Cousin I/II and the Levi problem in ${\mathbb{C}}^n$

Intuition Beginner

Suppose you want to build a function on a region of ${\mathbb{C}}^n$ that has a prescribed bad behaviour — a prescribed set of poles, say — laid out one patch at a time. On each small patch you write down a local recipe: here the function should blow up like $1/z_1$, over there like $1/(z_1{z}_2)$. The patches overlap, and on the overlaps your two recipes must agree up to something harmless (a function with no poles). The question is whether all these local recipes can be stitched into a single function defined on the whole region.

This is the Cousin I problem. It is the several-variable version of a question one variable answers with Mittag-Leffler: can you always prescribe principal parts? In one variable on a nice domain the answer is yes. In several variables the answer depends on the shape of the region.

The companion Cousin II problem prescribes zeros and poles multiplicatively, the way the Weierstrass product builds an entire function with assigned zeros. Here a second kind of obstruction appears, and it is not about analysis at all — it is about the topology of the region, the holes it has in the sense of loops and surfaces.

The deep fact is that the regions where Cousin I always works are exactly the "natural" regions for holomorphic functions: the domains of holomorphy. The Levi problem, settled by showing pseudoconvex regions are domains of holomorphy, is what makes Cousin I solvable on every region a working analyst cares about.

Visual Beginner

A picture of a region in ${\mathbb{C}}^n$ covered by overlapping patches. On each patch a small label shows a prescribed singular recipe (a fraction blowing up along a line). On the overlaps, arrows mark that two recipes differ by a smooth, pole-free correction. A single large arrow points to one global function defined over the whole region, the stitched-together answer. A side panel shows a second region shaped like a thick ring, with a loop drawn through its hole, illustrating that the multiplicative Cousin II problem can be blocked by a topological hole even when the additive Cousin I problem is fine.

Worked example Beginner

Take the polydisc ${\Delta }^2=\{(z_1,z_2):|z_1|<1,|z_2|<1\}$ in ${\mathbb{C}}^2$. Cover it by two patches: $U_1$ where $z_1\ne 0$ is allowed to be a pole and $U_2$ a neighbourhood where you want no pole. On $U_1$ prescribe the recipe $f_1=1/z_1$. On $U_2$ prescribe $f_2=0$.

On the overlap, the difference $f_1-f_2=1/z_1$ has a pole along $z_1=0$, so the two recipes do not match by a pole-free correction across that line. The honest Cousin I datum instead asks for a global function whose only singular part along $z_1=0$ is $1/z_1$.

The answer is direct: the single function $f=1/z_1$ already works on all of ${\Delta }^2$ minus the line $z_1=0$, and its principal part along that line is exactly the prescribed $1/z_1$. So the global solution is $f=1/z_1$ itself.

What this tells us: on the polydisc, a domain of holomorphy, the prescribed-poles problem is solvable, and here the solution is the obvious one — the recipe was already globally meaningful. The content of the theory is that solvability holds even when no single recipe works everywhere, because the obstruction lives in a cohomology group that vanishes on such domains.

Check your understanding Beginner

Formal definition Intermediate+

Let $\Omega \subseteq {\mathbb{C}}^n$ be a domain. Write ${\mathcal{O}}_{\Omega }$ for the sheaf of holomorphic functions, ${\mathcal{M}}_{\Omega }$ for the sheaf of meromorphic functions, ${\mathcal{O}}_{\Omega }^{\ast }$ and ${\mathcal{M}}_{\Omega }^{\ast }$ for the multiplicative sheaves of nowhere-vanishing holomorphic and not-identically-zero meromorphic functions.

Cousin I data. A Cousin I datum is an open cover $\{U_i\}$ of $\Omega$ together with $f_i\in {\mathcal{M}}_{\Omega }(U_i)$ such that $f_i-f_j\in {\mathcal{O}}_{\Omega }(U_i\cap U_j)$ on every overlap. A solution is $f\in {\mathcal{M}}_{\Omega }(\Omega )$ with $f-f_i\in {\mathcal{O}}_{\Omega }(U_i)$ for every $i$. Equivalently, the datum is a global section of the principal-parts sheaf ${\mathcal{P}}_{\Omega }={\mathcal{M}}_{\Omega }/{\mathcal{O}}_{\Omega }$, and a solution is a meromorphic lift of that section.

Cousin II data. A Cousin II datum is a cover $\{U_i\}$ with $f_i\in {\mathcal{M}}_{\Omega }^{\ast }(U_i)$ such that $f_i/f_j\in {\mathcal{O}}_{\Omega }^{\ast }(U_i\cap U_j)$. A solution is $f\in {\mathcal{M}}_{\Omega }^{\ast }(\Omega )$ with $f/f_i\in {\mathcal{O}}_{\Omega }^{\ast }(U_i)$. Equivalently, the datum is a global section of the divisor sheaf ${\mathcal{D}}_{\Omega }={\mathcal{M}}_{\Omega }^{\ast }/{\mathcal{O}}_{\Omega }^{\ast }$, and a solution is a global meromorphic function realising that divisor.

The differences $g_{ij}=f_i-f_j$ of a Cousin I datum form a Čech 1-cocycle in ${\mathcal{O}}_{\Omega }$; the datum is solvable when its class in $H^1(\Omega ,{\mathcal{O}}_{\Omega })$ vanishes. The ratios $g_{ij}=f_i/f_j$ of a Cousin II datum form a multiplicative 1-cocycle in ${\mathcal{O}}_{\Omega }^{\ast }$; the datum is solvable when its class in $H^1(\Omega ,{\mathcal{O}}_{\Omega }^{\ast })=\mathrm{Pic}(\Omega )$ vanishes.

The exponential sheaf sequence

$$0\to \mathbb{Z}\stackrel{2\pi i}{\to }{\mathcal{O}}_{\Omega }\stackrel{\exp }{\to }{\mathcal{O}}_{\Omega }^{\ast }\to 0$$

yields the long-exact segment

$$H^1(\Omega ,{\mathcal{O}}_{\Omega })\to \mathrm{Pic}(\Omega )\stackrel{c_1}{\to }{H}^2(\Omega ,\mathbb{Z})\to H^2(\Omega ,{\mathcal{O}}_{\Omega }),$$

with $c_1$ the first Chern class. On a domain of holomorphy the outer analytic groups $H^1(\Omega ,{\mathcal{O}}_{\Omega })$ and $H^2(\Omega ,{\mathcal{O}}_{\Omega })$ vanish (Cartan's Theorem B), so $c_1:\mathrm{Pic}(\Omega )\to H^2(\Omega ,\mathbb{Z})$ is an isomorphism. This presentation follows Krantz ^{[Krantz Ch. 4]} and Hörmander ^{[Hörmander Ch. III]}.

Counterexamples to common slips

• ${\mathbb{C}}^2\setminus \{0\}$ is not a domain of holomorphy: Hartogs extension forces every holomorphic function to extend across the origin, and $H^1$ of the structure sheaf is non-zero, so Cousin I can fail there.
• Cousin II is not "Cousin I for the logarithm." The analytic part of the obstruction matches, but the integer class $c_1\in H^2(\Omega ,\mathbb{Z})$ is genuinely extra; ignoring it is the standard error.
• On the polydisc and the ball both problems are always solvable, so they are the wrong test cases for the topological obstruction; a domain with non-vanishing $H^2(\Omega ,\mathbb{Z})$ is needed to see Cousin II fail.

Key theorem with proof Intermediate+

Theorem (Oka 1937; Cousin I on a domain of holomorphy). Let $\Omega \subseteq {\mathbb{C}}^n$ be a domain of holomorphy. Then $H^1(\Omega ,{\mathcal{O}}_{\Omega })=0$, and consequently every Cousin I datum on $\Omega$ is solvable.

Proof. By the solution of the Levi problem 06.10.05, a domain of holomorphy is pseudoconvex, hence admits a $C^{\infty }$ strictly plurisubharmonic exhaustion $\phi$. The proof reduces solvability of Cousin I to the surjectivity of the $\bar{\partial }$ operator on $(0,1)$-forms, then invokes the Hörmander $L^2$ existence theorem 06.10.04.

Step 1 — Čech to $\bar{\partial }$. Let $\{U_i\}$ be the cover and $g_{ij}=f_i-f_j\in {\mathcal{O}}_{\Omega }(U_i\cap U_j)$ the cocycle. Choose a smooth partition of unity $\{{\chi }_i\}$ subordinate to $\{U_i\}$. Set, on each $U_i$,

$$h_i=\sum _k{\chi }_k{g}_{ki},g_{ki}=f_k-f_i,$$

a smooth function on $U_i$ (the sum extends by zero outside $U_k$). On an overlap $U_i\cap U_j$, the cocycle identity $g_{ki}-g_{kj}=g_{ji}$ gives $h_i-h_j={\sum }_k{\chi }_k(g_{ki}-g_{kj})=g_{ji}{\sum }_k{\chi }_k=g_{ji}$. Thus the smooth functions $h_i$ differ by the holomorphic cocycle, so $\bar{\partial }{h}_i=\bar{\partial }{h}_j$ on overlaps, and they patch to a global smooth $(0,1)$-form $\psi$ with $\psi {|}_{U_i}=\bar{\partial }{h}_i$. Since each $g_{ij}$ is holomorphic, $\psi$ is $\bar{\partial }$-closed.

Step 2 — solve $\bar{\partial }u=\psi$. The form $\psi$ is a $\bar{\partial }$-closed $(0,1)$-form. Because $\Omega$ is pseudoconvex with the exhaustion $\phi$, the Hörmander $L^2$ existence theorem 06.10.04 supplies $u\in L_{\mathrm{loc}}^2(\Omega )$ with $\bar{\partial }u=\psi$; elliptic regularity of $\bar{\partial }$ (the form $\psi$ is smooth) makes $u$ smooth.

Step 3 — assemble the meromorphic solution. Put $f=h_i-u$ on $U_i$. On $U_i$, $\bar{\partial }(h_i-u)=\bar{\partial }{h}_i-\bar{\partial }u=\psi {|}_{U_i}-\psi {|}_{U_i}=0$, so $h_i-u$ is holomorphic on the part of $U_i$ where $h_i$ is finite. The definition is consistent across overlaps because $h_i-h_j=g_{ji}$ is holomorphic and $u$ is global: $(h_i-u)-(h_j-u)=h_i-h_j=g_{ji}$, matching the prescribed jumps. Now $f$ has the same principal part as $f_i$ on $U_i$: indeed $f-f_i=(h_i-u)-f_i$, and $h_i={\sum }_k{\chi }_k(f_k-f_i)$ has principal part $-f_i{\sum }_k{\chi }_k=-f_i$ added to a pole-free combination of the $f_k$, while $u$ is holomorphic, so $f-f_i\in {\mathcal{O}}_{\Omega }(U_i)$. Therefore $f\in {\mathcal{M}}_{\Omega }(\Omega )$ solves the datum.

The cohomology statement $H^1(\Omega ,{\mathcal{O}}_{\Omega })=0$ is the same argument read at the level of the cocycle $\{g_{ij}\}$: it is a $\bar{\partial }$-coboundary, hence a Čech coboundary by the Dolbeault-Čech comparison. $\square$

Bridge. This proof builds toward the Cousin II obstruction analysed in the Advanced results, where the same $\bar{\partial }$-vanishing handles the analytic half but a topological residue survives. The foundational reason Cousin I is unconditional on a domain of holomorphy is that solvability is the vanishing of $H^1(\Omega ,{\mathcal{O}}_{\Omega })$, and that vanishing is exactly the $\bar{\partial }$-$L^2$ surjectivity proven on pseudoconvex domains — putting these together, the Levi problem and the Cousin problem are two faces of one theorem. The Čech-to-$\bar{\partial }$ passage used here generalises: it is the mechanism by which every sheaf-cohomological obstruction on a Stein domain dissolves into a PDE one, and it appears again in the Oka-Weil approximation and in the exponential-sequence analysis of Cousin II. The bridge is the partition-of-unity patching that turns a holomorphic cocycle into a smooth $\bar{\partial }$-closed form; this is exactly the construction by which the dimension-one Mittag-Leffler theorem 06.09.04 generalises to ${\mathbb{C}}^n$, and it is dual to the way the residue calculus controls the one-variable case.

Exercises Intermediate+

Lean formalization Intermediate+

Mathlib does not currently formalise meromorphic-function sheaves on domains in ${\mathbb{C}}^n$, sheaf cohomology of the structure sheaf, or the predicate of being a domain of holomorphy, so neither Cousin problem can be stated. A proposed signature, sketching the Cousin I target:

-- Sketch only; no current Mathlib coverage. See lean_mathlib_gap.
import Mathlib.Analysis.Analytic.Within
import Mathlib.Geometry.Manifold.Sheaf.Basic

namespace Codex.RiemannSurfaces.SeveralVariables

variable {n : ℕ} (Ω : Set (Fin n → ℂ)) (hΩ : IsDomainOfHolomorphy Ω)

-- Cousin I: on a domain of holomorphy the first cohomology of the
-- structure sheaf vanishes, so every additive Cousin datum lifts.
theorem cousinI_solvable
    (𝒟 : CousinIDatum Ω) :
    ∃ f : Meromorphic Ω, 𝒟.IsSolvedBy f := by
  -- reduce to H¹(Ω, 𝒪) = 0, then to ∂̄-surjectivity on (0,1)-forms,
  -- then to the Hörmander L² existence theorem on pseudoconvex Ω
  sorry

end Codex.RiemannSurfaces.SeveralVariables

The proof depends on names absent from Mathlib (the meromorphic and principal-parts sheaves, sheaf cohomology on a domain, the domain-of-holomorphy predicate, and the $\bar{\partial }$-$L^2$ surjectivity already missing per 06.10.04). Until those land, the unit ships lean_status: none.

Advanced results Master

The two Cousin problems split along the exponential sequence into an analytic part, common to both, and a topological part, exclusive to Cousin II. The analytic part is the vanishing $H^1(\Omega ,{\mathcal{O}}_{\Omega })=0$ on a domain of holomorphy, proven above through the $\bar{\partial }$-$L^2$ engine. The topological part is the image of $c_1:\mathrm{Pic}(\Omega )\to H^2(\Omega ,\mathbb{Z})$, which on a domain of holomorphy is an isomorphism, so a Cousin II datum is solvable precisely when its first Chern class vanishes in $H^2(\Omega ,\mathbb{Z})$.

The Oka principle. Oka 1939 ^{[Oka 1939]} discovered that on a domain of holomorphy — and Grauert later on every Stein manifold — the holomorphic Cousin II problem is solvable if and only if the underlying continuous (topological) problem is. Concretely: a holomorphic line bundle on a Stein manifold is holomorphically the identity element of $\mathrm{Pic}$ if and only if it is topologically the identity. The Cousin II obstruction is therefore not analytic at all; the analysis collapses entirely onto $c_1\in H^2(\Omega ,\mathbb{Z})$. Grauert 1958 extended this to vector bundles of all ranks — the Oka-Grauert principle — making the holomorphic and topological classifications of complex vector bundles on a Stein manifold coincide as bijections.

Comparison with dimension one. On a non-compact Riemann surface $X$ both obstructions vanish unconditionally: $H^1(X,{\mathcal{O}}_X)=0$ by Behnke-Stein, and $H^2(X,\mathbb{Z})=0$ because a connected non-compact $2$-manifold has vanishing top integer cohomology 06.09.04. This is why Cousin I and Cousin II are both always solvable in dimension one. The genuine several-variable phenomenon is that $H^2(\Omega ,\mathbb{Z})$ can be non-zero for a domain of holomorphy in ${\mathbb{C}}^n$, $n\ge 2$, so Cousin II acquires a genuine topological gate that has no one-variable analogue.

Oka-Weil approximation. The constructive backbone of Cousin I in ${\mathbb{C}}^n$ is the Oka-Weil theorem: on a domain of holomorphy $\Omega$, functions holomorphic near a holomorphically convex compactum $K={\hat{K}}_{\Omega }$ are uniform limits on $K$ of global holomorphic functions. This replaces the one-variable Runge convergence factors and lets the exhaustion-and-correction scheme converge. Oka-Weil itself is a corollary of the solved Levi problem 06.10.05: holomorphic convexity of $\Omega$, plus $\bar{\partial }$-solvability, gives the approximation. Approximation, Cousin I, and the Levi problem are three readings of the same vanishing theorem.

Failure on non-Stein domains. On ${\mathbb{C}}^2\setminus \{0\}$, Hartogs extension forces $\mathcal{O}({\mathbb{C}}^2\setminus \{0\})=\mathcal{O}({\mathbb{C}}^2)$, and $H^1({\mathbb{C}}^2\setminus \{0\},\mathcal{O})\ne 0$; a Cousin I datum prescribing a pole that does not extend is unsolvable. This is the cohomological signature of the domain not being Stein, and it is the precise point at which the equivalence chain — domain of holomorphy ⟺ holomorphically convex ⟺ pseudoconvex — breaks.

Synthesis. The Cousin problems in ${\mathbb{C}}^n$ are the cohomological shadow of the Levi problem: solvability of Cousin I is the vanishing $H^1(\Omega ,{\mathcal{O}}_{\Omega })=0$, which is exactly the $\bar{\partial }$-$L^2$ surjectivity that proves pseudoconvex domains are domains of holomorphy. This identifies analysis with topology in a sharp way — the Oka principle says the only obstruction to Cousin II on a Stein domain is the integer class $c_1\in H^2(\Omega ,\mathbb{Z})$, so the holomorphic classification of line bundles is the continuous one. Putting these together, the additive problem, the multiplicative problem, and Oka-Weil approximation are three readings of a single vanishing theorem, and the central insight is that the domain-of-holomorphy condition is what makes all three simultaneously solvable; the bridge is the $\bar{\partial }$ engine of 06.10.04, which dissolves each sheaf-cohomology obstruction into a PDE. The dimension-one theory 06.09.04 is the degenerate case where the topological gate is open for free, and the several-variable theory generalises it precisely by keeping that gate visible.

Full proof set Master

Proposition (Cousin II obstruction is exactly $c_1$ on a domain of holomorphy). Let $\Omega \subseteq {\mathbb{C}}^n$ be a domain of holomorphy. A Cousin II datum with associated line bundle $L\in \mathrm{Pic}(\Omega )$ is solvable if and only if its first Chern class $c_1(L)\in H^2(\Omega ,\mathbb{Z})$ vanishes.

Proof. A Cousin II datum $\{(U_i,f_i)\}$ has multiplicative cocycle $g_{ij}=f_i/f_j\in {\mathcal{O}}_{\Omega }^{\ast }(U_i\cap U_j)$, defining a class $[L]\in H^1(\Omega ,{\mathcal{O}}_{\Omega }^{\ast })=\mathrm{Pic}(\Omega )$. The datum is solvable if and only if $[L]=0$: a global solution $f$ gives $f/f_i\in {\mathcal{O}}_{\Omega }^{\ast }(U_i)$, a $0$-cochain splitting the cocycle as a coboundary, and conversely a splitting $f_i=h_{i}f$ with $h_i\in {\mathcal{O}}_{\Omega }^{\ast }(U_i)$ produces the global solution $f$.

Now read the exponential long exact sequence

$$H^1(\Omega ,{\mathcal{O}}_{\Omega })\stackrel{{\exp }_{\ast }}{\to }\mathrm{Pic}(\Omega )\stackrel{c_1}{\to }{H}^2(\Omega ,\mathbb{Z})\to H^2(\Omega ,{\mathcal{O}}_{\Omega }).$$

Because $\Omega$ is a domain of holomorphy, Cartan's Theorem B gives $H^q(\Omega ,{\mathcal{O}}_{\Omega })=0$ for all $q\ge 1$; in particular $H^1(\Omega ,{\mathcal{O}}_{\Omega })=0$ and $H^2(\Omega ,{\mathcal{O}}_{\Omega })=0$. Exactness then makes $c_1$ injective (its kernel is the image of $H^1(\Omega ,{\mathcal{O}}_{\Omega })=0$) and surjective (its cokernel injects into $H^2(\Omega ,{\mathcal{O}}_{\Omega })=0$). Hence $[L]=0$ in $\mathrm{Pic}(\Omega )$ if and only if $c_1([L])=0$ in $H^2(\Omega ,\mathbb{Z})$, which is the claim. $\square$

Proposition (Cousin I cocycle is a $\bar{\partial }$-coboundary on a pseudoconvex domain). Let $\Omega$ be pseudoconvex and $\{g_{ij}\}$ a Čech 1-cocycle in ${\mathcal{O}}_{\Omega }$ for a cover $\{U_i\}$. Then $\{g_{ij}\}$ is a Čech coboundary.

Proof. With a partition of unity $\{{\chi }_k\}$ subordinate to $\{U_i\}$, set $h_i={\sum }_k{\chi }_k{g}_{ki}$ as in the Key theorem. The smooth functions $h_i$ satisfy $h_i-h_j=g_{ji}$ on overlaps, so $\bar{\partial }{h}_i=\bar{\partial }{h}_j$ there and they glue to a global $\bar{\partial }$-closed $(0,1)$-form $\psi$. By the Hörmander $L^2$ existence theorem on pseudoconvex domains 06.10.04 there is $u\in C^{\infty }(\Omega )$ with $\bar{\partial }u=\psi$. Then $c_i:=h_i-u$ is holomorphic on $U_i$ (as $\bar{\partial }{c}_i=\psi {|}_{U_i}-\psi {|}_{U_i}=0$), and $c_i-c_j=h_i-h_j=g_{ji}$, exhibiting $\{g_{ij}\}$ as the coboundary $\{c_j-c_i\}$. $\square$

Corollary (Cousin I solvability). On a domain of holomorphy every Cousin I datum is solvable. A domain of holomorphy is pseudoconvex 06.10.05; apply the previous Proposition to the difference cocycle $g_{ij}=f_i-f_j$ to obtain holomorphic $c_i$ with $c_i-c_j=f_i-f_j$, whence $f:=f_i-c_i$ is independent of $i$ and meromorphic on $\Omega$ with the prescribed principal parts. $\square$

Corollary (Oka-Weil from holomorphic convexity). If $\Omega$ is a domain of holomorphy and $K={\hat{K}}_{\Omega }⋐\Omega$, every function holomorphic near $K$ is a uniform limit on $K$ of elements of $\mathcal{O}(\Omega )$. The holomorphically convex hull condition ${\hat{K}}_{\Omega }=K$ together with $\bar{\partial }$-solvability lets a local holomorphic function be corrected by a globally $\bar{\partial }$-solved term that is small on $K$; iterating over an exhaustion gives the uniform approximation. The full argument is Krantz ^{[Krantz Ch. 4]} and Hörmander ^{[Hörmander Ch. III]}. $\square$

Connections Master

• The $\bar{\partial }$ equation with Hörmander $L^2$ estimates 06.10.04. The single engine behind Cousin I: the additive obstruction $H^1(\Omega ,{\mathcal{O}}_{\Omega })=0$ is precisely the surjectivity of $\bar{\partial }$ on $(0,1)$-forms over a pseudoconvex domain. Every step of the Cousin I proof reduces to this existence theorem; the partition-of-unity passage from a holomorphic cocycle to a $\bar{\partial }$-closed form is the standard interface.

• Solution of the Levi problem 06.10.05. Cousin I solvability and the Levi problem are two readings of the same vanishing theorem. A domain of holomorphy is pseudoconvex (Levi), pseudoconvexity gives $\bar{\partial }$-$L^2$ solvability, and that solvability is the Cousin I vanishing. Oka-Weil approximation, used in the constructive Cousin I scheme, is itself a corollary of the solved Levi problem.

• Cousin I (additive) on a Riemann surface 06.09.04. The dimension-one ancestor. There the obstruction $H^1(X,{\mathcal{O}}_X)$ vanishes by Behnke-Stein on every non-compact surface and the topological gate $H^2(X,\mathbb{Z})=0$ is open for free. The ${\mathbb{C}}^n$ theory generalises this precisely by keeping the topological obstruction in $H^2(\Omega ,\mathbb{Z})$ visible for Cousin II.

• Cousin II (multiplicative) on a Riemann surface 06.09.05. The multiplicative dimension-one ancestor, where $\mathrm{Pic}(X)=0$ makes Cousin II unconditional. In ${\mathbb{C}}^n$ the exponential sequence keeps $c_1\in H^2(\Omega ,\mathbb{Z})$ as a genuine gate, and the Oka principle says it is the only one on a Stein domain.

• Sheaf cohomology 04.03.01. The entire Cousin framework lives in coherent sheaf cohomology: principal-parts and divisor sheaves, the exponential short exact sequence, the long exact sequence with the Chern-class connecting map, and Cartan's Theorem B vanishing on domains of holomorphy.

Historical & philosophical context Master

Pierre Cousin posed both problems in his 1895 thesis Sur les fonctions de $n$ variables complexes ^{[Cousin 1895]} (Acta Math. 19, 1–61), generalising the Mittag-Leffler prescribed-principal-parts problem and the Weierstrass prescribed-zeros problem from one variable to products of planar domains in ${\mathbb{C}}^n$. Cousin solved both on products of one-dimensional domains by iterating the one-variable constructions through the Cauchy integral. The subtlety that solvability depends on the shape of the domain — not merely its dimension — was not visible in Cousin's product setting and emerged only with the Hartogs phenomenon and the notion of a domain of holomorphy.

Kiyoshi Oka settled the general additive problem in 1937 ^{[Oka 1937]} (J. Sci. Hiroshima Univ. Ser. A, 7), proving Cousin I solvable on every domain of holomorphy, and in 1939 ^{[Oka 1939]} (J. Sci. Hiroshima Univ. Ser. A, 9) turned to Cousin II, where he found the decisive new phenomenon: a topological obstruction can block the multiplicative problem even on a domain of holomorphy. The same 1939 paper formulated the Oka principle — that the holomorphic Cousin II problem on a domain of holomorphy is solvable exactly when its continuous counterpart is — locating the entire residual difficulty in integer second cohomology. Henri Cartan's 1951–53 seminar recast Oka's results in the language of coherent sheaves and Theorem B, and Hans Grauert's 1958 Analytische Faserungen über holomorph-vollständigen Räumen (Math. Ann. 135) extended the Oka principle to vector bundles of arbitrary rank on Stein manifolds. Lars Hörmander's 1965 $L^2$ method ^{[Hörmander 1965]} (Acta Math. 113) supplied the PDE engine that mechanises the Cousin I vanishing on every pseudoconvex domain, the form in which Krantz ^{[Krantz Ch. 4]} presents the theory.

Bibliography Master

@article{Cousin1895,
  author  = {Cousin, Pierre},
  title   = {Sur les fonctions de $n$ variables complexes},
  journal = {Acta Math.},
  volume  = {19},
  year    = {1895},
  pages   = {1--61}
}

@article{Oka1937CousinI,
  author  = {Oka, Kiyoshi},
  title   = {Sur les fonctions analytiques de plusieurs variables. II. Domaines d'holomorphie},
  journal = {J. Sci. Hiroshima Univ. Ser. A},
  volume  = {7},
  year    = {1937},
  pages   = {115--130}
}

@article{Oka1939CousinII,
  author  = {Oka, Kiyoshi},
  title   = {Sur les fonctions analytiques de plusieurs variables. III. Deuxi{\`e}me probl{\`e}me de Cousin},
  journal = {J. Sci. Hiroshima Univ. Ser. A},
  volume  = {9},
  year    = {1939},
  pages   = {7--19}
}

@article{Grauert1958,
  author  = {Grauert, Hans},
  title   = {Analytische Faserungen {\"u}ber holomorph-vollst{\"a}ndigen R{\"a}umen},
  journal = {Math. Ann.},
  volume  = {135},
  year    = {1958},
  pages   = {263--273}
}

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

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

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

@book{GrauertRemmertStein,
  author    = {Grauert, Hans and Remmert, Reinhold},
  title     = {Theory of Stein Spaces},
  series    = {Grundlehren der mathematischen Wissenschaften},
  volume    = {236},
  publisher = {Springer},
  year      = {1979}
}