a defining geometry in which there is a holomorphic polynomial $h$ on a neighbourhood $U$ of $p_{0}$ with $h (p_{0}) = 0$ and $Re h < 0$ on $U \cap Ω ∖ {p_{0}}$ to second order — the Levi polynomial of a smooth strictly pseudoconvex defining function, $h (z) = \sum_{j} ρ_{z_{j}} (p_{0}) (z_{j} - p_{0, j}) + \frac{1}{2} \sum_{j, k} ρ_{z_{j} z_{k}} (p_{0}) (z_{j} - p_{0, j}) (z_{k} - p_{0, k})$ . Then $s = 1/ h$ is holomorphic on $U \cap Ω$ and $∣ s ∣ \to \infty$ as $z \to p_{0}$ . (For a general pseudoconvex $Ω$ one first exhausts by smooth strictly pseudoconvex subdomains, by the smooth strictly PSH approximation of 06.10.02, and runs the construction on each.)

Step 2: cut off. Let $χ \in C_{c}^{\infty} (U)$ equal $1$ near $p_{0}$ . The function $χs$ is smooth on $Ω$ and agrees with $s$ near $p_{0}$ , but it is not holomorphic: $g := \overset{ˉ}{\partial} (χs) = (\overset{ˉ}{\partial} χ) s$ . Because $\overset{ˉ}{\partial} χ$ is supported in the annular region where $χ$ transitions — bounded away from $p_{0}$ — and $s$ is bounded there, $g$ is a smooth $\overset{ˉ}{\partial}$ -closed $(0, 1)$ -form, supported off $p_{0}$ , with $g \in L^{2}$ against any weight.

Step 3: correct. Choose the weight $φ$ so heavily plurisubharmonic near $p_{0}$ that $$ \int_\Omega |g|^2 e^{-\varphi}, dV < \infty \quad\text{while}\quad e^{-\varphi} \ \text{is non-integrable near } p_0 . $$ Concretely add to a strictly PSH exhaustion a term $N lo g ∣ h ∣^{- 2}$ or $2 n lo g ∣ z - p_{0} ∣^{- 1} \cdot$ (a PSH bump), forcing a pole-strength singularity of the weight at $p_{0}$ ; since $g$ vanishes near $p_{0}$ this leaves $\int ∣ g ∣^{2} e^{- φ} < \infty$ . The $L^{2}$ existence hypothesis yields $u \in L_{loc}^{2}$ with $\overset{ˉ}{\partial} u = g$ and $\int_{Ω} ∣ u ∣^{2} e^{- φ} d V < \infty$ . The non-integrability of $e^{- φ}$ at $p_{0}$ forces $u (p_{0}) = 0$ in the strong sense that $u \to 0$ along $z_{ν}$ , because a function with $\int ∣ u ∣^{2} e^{- φ} < \infty$ cannot be bounded away from $0$ where $e^{- φ}$ blows up.

Step 4: assemble. Set $f = χs - u$ . Then $$ \bar\partial f = \bar\partial(\chi s) - \bar\partial u = g - g = 0, $$ so $f$ is holomorphic on $Ω$ (a distributional solution of $\overset{ˉ}{\partial} f = 0$ is holomorphic, by Weyl/elliptic regularity, 06.04.05). Near $p_{0}$ , $f = s - u$ , and along $z_{ν} \to p_{0}$ one has $∣ s (z_{ν}) ∣ \to \infty$ while $u (z_{ν}) \to 0$ ; hence $∣ f (z_{ν}) ∣ \to \infty$ . This is the separating function. Sweeping $p_{0}$ over $\partial Ω$ gives $\hat{K}_{Ω} ⋐ Ω$ for every compact $K$ , so $Ω$ is holomorphically convex and therefore a domain of holomorphy. $□$

Bridge. This proof builds toward every later theorem of the chapter that runs on the $\overset{ˉ}{\partial}$ machine, and the cut-off-and-correct pattern appears again in the Cousin problems and the integral-kernel solution operators. The foundational reason a purely local curvature condition produces a global function is the positivity in the $L^{2}$ estimate: this is exactly the Levi form of the weight $φ$ , the same Hermitian form whose nonnegativity defined plurisubharmonicity in 06.10.02, now carrying enough strict positivity to dominate the defect $g$ . Putting these together, the Levi problem identifies pseudoconvexity with holomorphic convexity — the geometric input with the function-theoretic output — and the bridge is the weighted solvability of $\overset{ˉ}{\partial} u = g$ , which converts a local peak $1/ h$ into a global holomorphic separator. The central insight is that the overdetermined Cauchy–Riemann system, the obstruction that made several-variable theory hard, becomes the instrument of its solution once one controls $\overset{ˉ}{\partial}$ in $L^{2}$ ; this generalises the dimension-one fact that any $\overset{ˉ}{\partial}$ -equation on a planar domain is solvable, which is why every planar region is a domain of holomorphy.

Exercises Intermediate+

Exercise 1 (easy, short-answer).

State precisely which direction of the equivalence "domain of holomorphy $⟺$ pseudoconvex" is the easy half and name the tool that proves it.

Hint

One direction is a Kontinuitätssatz / Hartogs-figure argument on $- lo g d (\cdot, \partial Ω)$ ; the other is the deep one.

Answer

The easy half is domain of holomorphy $\Rightarrow$ pseudoconvex: if $Ω$ is a domain of holomorphy then $- lo g d (z, \partial Ω)$ is plurisubharmonic, proved by the continuity theorem (Kontinuitätssatz) — a family of analytic discs whose boundaries stay in $Ω$ cannot have interiors escaping, which is the plurisubharmonicity of the boundary-distance height. The deep half, the converse, is the Levi problem proper. Rubric: full credit for naming the correct direction and the Kontinuitätssatz / boundary-distance PSH tool.

Exercise 3 (medium, short-answer).

In Step 3 of the Key-theorem proof the weight $φ$ is engineered to be non-integrable at $p_{0}$ . Explain precisely why this forces the correction $u$ to vanish at $p_{0}$ , and where the hypothesis that $g$ is supported away from $p_{0}$ is used.

Hint

A finite value $\int ∣ u ∣^{2} e^{- φ} d V$ near a point where $e^{- φ}$ diverges constrains $u$ ; and $\int ∣ g ∣^{2} e^{- φ}$ must stay finite for the existence theorem to apply.

Answer

The existence theorem returns $u$ with $\int_{Ω} ∣ u ∣^{2} e^{- φ} d V < \infty$ . If $e^{- φ}$ is non-integrable on every neighbourhood of $p_{0}$ , then any measurable $u$ with finite weighted norm must satisfy $lim inf_{z \to p_{0}} ∣ u (z) ∣ = 0$ — were $∣ u ∣ \geq ε > 0$ on a neighbourhood, the integral would inherit the divergence of $\int e^{- φ}$ . After the elliptic regularity that makes $f$ (hence $u = χs - f$ ) holomorphic near $p_{0}$ , this $lim inf$ upgrades to a genuine limit $u (z_{ν}) \to 0$ . The support hypothesis on $g$ is exactly what keeps $\int ∣ g ∣^{2} e^{- φ} < \infty$ despite the singular weight: $g$ vanishes near $p_{0}$ , so the divergence of $e^{- φ}$ at $p_{0}$ never multiplies a nonzero $g$ . Rubric: full credit for the non-integrability argument forcing $u \to 0$ and for locating the support hypothesis in the finiteness of $\int ∣ g ∣^{2} e^{- φ}$ .

Exercise 4 (medium, short-answer).

Show that an increasing union $Ω = ⋃_{ν} Ω_{ν}$ of domains of holomorphy with $Ω_{ν} ⋐ Ω_{ν + 1}$ is pseudoconvex, and explain why this lets the Levi-problem proof reduce a general pseudoconvex $Ω$ to bounded strictly pseudoconvex pieces.

Hint

$- lo g d (\cdot, \partial Ω) = sup_{ν} (- lo g d (\cdot, \partial Ω_{ν}))$ is an increasing limit; and each domain of holomorphy is pseudoconvex.

Answer

Each $Ω_{ν}$ is pseudoconvex (domain of holomorphy $\Rightarrow$ pseudoconvex), so $- lo g d (\cdot, \partial Ω_{ν})$ is PSH. On compacts, $- lo g d (\cdot, \partial Ω)$ is the increasing limit of $- lo g d (\cdot, \partial Ω_{ν})$ , and an increasing limit of PSH functions that stays upper semicontinuous is PSH; hence $Ω$ is pseudoconvex. Conversely, every pseudoconvex $Ω$ is the increasing union of its strictly-pseudoconvex sublevel sets ${φ < c}$ of a smooth strictly PSH exhaustion (regularise the exhaustion by 06.10.02 and add $ε ∣ z ∣^{2}$ ). The Behnke–Stein theorem then propagates the domain-of-holomorphy property from the pieces to the union, so it suffices to solve the Levi problem on bounded strictly pseudoconvex domains. Rubric: full credit for the increasing-limit PSH argument and the reduction to bounded strictly pseudoconvex exhausting pieces.

Exercise 5 (medium, short-answer).

Give the analytic-polyhedron / Oka–Weil consequence: deduce from holomorphic convexity that on a domain of holomorphy $Ω$ , every $f$ holomorphic on a neighbourhood of a compact $K$ with $K = \hat{K}_{Ω}$ is a uniform limit on $K$ of functions in $O (Ω)$ .

Hint

When $K$ equals its hull, $K$ is cut out by finitely many ${∣ f_{i} ∣ \leq 1}$ up to $ε$ ; expand in the corresponding polydisc coordinates.

Answer

Since $K = \hat{K}_{Ω}$ , for any neighbourhood $V \supseteq K$ there are $f_{1}, \dots, f_{m} \in O (Ω)$ with $K \subseteq P = {z \in Ω : ∣ f_{i} (z) ∣ < 1} ⋐ V$ , an analytic polyhedron. The map $z \mapsto (z, f_{1} (z), \dots, f_{m} (z))$ embeds a neighbourhood of $\overline{P}$ into a polydisc in $C^{n + m}$ as a closed submanifold; a function holomorphic near $K$ pulls back from a polydisc neighbourhood, where it has a convergent power series, and the partial sums — polynomials in $z$ and the $f_{i}$ , hence elements of $O (Ω)$ — converge uniformly on $K$ . This is the Oka–Weil theorem; it is the approximation companion of the Levi-problem solution. Rubric: full credit for the analytic-polyhedron embedding and the polynomial-in- $(z, f_{i})$ approximation.

Exercise 6 (hard, short-answer).

Prove the globalisation step: assuming $\overset{ˉ}{\partial} u = g$ is solvable in $L^{2} (Ω, e^{- φ})$ for every smooth strictly PSH weight $φ$ and every $\overset{ˉ}{\partial}$ -closed $g \in L^{2}$ , show that for each boundary point and each approach sequence there is a single $F \in O (Ω)$ unbounded on a dense set of $\partial Ω$ simultaneously.

Hint

Take a countable dense set of boundary points, build a separating $f_{k}$ for each with $sup_{K_{k}} ∣ f_{k} ∣$ small, and sum a rapidly weighted series.

Answer

Let ${p_{k}}$ be a countable dense subset of $\partial Ω$ and let $K_{1} ⋐ K_{2} ⋐ \dots$ exhaust $Ω$ by compacts. For each $k$ the Key theorem gives $f_{k} \in O (Ω)$ with $∣ f_{k} ∣ \to \infty$ along a sequence approaching $p_{k}$ ; rescaling and subtracting a constant, arrange $sup_{K_{k}} ∣ f_{k} ∣ \leq 2^{- k}$ and pick points $w_{k} \to p_{k}$ with $∣ f_{k} (w_{k}) ∣ \geq k + \sum_{j < k} ∣ f_{j} (w_{k}) ∣$ . Set $F = \sum_{k} f_{k}$ . The bound $sup_{K_{k}} ∣ f_{k} ∣ \leq 2^{- k}$ makes the tail $\sum_{j \geq k} f_{j}$ converge uniformly on $K_{k - 1}$ , so $F \in O (Ω)$ . At $w_{k}$ , $∣ F (w_{k}) ∣ \geq ∣ f_{k} (w_{k}) ∣ - \sum_{j < k} ∣ f_{j} (w_{k}) ∣ - \sum_{j > k} ∣ f_{j} (w_{k}) ∣ \geq k - 1$ , so $F$ is unbounded near every $p_{k}$ , hence near a dense subset of $\partial Ω$ . Such an $F$ has $Ω$ as its exact domain of holomorphy. Rubric: full credit for the weighted dense-summation construction and the lower bound forcing unboundedness at each $p_{k}$ .

Exercise 7 (hard, short-answer).

The Hartogs triangle $T = {(z_{1}, z_{2}) : ∣ z_{1} ∣ < ∣ z_{2} ∣ < 1}$ is a domain of holomorphy, yet it has a boundary point where no smooth defining function exists. Reconcile this with the Levi-form definition of pseudoconvexity, and state which characterisation of pseudoconvexity certifies $T$ .

Hint

The Levi-form test needs $C^{2}$ boundary; the PSH-exhaustion test does not.

Answer

The Levi-form condition presupposes a $C^{2}$ defining function, which fails at the boundary point $(0, 0)$ of $T$ where the boundary is not a smooth hypersurface. The correct, fully general certificate is the PSH-exhaustion (Hartogs) characterisation, which makes no smoothness assumption: $T$ admits the continuous PSH exhaustion $φ (z) = max (- lo g (∣ z_{2} ∣ - ∣ z_{1} ∣), - lo g (1 - ∣ z_{2} ∣), ∣ z ∣^{2})$ , whose sublevel sets are relatively compact in $T$ and which is PSH as a maximum of PSH functions. So $T$ is pseudoconvex by the exhaustion criterion, and by the Levi-problem theorem it is a domain of holomorphy — consistent with the classical fact that holomorphic functions on $T$ do not all extend across $(0, 0)$ . The example shows why the analytic (exhaustion) formulation, not the geometric (Levi-form) one, is the load-bearing definition in the proof. Rubric: full credit for locating the failure in the $C^{2}$ hypothesis and naming the PSH-exhaustion certificate.

Advanced results Master

The Levi-problem theorem is the analytic core of a wider equivalence and the seed of two structural generalisations: Grauert's solution on complex manifolds and the sheaf-theoretic reformulation through Cartan's Theorems A and B.

The full equivalence and its proof economy. On $Ω \subseteq C^{n}$ the four conditions — domain of holomorphy; holomorphic convexity; existence of a continuous PSH exhaustion (pseudoconvexity); and, for $C^{2}$ boundary, Levi pseudoconvexity — are equivalent. Three of the implications are elementary: holomorphic convexity $\Rightarrow$ domain of holomorphy is Cartan–Thullen 06.10.01; domain of holomorphy $\Rightarrow$ pseudoconvex is the Kontinuitätssatz; Levi pseudoconvex $\Rightarrow$ PSH-exhaustible uses the Levi form of $- lo g d$ . The single deep implication is pseudoconvex $\Rightarrow$ holomorphically convex, the Key theorem, and the whole weight of the subject rests on it. The economy of the $L^{2}$ proof is that it never constructs the separating function explicitly; it constructs only the correction $u$ , abstractly, from a positivity estimate, and lets $χs - u$ inherit holomorphy from $\overset{ˉ}{\partial}$ -closedness.

Strict pseudoconvexity and the Levi polynomial. When $\partial Ω$ is $C^{2}$ and the Levi form is strictly positive, the Levi polynomial $h_{p_{0}}$ of Step 1 is a genuine local holomorphic support function: $Re h_{p_{0}} (z) \leq - c ∣ z - p_{0} ∣^{2}$ on $\partial Ω$ near $p_{0}$ , so $1/ h_{p_{0}}$ peaks at $p_{0}$ from the start and the correction $u$ is needed only to globalise, not to create, the singularity. This is the regime where the integral-kernel methods of Henkin and Ramirez later supply an explicit solution operator for $\overset{ˉ}{\partial}$ with sup-norm and Hölder estimates, replacing the abstract Hilbert-space $u$ by a kernel integral. The weakly pseudoconvex case (Levi form merely $\geq 0$ , e.g. ${∣ z_{1} ∣^{2} + ∣ z_{2} ∣^{4} < 1}$ ) has no holomorphic support function in general, and the $L^{2}$ method is then not a convenience but a necessity.

Grauert's theorem and Stein manifolds. Grauert (1958) extended the solution from domains in $C^{n}$ to complex manifolds: a complex manifold carrying a strictly PSH exhaustion is Stein, and the Levi problem on manifolds asks whether a holomorphically convex manifold with no compact positive-dimensional analytic subsets is Stein. Grauert's affirmative answer runs the same $\overset{ˉ}{\partial}$ engine, now with cohomology vanishing $H^{q} (X, F) = 0$ for $q \geq 1$ and coherent $F$ on Stein $X$ standing in for the single-function separation. The dimension-one shadow of this is in 06.09.02: every open Riemann surface is Stein, and Theorems A and B hold there.

Bremermann's and Norguet's routes. Oka's 1953 proof, Bremermann's and Norguet's independent 1954 proofs, and the later Hörmander $L^{2}$ proof differ in the analytic engine but share the architecture: reduce to local peak functions, solve a $\overset{ˉ}{\partial}$ -type problem with a plurisubharmonic budget, globalise. Oka used his own coherence and the "Oka principle" of trading analytic for topological data; Hörmander mechanised the budget into the single weighted inequality used in the Key theorem. The kernel route (Grauert–Lieb, Henkin) and the $\overset{ˉ}{\partial}$ -Neumann route (Kohn) give the regularity refinements that the abstract $L^{2}$ solution alone does not.

Synthesis. The solution of the Levi problem is the foundational reason several-variable complex analysis is a structural theory rather than a catalogue of examples: it identifies a purely local boundary-curvature condition with a global function-theoretic one, and the identification is mediated entirely by the solvability of $\overset{ˉ}{\partial}$ against a plurisubharmonic weight. Putting these together, the three pictures that 06.10.01, 06.10.02, and 06.10.03 developed in parallel — holomorphic convexity, plurisubharmonic exhaustion, Levi-form positivity — collapse onto one another, and the central insight is that the overdetermined Cauchy–Riemann operator, which made the subject hard, is exactly the operator whose controlled invertibility makes it tractable. This is dual to the Cartan–Thullen picture: where Cartan–Thullen reads holomorphic convexity off the radius of convergence of power series, the Levi solution reads it off the Levi form of a weight, and the two readings agree because both are measuring the same plurisubharmonic budget. The pattern recurs across the rest of the chapter: the Cousin problems, the Oka–Weil approximation theorem, and the integral-kernel constructions are all instances of "solve $\overset{ˉ}{\partial}$ with a PSH weight, then globalise," and each builds toward the boundary-regularity theory of strongly pseudoconvex domains where the abstract solution is sharpened to an explicit kernel.

Full proof set Master

Proposition 1 (easy direction via the Kontinuitätssatz). If $Ω \subseteq C^{n}$ is a domain of holomorphy, then $φ = - lo g d (\cdot, \partial Ω)$ is plurisubharmonic, so $Ω$ is pseudoconvex.

Proof. Fix a complex line $λ \mapsto a + λb$ and a closed disc $\overline{D}$ in its parameter on which $a + λb \in Ω$ . Write $g (λ) = φ (a + λb) = - lo g d (a + λb, \partial Ω)$ ; subharmonicity of $g$ is what must be shown. Suppose not: then there is a harmonic $H$ on $D$ , continuous up to $\partial D$ , with $g \leq H$ on $\partial D$ but $g (λ_{0}) > H (λ_{0})$ for some interior $λ_{0}$ . Let $h$ be a holomorphic function on $D$ with $Re h = H$ . The family of analytic discs $ζ \mapsto a + λb + ζ e^{- h (λ)} v$ , for unit vectors $v$ and $∣ ζ ∣$ small, has boundaries (as $λ$ runs over $\partial D$ ) staying inside $Ω$ because $∣ e^{- h (λ)} ∣ = e^{- H (λ)} \leq e^{- g (λ)} = d (a + λb, \partial Ω)$ keeps the perturbed point within the boundary distance. By the continuity theorem for domains of holomorphy (a Hartogs figure cannot be a domain of holomorphy), the centres $a + λb$ over the interior also satisfy the distance bound $d \geq e^{- H}$ , i.e. $g \leq H$ on $D$ , contradicting $g (λ_{0}) > H (λ_{0})$ . Hence $g$ is subharmonic, $φ$ is PSH, and $Ω$ is pseudoconvex. $□$

Proposition 2 (separation forces holomorphic convexity). Let $Ω \subseteq C^{n}$ be open. Suppose that for every $p_{0} \in \partial Ω$ and every sequence $z_{ν} \to p_{0}$ in $Ω$ there is $f \in O (Ω)$ with $sup_{ν} ∣ f (z_{ν}) ∣ = \infty$ . Then $\hat{K}_{Ω} ⋐ Ω$ for every compact $K ⋐ Ω$ , i.e. $Ω$ is holomorphically convex.

Proof. The hull $\hat{K}_{Ω}$ is closed in $Ω$ and bounded (it lies in the polynomial hull of $K$ , which is compact when $Ω$ is bounded; the unbounded case is handled by also separating the point at infinity). It remains to show $\hat{K}_{Ω}$ has positive distance to $\partial Ω$ . If not, there is a sequence $w_{ν} \in \hat{K}_{Ω}$ with $w_{ν} \to p_{0} \in \partial Ω$ . By hypothesis pick $f \in O (Ω)$ with $∣ f (w_{ν}) ∣ \to \infty$ along a subsequence. But $w_{ν} \in \hat{K}_{Ω}$ means $∣ f (w_{ν}) ∣ \leq sup_{K} ∣ f ∣ < \infty$ for the fixed $f \in O (Ω)$ , a contradiction. Therefore $d (\hat{K}_{Ω}, \partial Ω) > 0$ , and being closed and bounded with positive boundary distance, $\hat{K}_{Ω}$ is relatively compact in $Ω$ . $□$

Proposition 3 (the singular weight kills the correction). Let $Ω$ be pseudoconvex, $p_{0} \in \partial Ω$ , and let $φ$ be a weight with $\int_{B (p_{0}, r) \cap Ω} e^{- φ} d V = \infty$ for all $r > 0$ . If $u \in L_{loc}^{2} (Ω)$ satisfies $\int_{Ω} ∣ u ∣^{2} e^{- φ} d V < \infty$ and $u$ is holomorphic on $B (p_{0}, r_{0}) \cap Ω$ , then $u (z) \to 0$ as $z \to p_{0}$ within $Ω$ .

Proof. Suppose, for contradiction, that there is $ε > 0$ and a sequence $z_{ν} \to p_{0}$ with $∣ u (z_{ν}) ∣ \geq ε$ . By holomorphy of $u$ near $p_{0}$ and the mean-value property, $∣ u ∣$ cannot oscillate faster than its sup over small balls; concretely there is $ρ_{ν} ↓ 0$ and $c > 0$ with $∣ u ∣ \geq ε /2$ on $B (z_{ν}, c ρ_{ν}) \cap Ω$ , balls of comparable size that accumulate at $p_{0}$ . Choosing a subsequence so these balls are disjoint and contained in $B (p_{0}, r) \cap Ω$ , $$ \int_\Omega |u|^2 e^{-\varphi}, dV \ \ge\ \sum_\nu \int_{B(z_\nu, c\rho_\nu)} \frac{\varepsilon^2}{4}, e^{-\varphi}, dV \ =\ \frac{\varepsilon^2}{4}\sum_\nu \int_{B(z_\nu, c\rho_\nu)} e^{-\varphi}, dV . $$ The balls fill a positive fraction of every neighbourhood of $p_{0}$ (a Whitney-type covering of the approach region), so the right side inherits the divergence $\int_{B (p_{0}, r) \cap Ω} e^{- φ} = \infty$ and the left side is forced to be infinite — contradicting $\int ∣ u ∣^{2} e^{- φ} < \infty$ . Hence no such $ε$ exists and $u (z) \to 0$ as $z \to p_{0}$ . $□$

Proposition 4 (holomorphic convexity is biholomorphically invariant; the conclusion transfers). If $Φ : Ω^{'} \to Ω$ is a biholomorphism and $Ω$ is a domain of holomorphy, then so is $Ω^{'}$ .

Proof. Pullback $f \mapsto f \circ Φ$ is a $C$ -algebra isomorphism $O (Ω) \to O (Ω^{'})$ commuting with sup-norms over corresponding compacts, since $Φ$ is a homeomorphism. For compact $K^{'} ⋐ Ω^{'}$ set $K = Φ (K^{'}) ⋐ Ω$ . For $z^{'} \in Ω^{'}$ and every $g = f \circ Φ \in O (Ω^{'})$ , $∣ g (z^{'}) ∣ = ∣ f (Φ (z^{'})) ∣$ and $sup_{K^{'}} ∣ g ∣ = sup_{K} ∣ f ∣$ , so $z^{'} \in \hat{K^{'}}_{Ω^{'}}$ iff $Φ (z^{'}) \in \hat{K}_{Ω}$ ; that is, $\hat{K^{'}}_{Ω^{'}} = Φ^{- 1} (\hat{K}_{Ω})$ . Since $Ω$ is a domain of holomorphy, $\hat{K}_{Ω} ⋐ Ω$ , and applying the homeomorphism $Φ^{- 1}$ to a relatively compact set gives $\hat{K^{'}}_{Ω^{'}} ⋐ Ω^{'}$ . Thus $Ω^{'}$ is holomorphically convex, hence a domain of holomorphy by Cartan–Thullen. The Levi-problem theorem is therefore a biholomorphic invariant, as pseudoconvexity is, consistent with the equivalence. $□$

Connections Master

Domains of holomorphy and holomorphic convexity 06.10.01. The Levi-problem theorem supplies the missing implication of the Cartan–Thullen circle: pseudoconvex $\Rightarrow$ holomorphically convex. Where 06.10.01 characterised domains of holomorphy through power-series radius equality and the convex hull $\hat{K}_{Ω}$ , this unit certifies that property from boundary geometry alone, closing the equivalence chain that organises the chapter.
Plurisubharmonic functions 06.10.02. The proof runs entirely on the PSH weight $φ$ and its strictly positive Levi form; the singular-weight construction of Step 3 and the kill-the-correction Proposition 3 are pure pluripotential theory. The boundary-distance height $- lo g d (\cdot, \partial Ω)$ introduced there is both the certificate of pseudoconvexity and the raw material of the exhaustion the proof exhausts by.
Pseudoconvexity and the Levi form 06.10.03. This unit is the structural payoff of the pseudoconvexity theory: it shows the local Levi-form condition is not merely necessary but sufficient for domain-of-holomorphy status. The Levi polynomial used in Step 1 and the strictly-pseudoconvex sharpening in Advanced results are read directly off the Levi form developed there.
The $\overset{ˉ}{\partial}$ problem with $L^{2}$ estimates 06.10.04. The Key theorem cites the Hörmander weighted existence theorem as its sole analytic input; that unit proves the input via the Bochner–Kodaira–Morrey–Kohn identity. The Levi problem is the first and most important application of the $L^{2}$ machine, and the smooth strictly-PSH-weight regularisation needed there is supplied by 06.10.02.
The one-variable $\overset{ˉ}{\partial}$ -Hilbert PDE 06.04.05. The elliptic regularity that turns the distributional solution $\overset{ˉ}{\partial} f = 0$ into a genuine holomorphic $f$ , and the dimension-one solvability that makes every planar region a domain of holomorphy, are the $n = 1$ ancestors of the several-variable argument. The several-variable theory is exactly the obstruction-laden generalisation of that solvable one-variable equation.

Historical & philosophical context Master

Eugenio Elia Levi, in a 1910 study of essential singularities of functions of two or more complex variables ^{[Krantz Ch. 4]}, isolated the second-order boundary condition now bearing his name and proved its necessity for a domain of holomorphy. The sufficiency — whether Levi's local condition forces the global property — became known as the Levi problem and resisted resolution for over four decades. Kiyoshi Oka announced and proved the affirmative answer for $C^{2}$ in the 1940s and for general $C^{n}$ in his ninth memoir ^{[Oka 1953]} (Japanese Journal of Mathematics 23, 97–155), working through his own coherence theory and the device of attaching to a pseudoconvex domain a sequence of analytic polyhedra. Independently and almost simultaneously, Hans-Joachim Bremermann ^{[Bremermann 1954]} (Mathematische Annalen 128, 63–91) and François Norguet ^{[Norguet 1954]} (Bulletin de la Société Mathématique de France 82, 137–159) gave their own proofs in 1954, and the three names are jointly attached to the theorem.

The proof presented here is not the original. It is the streamlined $L^{2}$ proof made possible by Lars Hörmander's 1965 existence theorem for the $\overset{ˉ}{\partial}$ operator ^{[Hörmander 1965]} (Acta Mathematica 113, 89–152), which replaced Oka's coherence-and-polyhedra machinery with a single weighted inequality of Bochner–Kodaira type. Hans Grauert had meanwhile, in 1958, transported the whole question to complex manifolds and solved the Levi problem there, identifying the strictly-PSH-exhaustible manifolds with the Stein manifolds and thereby fusing the analytic Levi problem with Henri Cartan's sheaf-cohomological Theorems A and B. The convergence of these routes — Oka's coherence, Hörmander's $L^{2}$ estimates, Grauert's bumping and Cartan's sheaves — is why a single boundary-curvature inequality came to control the entire global function theory of pseudoconvex domains.

Bibliography Master

@article{Oka1953,
  author  = {Oka, Kiyoshi},
  title   = {Sur les fonctions analytiques de plusieurs variables. IX.
             {Domaines} finis sans point critique int\'erieur},
  journal = {Jpn. J. Math.},
  volume  = {23},
  year    = {1953},
  pages   = {97--155}
}

@article{Bremermann1954,
  author  = {Bremermann, Hans J.},
  title   = {\"Uber die \"Aquivalenz der pseudokonvexen Gebiete und der
             Holomorphiegebiete im Raum von $n$ komplexen
             Ver\"anderlichen},
  journal = {Math. Ann.},
  volume  = {128},
  year    = {1954},
  pages   = {63--91}
}

@article{Norguet1954,
  author  = {Norguet, Fran\c{c}ois},
  title   = {Sur les domaines d'holomorphie des fonctions uniformes de
             plusieurs variables complexes (passage du local au global)},
  journal = {Bull. Soc. Math. France},
  volume  = {82},
  year    = {1954},
  pages   = {137--159}
}

@article{Hormander1965,
  author  = {H\"ormander, Lars},
  title   = {$L^2$ estimates and existence theorems for the
             $\bar\partial$ operator},
  journal = {Acta Math.},
  volume  = {113},
  year    = {1965},
  pages   = {89--152}
}

@article{Grauert1958,
  author  = {Grauert, Hans},
  title   = {On {Levi's} problem and the imbedding of real-analytic
             manifolds},
  journal = {Ann. of Math. (2)},
  volume  = {68},
  year    = {1958},
  pages   = {460--472}
}

@article{Levi1910,
  author  = {Levi, Eugenio Elia},
  title   = {Studii sui punti singolari essenziali delle funzioni
             analitiche di due o pi\`u variabili complesse},
  journal = {Ann. Mat. Pura Appl. (3)},
  volume  = {17},
  year    = {1910},
  pages   = {61--87}
}

@book{KrantzSCV,
  author    = {Krantz, Steven G.},
  title     = {Function Theory of Several Complex Variables},
  edition   = {2},
  series    = {AMS Chelsea Publishing},
  volume    = {340},
  publisher = {American Mathematical Society},
  year      = {2001}
}

@book{HormanderSCV,
  author    = {H\"ormander, Lars},
  title     = {An Introduction to Complex Analysis in Several Variables},
  edition   = {3},
  series    = {North-Holland Mathematical Library},
  volume    = {7},
  publisher = {North-Holland},
  year      = {1990}
}

Prerequisites

06.10.01
06.10.02
06.07.01
06.01.04

Tier anchors

beginner: Krantz *Function Theory of Several Complex Variables* Ch. 4, informal sections; Range *Holomorphic Functions and Integral Representations* Ch. IV intro
intermediate: Krantz *Function Theory of Several Complex Variables* §4.2; Hörmander *An Introduction to Complex Analysis in Several Variables* §4.2; Range Ch. IV
master: Oka 1953 / Bremermann 1954 / Norguet 1954 (originators); Krantz Ch. 4; Hörmander §4.2; Range Ch. IV

References

Oka 1953 — Sur les fonctions analytiques de plusieurs variables, IX: Domaines finis sans point critique intérieur · Japanese J. Math. 23, 97–155 (originator: pseudoconvex domains in C^n are domains of holomorphy)
Bremermann 1954 — Über die Äquivalenz der pseudokonvexen Gebiete und der Holomorphiegebiete im Raum von n komplexen Veränderlichen · Math. Ann. 128, 63–91 (independent solution of the Levi problem)
Norguet 1954 — Sur les domaines d'holomorphie des fonctions uniformes de plusieurs variables complexes · Bull. Soc. Math. France 82, 137–159 (independent solution of the Levi problem)
Hörmander 1965 — L^2 estimates and existence theorems for the dbar operator · Acta Math. 113, 89–152 (the L^2 method that mechanises the Levi-problem solution)
Krantz — Function Theory of Several Complex Variables · AMS Chelsea 340, 2nd ed., Ch. 4 (solution of the Levi problem via dbar)
Hörmander — An Introduction to Complex Analysis in Several Variables · North-Holland, 3rd ed. 1990, §4.2 (pseudoconvex domains are domains of holomorphy)
Range — Holomorphic Functions and Integral Representations in Several Complex Variables · Springer GTM 108, Ch. IV (the Levi problem and the Oka–Grauert solution)

Estimated time

beginner: 16m
intermediate: 42m
master: 70m