## On a Problem of Bremermann Concerning Runge Domains This work was supported by a Marie Curie International Reintegration Grant

### Cezar Joiţ a

Abstract
In this paper we give an example of a bounded Stein domain in ${\mathbb{C}}^{n}$  , with smooth boundary, which is not Runge and whose intersection with every complex line is simply connected.

1 Introduction

In [1Bremermann asked the following question:
”Suppose that $D$  is a Stein domain in ${\mathbb{C}}^{n}$  such that for every complex line $l$  in ${\mathbb{C}}^{n}$  , $l\D$  is connected. Is it true that $D$  is Runge in ${\mathbb{C}}^{n}$  ?” The question remained open and was mentioned again in a recent book by T. Ohsawa ([4, page 81). In this paper we will give a negative answer to Bremermann's question. Namely, we will give an example of a bounded, strictly pseudoconvex domain in ${\mathbb{C}}^{n}$  with real analytic boundary which is not Runge in ${\mathbb{C}}^{n}$  but whose intersection with every complex line is simply connected.
Note that if $D$  is bounded the hypothesis of the problem means simply that for every complex line $l$  , $l\cap D$  is Runge in $l$  . If, in addition, one requires that $l\cap D$  is connected as well then it does follow that $D$  is Runge. See for example [2, page 309, Theorem 4.7.8.
For simplicity our construction will be done in ${\mathbb{C}}^{2}$  but it can be easily adapted to ${\mathbb{C}}^{n}$  for $n\ge 2$  . To produce our example we will construct first a bounded, strictly pseudoconvex domain $W\subset {\mathbb{C}}^{2}$  with smooth, real analytic boundary which is Runge but its closure is not holomorphically convex. (Note that this is not possible in $\mathbb{C}$  .) Next we show that, in fact, we can construct $W$  as above and moreover it has the following geometric property: for every complex line $l$  the set of points where $l$  is tangent to $\partial W$  is at most finite. If this is the case, then one can show that $l\cap \overline{W}$  is polynomially convex, again for every complex line $l$  . Finally, we show that an appropriate neighborhood of $\overline{W}$  is a counterexample to Bremermann's problem.

2 The Example

The construction will be done in several steps.
First we prove that there exists a bounded domain in ${\mathbb{C}}^{2}$  with smooth, real analytic boundary which is strictly pseudoconvex, Runge in ${\mathbb{C}}^{2}$  , and its closure is not polynomially convex.
J. Wermer [5proved that there exists a biholomorphic map $F$  from a polydisc $P=\left\{\left({z}_{1},{z}_{2}\right)\in {\mathbb{C}}^{2}:|{z}_{1}|  in ${\mathbb{C}}^{2}$  onto an open set $F\left(P\right)$  of ${\mathbb{C}}^{2}$  such that $F\left(P\right)$  is not polynomially convex. (Wermer's original result was in ${\mathbb{C}}^{3}$  but it can be modified to hold in ${\mathbb{C}}^{2}$  as well; see [3or [4.) We start with such a map and let ${U}_{n}:=\left\{z\in {\mathbb{C}}^{2}:|\frac{{z}_{1}}{a}{|}^{n}+|\frac{{z}_{2}}{b}{|}^{n}<1\right\}$  . Since ${U}_{n}\subset {U}_{n+1}\subset P$  and $\cup {U}_{n}=P$  , it follows that there exists $m\in \mathbb{N}$  such that $F\left({U}_{m}\right)$  is not polynomially convex. Set $U:={U}_{m}$  and $V=F\left(U\right)$  . If we define $\phi :U\to \mathbb{R}$  by $\phi \left(z\right)=\frac{1}{1-|{z}_{1}/a{|}^{2m}-|{z}_{2}/b{|}^{2m}}+|\frac{{z}_{1}}{a}{|}^{2}+|\frac{{z}_{2}}{b}{|}^{2}-1$  then $\phi$  is a strictly plurisubharmonic real analytic function and has only one critical point. Since $F$  is a biholomorphism, $\phi \circ {F}^{-1}:V\to \mathbb{R}$  has the same properties and it is an exhaustion function for $V$  . For $\alpha >0$  let ${V}_{\alpha }=\left\{z\in V:\phi \circ {F}^{-1}\left(z\right)<\alpha \right\}$  . It follows that there exists $\alpha >0$  such that ${V}_{\alpha }$  is not polynomially convex. On the other hand, if ${z}_{0}=F\left(0\right)$  (this is the minimum point and the only critical point of $\phi \circ {F}^{-1}$  and $\phi \circ {F}^{-1}\left({z}_{0}\right)=0$  ) and we choose $B\subset V$  a ball centered at ${z}_{0}$  , then there exists $\alpha >0$  such that ${V}_{\alpha }\subset B$  . It follows that ${V}_{\alpha }$  is Runge in $B$  (because $\phi \circ {F}^{-1}$  is defined on $B$  ) and therefore is polynomially convex.
Put $r:=sup\left\{\alpha \in \mathbb{R}:{V}_{\alpha }\text{is polynomially convex}\right\}$  . From the above observations we deduce that $0  .
We claim that ${V}_{r}$  is the example that we are looking for. Indeed ${V}_{r}$  is Runge in ${\mathbb{C}}^{2}$  as an increasing union of Runge domains and it has smooth, real analytic boundary because $\phi \circ {F}^{-1}$  has no critical point on the $\partial {V}_{r}$  . We only need to convince ourselves that ${\overline{V}}_{r}=\left\{z\in V:\phi \circ {F}^{-1}\left(z\right)\le r\right\}$  is not polynomially convex. If ${\overline{V}}_{r}$  was polynomially convex then it would have a Runge (in ${\mathbb{C}}^{2}$  ) neighborhood $W$  with $W\subset V$  . If this was the case then for $\epsilon >0$  small enough ${V}_{r+\epsilon }\subset W$  and ${V}_{r+\epsilon }$  would be Runge in $W$  and therefore in ${\mathbb{C}}^{2}$  . This would contradict the choice of $r$  .
Let us refrase what we have done so far. We proved that if $V$  is a domain in ${\mathbb{C}}^{2}$  and $\phi :V\to \mathbb{R}$  is a strictly plurisubharmonic function such that there exist ${a}_{0}<{a}_{1}$  real numbers with the following properties: $\begin{array}{ccccc}\left\{x\in V:\phi \left(x\right)<{a}_{1}\right\}\subset \subset V,& \left\{x\in V:\phi \left(x\right)<{a}_{0}\right\}\text{is connected and contains}C\left(\phi \right):=\text{the set of}& \text{critical points of}\phi ,& \left\{x\in V:\phi \left(x\right)<{a}_{0}\right\}\text{is Runge in}{\mathbb{C}}^{2},& \left\{x\in V:\phi \left(x\right)<{a}_{1}\right\}\text{is not Runge in}{\mathbb{C}}^{2}\end{array}\right\}\left(*\right)$  Then there exists a unique real number $r=r\left(\phi \right)\in \left[{a}_{0},{a}_{1}\right)$  such that ${V}_{r\left(\phi \right)}:=\left\{x\in V:\phi \left(x\right)  is Runge and ${\overline{V}}_{r\left(\phi \right)}=\left\{x\in V:\phi \left(x\right)\le r\left(\phi \right)\right\}$  is not holomorphically convex. Note that ${V}_{r\left(\phi \right)}$  must be connected since each of its components contains a (minimum) critical point, ${V}_{r\left(\phi \right)}$  contains $\left\{x\in V:\phi \left(x\right)<{a}_{0}\right\}$  which is connected and $\left\{x\in V:\phi \left(x\right)<{a}_{0}\right\}\supset C\left(\phi \right)$  .
We also proved that there exists a real analytic function $\phi$  satisfying (*). We fix such a $\phi$  . Shrinking $V$  we can assume that $\overline{V}$  is compact and that $\phi$  is defined on a neighborhood of $\overline{V}$  .
Next we want to show that there exists $\psi$  , a small perturbation of $\phi$  , which satisfies (*) and in addition it has the following geometric property: for every complex line $l$  the set $\mathcal{T}\left(\psi ,l\right):=\left\{x\in {V}_{r\left(\psi \right)}\cap l:l\text{is tangent to}\partial {V}_{r\left(\psi \right)}\text{at}x\right\}$  is finite.
Indeed: let $U$  be an open and connected set such that $C\left(\phi \right)\subset U\subset \subset {V}_{r\left(\phi \right)}$  and let $W$  be an open and relatively compact neighborhood of $\partial {V}_{r\left(\phi \right)}$  and $0<\delta <{\delta }^{\prime }<{a}_{1}-r\left(\phi \right)$  two real numbers such that $U\subset \subset \left\{x\in V:\phi \left(x\right)  and $\left\{x\in V:r\left(\phi \right)-\delta <\phi \left(x\right)  .
If $\epsilon >0$  is small enough then, for every $\psi :V\to \mathbb{R}$  , a ${\mathcal{C}}^{\infty }$  function, if the $sup$  norms on $\overline{V}$  of $\psi -\phi$  , $\frac{\partial \left(\psi -\phi \right)}{\partial {x}_{j}}$  , $\frac{{\partial }^{2}\left(\psi -\phi \right)}{\partial {x}_{j}\partial {x}_{k}}$  , $j,k=1,...,n$  (here we denote ${z}_{j}={x}_{2j+1}+i{x}_{2j}$  ) are less than $\epsilon$  then $\psi$  is strictly plurisubharmonic and satisfies (*). Moreover $C\left(\psi \right)\subset U$  , $r\left(\psi \right)\in \left[r\left(\phi \right)-\delta ,r\left(\phi \right)+\delta \right]$  and $\partial \left\{x\in V:\psi   for every $s\in \left[r\left(\phi \right)-\delta ,r\left(\phi \right)+\delta \right]$  . We claim that there exists a real analytic $\psi$  such that for every complex line $l$  and for every $s\in \left[r\left(\phi \right)-\delta ,r\left(\phi \right)+\delta \right]$  the following set $\mathcal{T}\left(\psi ,l,s\right):=\left\{x\in \partial \left\{x\in V:\psi \left(x\right)  is finite. Indeed, if ${x}_{0}$  is not isolated in $\mathcal{T}\left(\psi ,l,s\right)$  and if we denote by $u:={\psi }_{|l}$  then ${x}_{0}$  is not isolated in $\left\{z\in l:u\left(z\right)=u\left({x}_{0}\right),\nabla u\left(z\right)=0\right\}$  . On the other hand $u$  is real analytic and strictly subharmonic. It follows that around ${x}_{0}$  at least one of the sets $\left\{z\in l\cap V:\frac{\partial u}{\partial x}\left(z\right)=0\right\}$  or $\left\{z\in l\cap V:\frac{\partial u}{\partial y}\left(z\right)=0\right\}$  is smooth and then the smooth one is contained in the other one. Hence there exists around ${x}_{0}$  a smooth real analytic curve $C$  such that ${u}_{|C}=u\left({x}_{0}\right)$  and $\nabla {u}_{|C}=0$  . If $\left\{f=0\right\}$  is a local equation for $C$  it follows that ${\psi }_{|C}=u=u\left({x}_{0}\right)+{f}^{2}g$  . However it is not difficult to see that this condition is not satisfied by a generic real analytic function $\psi$  . For example one notices that $detHess\left(u\right)\left({x}_{0}\right)=0$  and after a linear change of coordinates we can assume that $f\left(x\right)=x+$  higher order terms and $g\left({x}_{0}\right)=1$  . Then a straightforward computation shows that $u$  must satisfy the following four conditions at ${x}_{0}$  : $\frac{{\partial }^{3}u}{\partial {y}^{3}}=0$  , $\frac{1}{4!}\frac{{\partial }^{4}u}{\partial {y}^{4}}=\left[\frac{1}{4}\frac{{\partial }^{3}u}{\partial x\partial {y}^{2}}{\right]}^{2}$  , $\frac{1}{5!}\frac{{\partial }^{5}u}{\partial {y}^{5}}=\left[\frac{1}{3!}\frac{{\partial }^{4}u}{\partial x\partial {y}^{3}}-\frac{1}{8}\frac{{\partial }^{3}u}{\partial x\partial {y}^{2}}\frac{{\partial }^{3}u}{\partial {x}^{2}\partial y}\right]\frac{1}{4}\frac{{\partial }^{3}u}{\partial x\partial {y}^{2}}$  .
We fix now a $\psi$  which satisfies (*) and the geometrical property mentioned above.
Our next goal will be to show that, for every complex line $l$  in ${\mathbb{C}}^{2}$  , $l\cap {\overline{V}}_{r\left(\psi \right)}$  is polynomially convex (although ${\overline{V}}_{r\left(\psi \right)}$  is not). Note that $\left(l\cap {\overline{V}}_{r\left(\psi \right)}\right)\\overline{l\cap {V}_{r\left(\psi \right)}}$  is a finite set (as a subset of $\mathcal{T}\left(\psi ,l\right)$  ). Hence it suffices to show that $\overline{l\cap {V}_{r\left(\psi \right)}}$  is polynomially convex. Let's assume that it is not. Note that $l\cap {V}_{r\left(\psi \right)}$  is Runge in $l$  (since ${V}_{r\left(\psi \right)}$  is Runge in ${\mathbb{C}}^{2}$  ) and that it has a smooth boundary except at a finite set of points (the set of points of non-smoothness is also a subset of $\mathcal{T}\left(\psi ,l\right)$  ). As we assumed that $\overline{l\cap {V}_{r\left(\psi \right)}}$  is not polynomially convex it follows that there exists a rectifiable loop $\gamma$  in $l$  such that $\gamma \\left(l\cap {V}_{r\left(\psi \right)}\right)$  contains only points where the boundary of $l\cap {V}_{r\left(\psi \right)}$  in $l$  is not smooth and therefore is finite and $\stackrel{^}{\gamma }\cap \left(l\\left(\overline{l\cap {V}_{r\left(\psi \right)}}\right)\right)\ne \varnothing$  (in fact it has a nonempty interior). Using again the finiteness of $\mathcal{T}\left(\psi ,l\right)$  it follows that $\stackrel{^}{\gamma }\cap \left({\mathbb{C}}^{2}\{\overline{V}}_{r\left(\psi \right)}\right)\ne \varnothing$  . We claim that there exists a ${\mathcal{C}}^{\infty }$  family of biholomorphisms $\left\{{f}_{\epsilon }:{\mathbb{C}}^{2}\to {\mathbb{C}}^{2}{\right\}}_{\epsilon \in \mathbb{R}}$  such that ${f}_{0}$  is the identity and for $\epsilon >0$  small enough ${f}_{\epsilon }\left(\gamma \right)\subset {V}_{r\left(\psi \right)}$  . Without loss of generality we can assume that $l=\left\{z=\left({z}_{1},{z}_{2}\right)\in {\mathbb{C}}^{2}:{z}_{2}=0\right\}$  . We write $\gamma \\left(l\cap {V}_{r\left(\psi \right)}\right)=:\left\{\left({p}_{1},0\right),...,\left({p}_{s},0\right)\right\}$  and we denote by $\left(0,{q}_{1}\right),...\left(0,{q}_{s}\right)$  the unit inner normals to $\partial {V}_{r\left(\psi \right)}$  . We choose $h:\mathbb{C}\to \mathbb{C}$  a holomorphic function such that $h\left({p}_{j}\right)={q}_{j}$  and we define ${f}_{\epsilon }\left(z\right)=\left({z}_{1},{z}_{2}+\epsilon h\left({z}_{1}\right)\right)$  . It is obvious that ${f}_{\epsilon }$  are biholomorphisms and since $\frac{d{f}_{\epsilon }}{d\epsilon }\left({p}_{j},0\right)=\left(0,{q}_{j}\right)$  it follows that ${f}_{\epsilon }$  have the sought properties. Because $\stackrel{^}{\gamma }\cap \left({\mathbb{C}}^{2}\{\overline{V}}_{r\left(\psi \right)}\right)\ne \varnothing$  and $\left\{{f}_{\epsilon }\right\}$  is a continuous family we deduce that for $\epsilon$  small enough ${f}_{\epsilon }\left(\stackrel{^}{\gamma }\right)\not\subset {\overline{V}}_{r\left(\psi \right)}$  . On the other hand ${f}_{\epsilon }\left(\stackrel{^}{\gamma }\right)=\stackrel{^}{{f}_{\epsilon }\left(\gamma \right)}$  and ${f}_{\epsilon }\left(\gamma \right)\subset {V}_{r\left(\psi \right)}\cap {f}_{\epsilon }\left(l\right)$  . It follows from here that ${V}_{r\left(\psi \right)}\cap {f}_{\epsilon }\left(l\right)$  is not Runge in ${f}_{\epsilon }\left(l\right)$  which is a contradiction since ${V}_{r\left(\psi \right)}$  is Runge in ${\mathbb{C}}^{2}$  and ${f}_{\epsilon }\left(l\right)$  is a closed analytic submanifold in ${\mathbb{C}}^{2}$  .
We are now ready to produce our example. For $\epsilon >0$  we set ${W}_{\epsilon }:=\left\{x\in V:\psi \left(x\right)  . It follows from the definition of $r\left(\psi \right)$  that ${W}_{\epsilon }$  is not Runge in ${\mathbb{C}}^{2}$  . We wish to prove that there exists $\epsilon >0$  such that for every complex line $l$  , ${W}_{\epsilon }\cap l$  is Runge in $l$  .
Suppose that this is not the case. Then for $n\in \mathbb{N}$  large enough there exists a complex line ${l}_{n}$  such that ${W}_{\frac{1}{n}}\cap {l}_{n}$  is not Runge in ${l}_{n}$  .Note that $\left\{{l}_{n}\right\}$  is a sequence of lines that intersect a given compact subset of ${\mathbb{C}}^{2}$  . It contains then a convergent subsequence. By passing to this subsequence we can assume that $\left\{{l}_{n}\right\}$  converges to a line $l$  .
We already proved that $l\cap {\overline{V}}_{r\left(\psi \right)}$  is holomorphically convex and this implies that there exists $\Omega$  a Runge open subset of ${\mathbb{C}}^{2}$  such that $l\cap {\overline{V}}_{r\left(\psi \right)}\subset \Omega \subset V$  .
As $\cap {W}_{\frac{1}{n}}={\overline{V}}_{r\left(\psi \right)}$  and ${l}_{n}$  converges to $l$  we deduce that there exists ${n}_{0}\in \mathbb{N}$  such that for every $n\ge {n}_{0}$  ${W}_{\frac{1}{n}}\cap {l}_{n}\subset \Omega$  . Hence ${W}_{\frac{1}{n}}\cap {l}_{n}=\left({W}_{\frac{1}{n}}\cap \Omega \right)\cap {l}_{n}$  .
On the other hand, $\psi$  is a plurisubharmonic function defined on the whole $\Omega$  which is Stein and therefore ${W}_{\frac{1}{n}}\cap \Omega$  , which is a level set for ${\psi }_{|\Omega }$  , is Runge in $\Omega$  . Since $\Omega$  is Runge in ${\mathbb{C}}^{2}$  it follows that ${W}_{\frac{1}{n}}\cap \Omega$  is also Runge in ${\mathbb{C}}^{2}$  and from here we obtain that ${W}_{\frac{1}{n}}\cap {l}_{n}$  is Runge in ${l}_{n}$  . This contradicts our assumption.
In conclusion, we proved that for $\epsilon >0$  small enough ${W}_{\epsilon }$  is bounded, strictly pseudoconvex, is not Runge in ${\mathbb{C}}^{2}$  and for every complex line $l$  in ${\mathbb{C}}^{2}$  , ${W}_{\epsilon }\cap l$  is Runge in $l$  . In the same way as before ${W}_{\epsilon }$  must be connected since each of its components contains a critical point of $\psi$  .
$\mathbf{A}\mathbf{c}\mathbf{k}\mathbf{n}\mathbf{o}\mathbf{w}\mathbf{l}\mathbf{e}\mathbf{d}\mathbf{g}\mathbf{m}\mathbf{e}\mathbf{n}\mathbf{t}\mathbf{s}:$  I am very grateful to Professor Mihnea Colţ oiu for bringing Bremermann's problem to my attention and to Professor Terrence Napier for very useful discussions.
References

1. Bremermann, H.J.: Die Charakterisierung Rungescher Gebiete durch plurisubharmonische Funktionen. Math. Ann. 136, 173–186, 1958.
2. Hörmander, L.: Notions of convexity. Progress in Mathematics, 127. Birkhäuser, 1994.
3. Fornæss, J.E.; Stensønes, B.: Lectures on Counterexamples in Several Complex Variables. Mathematical Notes, 33. Princeton University Press, 1987.
4. Ohsawa, T.: Analysis of several complex variables. Translations of Mathematical Monographs, 211. American Mathematical Society, 2002.
5. Wermer, J.: An example concerning polynomial convexity. Math. Ann. 139, 147–150, 1959.

Cezar Joiţ a Institute of Mathematics of the Romanian Academy P.O. Box 1-764, Bucharest 014700 ROMANIA E-mail address : Cezar.Joita@imar.ro