Pome

9.4 Reduced spaces and normalizationC16A2F5D9879

C16A2F5D9879

The main result of this section is the existence and uniqueness of the normalization of a reduced complex space (theorem 9.4.6), which we construct locally from weakly holomorphic functions.

A reduced complex space is a Hausdorff topological space $X$ with a sheaf $\mathcal O_ X$ of continuous complex-valued functions, its structure sheaf, such that every point has an open neighborhood $U$ and a homeomorphism $\phi $ of $U$ onto an analytic subset $A$ of a domain $\Omega \subset \mathbb C^ N$ with the following property: a function $g$ on an open set $V\subset U$ belongs to $\mathcal O_ X(V)$ if and only if $g\circ \phi ^{-1}$ is, near each point of $\phi (V)$, the restriction of a holomorphic function on an open subset of $\Omega $. In other words, $\phi $ identifies $\mathcal O_ X|_ U$ with $\mathcal O_ A=(\mathcal O_\Omega /\mathcal I_ A)|_ A$. We call $(U,\phi )$ a local model, or local embedding, of $X$; holomorphic functions on $X$ extend locally to the ambient domains. As stated in the Conventions, all complex spaces are reduced, paracompact and Hausdorff; in this appendix we also assume that they have a countable basis of open sets. A compact space with such local models has a countable basis automatically, since finitely many local models cover it.

A continuous map $F\colon Y\to X$ of complex spaces is holomorphic if $g\circ F\in \mathcal O_ Y$ for every local section $g$ of $\mathcal O_ X$, that is, if in local models of $X$ the components of $F$ belong to $\mathcal O_ Y$. By the definition of $\mathcal O_ Y$, these components extend near each point of $Y$ to holomorphic functions on an open subset of the ambient space of a local model of $Y$; consequently, compositions of holomorphic maps are holomorphic. An isomorphism is a holomorphic bijection whose inverse is holomorphic. Two local models at the same point differ by an isomorphism of analytic sets which, together with its inverse, is locally the restriction of holomorphic maps between the ambient domains. Such an isomorphism maps analytic subsets to analytic subsets, and it maps submanifolds to submanifolds, because a holomorphic map defined on a manifold that has a holomorphic left inverse is an injective immersion and a homeomorphism onto its image, so that its image is a submanifold by lemma 9.3.14. Hence regular and singular points, local branches and dimension are well defined on $X$, independently of the local model. An analytic subset of $X$ is a closed subset that is analytic in every local model; with the restricted structure sheaf it is again a complex space.

The structure sheaf is coherent as a module over itself. Indeed, in a local model let $f_1,\ldots ,f_ q$ be sections of $\mathcal O_ X$ with holomorphic lifts $F_1,\ldots ,F_ q$, and let $g_1,\ldots ,g_ N$ generate $\mathcal I_ A$ near a point, by theorem 9.3.10; the relations of the $f_ i$ over $\mathcal O_ X$ are the images of the projections to the first $q$ coordinates of the relations of $F_1,\ldots ,F_ q,g_1,\ldots ,g_ N$ over $\mathcal O_\Omega $, which are locally finitely generated by theorem 9.2.7. Thus lemma 9.2.6 applies to $\mathcal O_ X$-modules. Each local ring $\mathcal O_{X,x}$ is a quotient of $\mathcal O_ N$, hence Noetherian. The zero set $V(\mathfrak a)$ of a coherent ideal sheaf $\mathfrak a\subset \mathcal O_ X$ is an analytic subset; conversely, the ideal sheaf $\mathcal I_ Y\subset \mathcal O_ X$ of functions vanishing on an analytic subset $Y\subset X$ is locally generated by the images of generators of the ideal sheaf of $Y$ in an ambient domain, so it is coherent by theorem 9.3.10, lemma 9.2.6.

A holomorphic map $F\colon Y\to X$ is proper if inverse images of compact sets are compact, and finite if it is proper with finite fibers. It is bimeromorphic if there is a nowhere dense analytic subset $E\subset X$ such that $F^{-1}(E)$ is nowhere dense in $Y$ and $F$ restricts to an isomorphism $Y\setminus F^{-1}(E)\to X\setminus E$. A proper map is closed: if $C\subset Y$ is closed and $F(y_ j)\to x$ with $y_ j\in C$, the points $y_ j$ eventually lie in the compact inverse image of a compact neighborhood of $x$, so a subsequence converges to some $y\in C$, and $F(y)=x$.

Lemma 9.4.1.F56FCBEF9522 Let $F\colon Y\to X$ be a bimeromorphic holomorphic map of complex spaces. Then $F^{-1}(B)$ is nowhere dense in $Y$ for every nowhere dense analytic subset $B\subset X$. Consequently, the set $E$ in the definition may be enlarged by any nowhere dense analytic subset of $X$; if $U\subset X$ is the complement of a nowhere dense analytic subset, then $F^{-1}(U)$ is dense in $Y$; and every nonempty open subset of $Y$ contains a nonempty open set that $F$ maps isomorphically onto an open subset of $X$.

Proof. Let $E$ be as in the definition, and let $G\subset F^{-1}(B)$ be open in $Y$. Then $G\setminus F^{-1}(E)$ is an open subset of $Y\setminus F^{-1}(E)$ contained in $F^{-1}(B\setminus E)$, which corresponds under the isomorphism $Y\setminus F^{-1}(E)\to X\setminus E$ to the nowhere dense subset $B\setminus E$ of $X\setminus E$; so $G\setminus F^{-1}(E)$ is empty, and $G$, an open subset of the nowhere dense set $F^{-1}(E)$, is empty as well. In particular $F$ is also an isomorphism over $X\setminus (E\cup B)$, and $F^{-1}(E\cup B)=F^{-1}(E)\cup F^{-1}(B)$ is nowhere dense, as the union of two closed nowhere dense sets; so $E$ may be replaced by $E\cup B$. If $U=X\setminus B$, then $F^{-1}(U)$ is the complement of the closed nowhere dense set $F^{-1}(B)$, hence dense. Finally, a nonempty open set $G\subset Y$ meets the dense open set $Y\setminus F^{-1}(E)$, and $F$ maps the nonempty open set $G\setminus F^{-1}(E)$ isomorphically onto an open subset of $X\setminus E$. $\square$

A weakly holomorphic function on an open set $U\subset X$ is a holomorphic function on $U\cap X_{\mathrm{reg}}$ that is locally bounded near $U\cap X_{\mathrm{sing}}$: every point of $U$ has a neighborhood $V$ such that the function is bounded on $V\cap X_{\mathrm{reg}}$. These functions form a sheaf of rings $\widetilde{\mathcal O}_ X$, which contains $\mathcal O_ X$ because $X_{\mathrm{reg}}$ is dense in $X$ (corollary 9.3.16). They can be computed branch by branch.

Lemma 9.4.2.26BA2ABFD727 Let $x$ be a point of a complex space $X$, and let $X_1,\ldots ,X_ p$ be sufficiently small representatives of the branches of $(X,x)$ in a local model, on an open neighborhood $U$ of $x$ with $X\cap U=X_1\cup \cdots \cup X_ p$. Let $V\subset U$ be open. Then for each $i$ the restriction of a weakly holomorphic function $f$ on $X\cap V$ to $X_ i\cap X_{\mathrm{reg}}\cap V$ extends uniquely to a weakly holomorphic function $f_ i$ on $X_ i\cap V$, and $f\mapsto (f_1,\ldots ,f_ p)$ is a bijection onto the $p$-tuples of weakly holomorphic functions on the sets $X_ i\cap V$. In particular,

\[ \widetilde{\mathcal O}_{X,x}=\bigoplus _{i=1}^ p\widetilde{\mathcal O}_{X_ i,x}. \]

Proof. By corollary 9.3.13(1), $X_{\mathrm{reg}}\cap U$ is the disjoint union of the sets $X_ i\cap X_{\mathrm{reg}}=(X_ i)_{\mathrm{reg}}\setminus \bigcup _{j\neq i}X_ j$, each of which is open in $X_{\mathrm{reg}}$. The analytic subset $(X_ i)_{\mathrm{reg}}\cap \bigcup _{j\neq i}X_ j$ of the manifold $(X_ i)_{\mathrm{reg}}$ is nowhere dense in it, by corollary 9.3.6, applied to $(X_ i,x)$ and its subgerm $(X_ i\cap \bigcup _{j\neq i}X_ j,x)$, which differs from $(X_ i,x)$ because the branches are irredundant. Hence a bounded holomorphic function on $X_ i\cap X_{\mathrm{reg}}\cap V$ extends uniquely across this analytic subset to a holomorphic function on $(X_ i)_{\mathrm{reg}}\cap V$, by lemma 9.2.9 applied on connected coordinate balls of $(X_ i)_{\mathrm{reg}}$, and the extension is bounded by the same bound. Applied near each point of $X_ i\cap V$, this shows that the restriction of $f$ has a unique extension $f_ i$, which is weakly holomorphic on $X_ i\cap V$. Conversely, if $f_ i$ is weakly holomorphic on $X_ i\cap V$ for each $i$, the function equal to $f_ i$ on $X_ i\cap X_{\mathrm{reg}}\cap V$ is holomorphic and locally bounded on $X_{\mathrm{reg}}\cap V$, hence weakly holomorphic on $X\cap V$; by the uniqueness of the extensions, the two constructions are inverse to each other. Passing to germs at $x$ gives the last assertion. $\square$

The total ring of fractions of the reduced local ring $\mathcal O_{X,x}$ is obtained by inverting its nonzerodivisors. These are exactly the germs that are not identically zero on any branch of $(X,x)$. Indeed, if $h$ is such a germ and $hg=0$, then $g$ vanishes on the complement of $\{ h=0\} $ in each branch, which is dense by corollary 9.3.6; so $g=0$. If instead $h$ vanishes identically on the branch $X_1$, choose for each $j\neq 1$ a germ $g_ j$ vanishing on $X_ j$ but not on $X_1$, which exists because $(X_1,x)\not\subset (X_ j,x)$; the product $v=\prod _{j\neq 1}g_ j$ does not vanish identically on $X_1$, because $I_{X_1,x}$ is prime, and $hv=0$.

Proposition 9.4.3.36422A732754 Every point of a complex space $X$ has a neighborhood $U$ with a holomorphic function $h$ on $U$ whose germ at each $y\in U$ does not vanish identically on any branch of $(X,y)$, such that

\[ h_ y\widetilde{\mathcal O}_{X,y}\subset \mathcal O_{X,y} \qquad (y\in U). \]

For every $x\in X$, the weakly holomorphic germs at $x$ are precisely the elements of the total ring of fractions of $\mathcal O_{X,x}$ that are integral over $\mathcal O_{X,x}$, and they form a finitely generated $\mathcal O_{X,x}$-module. If $(X,x)$ is irreducible, every weakly holomorphic function near $x$ has a finite limit at $x$ along $X_{\mathrm{reg}}$.

Proof. Step 1: an irreducible germ. Let $(X,0)$ be irreducible, with a local parametrization as in theorem 9.3.4, its degree $q$, its primitive linear form $u$, its Weierstrass polynomial $W$ and $\delta =\operatorname {Disc}(W)$. We show that $\delta _ y\widetilde{\mathcal O}_{X,y}\subset \mathcal O_{X,y}$ for every $y$ in the parametrization polydisc. Let $f$ be weakly holomorphic near $y$ and bounded on $V\cap X_{\mathrm{reg}}$ for a neighborhood $V$ of $y$. Let $y=y^{(1)},y^{(2)},\ldots ,y^{(s)}$ be the points of $X$ over $y'$, choose disjoint neighborhoods $V_1\subset V,V_2,\ldots ,V_ s$ of them, and, using properness, a connected neighborhood $N'$ of $y'$ such that all points of $X$ over $N'$ lie in $\bigcup _ lV_ l$. Define $F$ on $X_{\mathrm{reg}}\cap \pi ^{-1}(N')$ by $F=f$ on $V_1$ and $F=0$ on the other $V_ l$.

For $z'\in N'\setminus \{ \delta =0\} $, let $z^{(1)},\ldots ,z^{(q)}$ be the points of $X$ over $z'$; the values $u(z^{(i)})$ are the distinct roots of $W(z',\cdot )$, by theorem 9.3.4, so there is a unique polynomial $\sum _{j<q}b_ j(z')T^ j$ taking the value $F(z^{(i)})$ at $T=u(z^{(i)})$ for every $i$. By Cramer's rule $\delta b_ j=D\cdot D_ j$, where $D$ is the Vandermonde determinant of the $u(z^{(i)})$, so that $D^2=\delta $ by (9.3.2), and $D_ j$ is obtained by replacing its $j$th column with the values $F(z^{(i)})$. The products $D\cdot D_ j$ do not depend on the order of the points; they are holomorphic, because the points depend holomorphically on $z'$ locally, and bounded, because $u$ and $F$ are bounded. By lemma 9.2.9 they extend holomorphically to $N'$. The holomorphic function $H(z)=\sum _ j(\delta b_ j)(z')u(z'')^ j$ on $N'\times \Delta ''$ equals $\delta F$ on the covering locus over $N'$, hence $\delta f$ on $X_{\mathrm{reg}}\cap V_1\cap \pi ^{-1}(N')$, by density of the covering locus and continuity. Thus $\delta f$ extends to the holomorphic germ of $H|_ X$ at $y$.

Step 2: several branches. Fix $x\in X$, and let $X_1,\ldots ,X_ p$ be sufficiently small representatives of the branches of $(X,x)$, and let $\delta _ j$ be a denominator for $X_ j$ as in Step 1, lifted to a holomorphic function of the ambient coordinates. Adding to $\delta _ j$ a linear combination $\sum _ la_ le_{jl}$ of local generators $e_{jl}$ of the ideal of $X_ j$ does not change it on $X_ j$. For each $k\neq j$ some $e_{jl}$ does not vanish identically on $X_ k$, so the coefficients $a$ for which $\delta _ j+\sum _ la_ le_{jl}$ vanishes identically on $X_ k$ form a proper affine subspace or the empty set. Choosing $a$ outside these finitely many subspaces, we obtain $\widetilde\delta _ j$, equal to $\delta _ j$ on $X_ j$ and not identically zero on any $X_ k$. Then $\eta =\prod _ j\widetilde\delta _ j$ is not identically zero on any $X_ k$, and it is a denominator for each $X_ j$ separately: if $f$ is weakly holomorphic on $X$ near a point $y$, the restriction of $f$ to $X_ j\cap X_{\mathrm{reg}}$ extends to a weakly holomorphic function $f_ j$ on $X_ j$ near $y$, by lemma 9.4.2, so by Step 1 $\eta f_ j=g_ j|_{X_ j}$ for some holomorphic $g_ j$ near $y$.

For each $j$ choose a product $v_ j$ of local equations of the branches $X_ k$, $k\neq j$, that does not vanish identically on $X_ j$, and put $h=\eta \sum _ jv_ j$. On $X_ j$ we have $h=\eta v_ j$, which is not identically zero, and $hf=\sum _ jv_ jg_ j$ holds on every $X_ j\cap X_{\mathrm{reg}}$, hence on $X_{\mathrm{reg}}$ near $y$. Thus $h_ y\widetilde{\mathcal O}_{X,y}\subset \mathcal O_{X,y}$ for all $y$ near $x$. Finally, $\{ h=0\} \cap X_ j$ has smaller dimension than $X_ j$ at $x$, hence at all points near $x$, by upper semicontinuity; every branch of $(X,y)$ is a branch of some $(X_ j,y)$ and has dimension $\dim X_ j$, by purity. So, after shrinking $U$, the germ of $h$ at every $y\in U$ is not identically zero on any branch of $(X,y)$.

Step 3: integrality, finiteness and limits. The map $f\mapsto (hf)/h$ embeds $\widetilde{\mathcal O}_{X,x}$ into the total ring of fractions of $\mathcal O_{X,x}$; it does not depend on the choice of $h$, and it is injective because $X_{\mathrm{reg}}\setminus \{ h=0\} $ is dense in $X_{\mathrm{reg}}$. Multiplication by $h$ identifies $\widetilde{\mathcal O}_{X,x}$ with an ideal of the Noetherian ring $\mathcal O_{X,x}$, so it is a finitely generated $\mathcal O_{X,x}$-module.

Every weakly holomorphic germ is integral. On an irreducible germ, in the notation of Step 1, the elementary symmetric functions of the values of $f$ at the $q$ points of $X$ over $z'\in \Delta '\setminus \{ \delta =0\} $ are bounded holomorphic functions, which extend to $\Delta '$ by lemma 9.2.9. They are the coefficients of a monic polynomial $P(z',T)$ with $P(z',f)=0$ on the covering locus, hence on $X_{\mathrm{reg}}$, and these coefficients belong to $\mathcal O_ d\subset \mathcal O_{X,x}$. For several branches, the product of the polynomials of the branches, with coefficients lifted to holomorphic functions, is a monic equation for $f$. Conversely, let $a/b$ be an element of the total ring of fractions, with $b$ a nonzerodivisor, that is integral over $\mathcal O_{X,x}$. On $X_{\mathrm{reg}}\setminus \{ b=0\} $ near $x$ the function $a/b$ satisfies a monic equation with bounded coefficients, so it is bounded by (9.2.1). The set $\{ b=0\} $ meets each connected open subset of the manifold $X_{\mathrm{reg}}$ in a nowhere dense analytic subset, so by lemma 9.2.9 $a/b$ extends to a holomorphic function on $X_{\mathrm{reg}}$ near $x$, which is weakly holomorphic and is mapped to $a/b$.

Let $(X,x)$ be irreducible and $f$ weakly holomorphic near $x$, satisfying the monic equation $P(z',f)=0$ just constructed. As $Q$ runs through the parametrization polydiscs around $x$, the sets $X_{\mathrm{reg}}\cap Q$ are the traces on $X_{\mathrm{reg}}$ of a basis of neighborhoods of $x$, and each of them is connected, because it lies between the connected covering locus and its closure. The cluster set of $f$ at $x$ is the intersection of the decreasing compact connected nonempty sets $\overline{f(X_{\mathrm{reg}}\cap Q)}$, hence it is nonempty, compact and connected. If $w$ is a cluster value, then $w=\lim f(z_ k)$ for points $z_ k\to x$ of $X_{\mathrm{reg}}$, and $P(x',w)=\lim P(z_ k',f(z_ k))=0$; so the cluster set is contained in the finite set of roots of $P(x',\cdot )$. Being connected, it is a single point, and $f$ has a finite limit at $x$. $\square$

A point $x\in X$ is normal if $(X,x)$ is irreducible and $\mathcal O_{X,x}$ is integrally closed in its field of fractions. By proposition 9.4.3, this means that $\widetilde{\mathcal O}_{X,x}=\mathcal O_{X,x}$. Conversely, this equality forces $(X,x)$ to be irreducible: otherwise the weakly holomorphic function equal to $1$ on one branch and to $0$ on the others would be holomorphic, hence continuous, at $x$, which is impossible because it takes both values at points arbitrarily close to $x$. Thus $x$ is normal if and only if $\widetilde{\mathcal O}_{X,x}=\mathcal O_{X,x}$. The space $X$ is normal if all its points are normal. Regular points are normal by lemma 9.2.9.

Lemma 9.4.4.D73CC9F08C05 The set of nonnormal points of a complex space $X$ is an analytic subset of $X$. If $X$ is normal and $E\subset X$ is a nowhere dense analytic subset, every holomorphic function on $X\setminus E$ that is locally bounded near $E$ extends uniquely to a holomorphic function on $X$.

Proof. We show that the set of nonnormal points is the support of a coherent sheaf. We first record two facts about coherent sheaves. If $\mathcal F$ and $\mathcal G$ are coherent $\mathcal O_ X$-modules, the sheaf $\mathcal Hom(\mathcal F,\mathcal G)$ of module homomorphisms is coherent, and its stalk at $x$ is $\operatorname {Hom}_{\mathcal O_{X,x}}(\mathcal F_ x,\mathcal G_ x)$. Indeed, $\mathcal F$ has local presentations $\mathcal O_ X^ a\to \mathcal O_ X^ b\to \mathcal F\to 0$, and $\mathcal Hom(\mathcal F,\mathcal G)$ is the kernel of the induced morphism $\mathcal G^ b\to \mathcal G^ a$; we apply lemma 9.2.6.

Next, the support of a coherent module $\mathcal F$ is analytic. In a local presentation as above, given by a $b\times a$ matrix $\Phi $ of holomorphic functions, $\mathcal F_ x=0$ exactly when the matrix $\Phi (x)$ has rank $b$. If $\Phi (x)$ has rank $b$, then $\mathcal F_ x=\mathfrak m_ x\mathcal F_ x$, where $\mathfrak m_ x$ is the maximal ideal, and the following determinant argument gives $\mathcal F_ x=0$: writing each of finitely many generators of $\mathcal F_ x$ as a combination of them with coefficients in $\mathfrak m_ x$, with coefficient matrix $C$, the adjugate identity shows that $\det (I-C)$ annihilates the generators, and this determinant is a unit. Conversely, if $\mathcal F_ x=0$, then $\Phi $ maps $\mathcal O_{X,x}^ a$ onto $\mathcal O_{X,x}^ b$, and reducing modulo $\mathfrak m_ x$ shows that $\Phi (x)$ has rank $b$. Hence the support of $\mathcal F$ is the common zero set of the minors of size $b$ of $\Phi $.

Let $h$ be a denominator as in proposition 9.4.3 on an open set $U$, let $\mathcal I$ be the coherent ideal sheaf of $\{ h=0\} $ in $U$, and put $\mathcal F=\mathcal Hom(\mathcal I,\mathcal I)$. Multiplication embeds $\mathcal O_ U$ in $\mathcal F$, because $\mathcal I$ contains the nonzerodivisor $h$. Every $\varphi \in \mathcal F_ y$ is multiplication by an element of the total ring of fractions: if $a=\varphi (h)$, then $h\varphi (g)=\varphi (hg)=ag$ for $g\in \mathcal I_ y$, so $\varphi $ is multiplication by $a/h$. Let $g_1,\ldots ,g_ N$ generate $\mathcal I_ y$, and write $(a/h)g_ i=\varphi (g_ i)=\sum _ jc_{ij}g_ j$ with $c_{ij}\in \mathcal O_{X,y}$. Multiplying the matrix $(a/h)I-(c_{ij})$ by its adjugate, we see that its determinant, a monic polynomial in $a/h$ with coefficients in $\mathcal O_{X,y}$, annihilates every $g_ i$, hence $\mathcal I_ y$ and in particular $h$; since $h$ is invertible in the total ring of fractions, this polynomial vanishes, and $a/h$ is integral over $\mathcal O_{X,y}$. Hence, by proposition 9.4.3,

\[ \mathcal O_ U\subset \mathcal F\subset \widetilde{\mathcal O}_ U. \]

We claim that the support of the coherent sheaf $\mathcal F/\mathcal O_ U$ is the set of nonnormal points of $U$; this proves the first assertion. At a normal point $\widetilde{\mathcal O}_{X,y}=\mathcal O_{X,y}$, so $\mathcal F_ y=\mathcal O_{X,y}$. Let $y$ be a nonnormal point, so that $\widetilde{\mathcal O}_{X,y}\neq \mathcal O_{X,y}$. By the analytic Nullstellensatz (corollary 9.3.8), applied in a local model, every element of $\mathcal I_ y$ has a power in $h\mathcal O_{X,y}$. Let $g_1,\ldots ,g_ N$ generate $\mathcal I_ y$, and choose $e$ with $g_ i^ e\in h\mathcal O_{X,y}$ for all $i$. Every monomial of degree $\ell =N(e-1)+1$ in the $g_ i$ contains some $g_ i$ at least $e$ times, so $\mathcal I_ y^\ell \subset h\mathcal O_{X,y}$, and therefore $\mathcal I_ y^\ell \widetilde{\mathcal O}_{X,y}\subset h\widetilde{\mathcal O}_{X,y}\subset \mathcal O_{X,y}$.

Let $k$ be the least integer with $\mathcal I_ y^ k\widetilde{\mathcal O}_{X,y}\subset \mathcal O_{X,y}$; then $k\geq 1$, and we may choose $w\in \mathcal I_ y^{k-1}\widetilde{\mathcal O}_{X,y}$ with $w\notin \mathcal O_{X,y}$. Then $w\mathcal I_ y\subset \mathcal O_{X,y}$, and every product $wg$, $g\in \mathcal I_ y$, vanishes on $\{ h=0\} $: at regular points because $g$ vanishes there and $w$ is holomorphic, and at singular points because $w$ is locally bounded, $g$ tends to zero, and $wg$ is continuous. Thus $w\mathcal I_ y\subset \mathcal I_ y$, so multiplication by $w$ belongs to $\mathcal F_ y$; it is not multiplication by an element of $\mathcal O_{X,y}$, since $w\notin \mathcal O_{X,y}$ and $h$ is a nonzerodivisor. This proves the claim.

For the last assertion, let $f$ be a holomorphic function on $X\setminus E$ that is locally bounded near $E$. Since $E$ is nowhere dense, an extension is unique. The set $E\cap X_{\mathrm{reg}}$ meets every connected open subset of the manifold $X_{\mathrm{reg}}$ in a nowhere dense analytic subset, so $f$ extends holomorphically to $X_{\mathrm{reg}}$ by lemma 9.2.9. The extension is locally bounded near $X_{\mathrm{sing}}$, because on a neighborhood of each point it is bounded by a bound of $f$ on the dense subset $X_{\mathrm{reg}}\setminus E$; hence it is weakly holomorphic, and therefore holomorphic, because $X$ is normal. It agrees with $f$ on $X\setminus E$, since both functions are continuous there and agree on the dense subset $X_{\mathrm{reg}}\setminus E$. $\square$

Normalizations are first constructed locally, as closures of graphs of weakly holomorphic functions.

Lemma 9.4.5.58E87F89D557 Let $X$ be an analytic subset of a domain $\Omega \subset \mathbb C^ N$, with coordinates $z$, let $x\in X$, and let $X_1,\ldots ,X_ p$ be sufficiently small representatives of the branches of $(X,x)$. There is an open neighborhood $W\subset \Omega $ of $x$ with the following properties. For each $i$ there are functions $f_{i1},\ldots ,f_{im_ i}$, weakly holomorphic on $X_ i\cap W$, and monic polynomials $P_{ij}(z,T)$ in $T$ with coefficients holomorphic on $W$, such that $P_{ij}(z,f_{ij})=0$ on $(X_ i)_{\mathrm{reg}}\cap W$. The closure $Y_ i$ in $W\times \mathbb C^{m_ i}$ of the graph of $(f_{i1},\ldots ,f_{im_ i})$ over $(X_ i)_{\mathrm{reg}}\cap W$ is a normal analytic subset of $W\times \mathbb C^{m_ i}$, on which $P_{ij}(z,w_ j)=0$ for all $j$, and the projection $\nu _ i\colon Y_ i\to X_ i\cap W$, $(z,w)\mapsto z$, is finite and bimeromorphic, is an isomorphism over $(X_ i)_{\mathrm{reg}}\cap W$, and has a single point over $x$. The map from the disjoint union of $Y_1,\ldots ,Y_ p$ to $X\cap W$ given by the $\nu _ i$ is finite and bimeromorphic, is an isomorphism over $X_{\mathrm{reg}}\cap W$, and has $p$ points over $x$. All these properties persist when $W$ is replaced by a smaller open neighborhood of $x$ and each $Y_ i$ by its part over it.

Proof. Step 1: the construction for an irreducible germ. Suppose first that $p=1$, so that $(X,x)$ is irreducible. Choose generators $f_1,\ldots ,f_ m$ of the $\mathcal O_{X,x}$-module $\widetilde{\mathcal O}_{X,x}$ and a denominator $h$ as in proposition 9.4.3, so that $hf_ j=g_ j$ with $g_ j$ holomorphic. Choose a neighborhood $W\subset \Omega $ of $x$ on which $h$ and the $g_ j$ are holomorphic, the $f_ j$ are weakly holomorphic, and each $f_ j$ satisfies a monic equation $P_ j(z,f_ j)=0$ whose coefficients are holomorphic on $W$. On $(X\cap W)\setminus \{ h=0\} $ the functions $g_ j/h$ are holomorphic and agree with $f_ j$ on $X_{\mathrm{reg}}$. Let $Y$ be the closure in $W\times \mathbb C^ m$ of the graph of $(g_1/h,\ldots ,g_ m/h)$ over $(X\cap W)\setminus \{ h=0\} $. This graph is the part outside $\{ h=0\} $ of the analytic set

\[ Z=\{ (z,w)\in W\times \mathbb C^ m:\ z\in X,\ h(z)w_ j=g_ j(z)\ \text{for all }j\} . \]

By lemma 9.3.7, applied in $W\times \mathbb C^ m$ to $Z$ and the analytic set $\{ h=0\} $, the closure $Y$ of $Z\setminus \{ h=0\} $ is analytic; near each point it is the union of the branches of $Z$ on which $h$ does not vanish identically.

The monic equations $P_ j(z,g_ j/h)=0$ hold on $X_{\mathrm{reg}}\setminus \{ h=0\} $, which is dense in $X\setminus \{ h=0\} $, and their left-hand sides are continuous there; so they hold on the graph, hence on its closure $Y$: $P_ j(z,w_ j)=0$. By (9.2.1) they bound $w$ over compact subsets of $X\cap W$ and leave finitely many possible values of $w$ over each point. Hence the projection $\nu \colon Y\to X\cap W$ is proper with finite fibers.

Over $X_{\mathrm{reg}}$, the set $Y$ is the graph of $f=(f_1,\ldots ,f_ m)$: a point of $Y$ over $z_0\in X_{\mathrm{reg}}$ is a limit of graph points over points $z_ k\to z_0$, and since the $f_ j$ are continuous at $z_0$, it equals $(z_0,f(z_0))$. Conversely, $(z_0,f(z_0))$ lies in $Y$, because $z_0$ is a limit of points of $X_{\mathrm{reg}}\setminus \{ h=0\} $: the germ of $h$ at $z_0$ is not identically zero on the manifold germ $(X,z_0)$, so $\{ h=0\} $ is nowhere dense in $X_{\mathrm{reg}}$ near $z_0$, by lemma 9.2.8. Hence $\nu $ is an isomorphism over $X_{\mathrm{reg}}\cap W$, with inverse $z\mapsto (z,f(z))$. This graph is dense in $Y$, because $X_{\mathrm{reg}}\setminus \{ h=0\} $ is dense in $X\setminus \{ h=0\} $ and $(g_1/h,\ldots ,g_ m/h)$ is continuous there; so $Y$ is the closure of the graph of $f$ over $X_{\mathrm{reg}}\cap W$, the set $\nu ^{-1}(X_{\mathrm{sing}})$ is nowhere dense, and $\nu $ is bimeromorphic. Every point of $Y$ over $x$ is a limit of points $(z_ k,f(z_ k))$ with $z_ k\in X_{\mathrm{reg}}$ and $z_ k\to x$; by the last assertion of proposition 9.4.3, the fiber over $x$ is the single point $y=(x,\lim _{z\to x}f(z))$.

Step 2: normality of $Y$ at $y$. Let $a$ be a weakly holomorphic function on $Y$ near $y$. Since $\nu ^{-1}(X_{\mathrm{reg}})$ is a manifold, it is contained in $Y_{\mathrm{reg}}$, and $b=a\circ (\nu |_{\nu ^{-1}(X_{\mathrm{reg}})})^{-1}$ is holomorphic on $X_{\mathrm{reg}}$ near $x$. It is bounded near $x$: by properness and since the fiber over $x$ is $\{ y\} $, the inverse image of a small neighborhood of $x$ lies in a neighborhood of $y$ on which $a$ is bounded. Thus $b\in \widetilde{\mathcal O}_{X,x}$, and $b=\sum _ ja_ jf_ j$ with $a_ j\in \mathcal O_{X,x}$. Lift the $a_ j$ to holomorphic functions $\widetilde a_ j$ near $x$ in $\Omega $. The function $\sum _ j\widetilde a_ j(z)w_ j$ is holomorphic near $y$, and it equals $a$ on $\nu ^{-1}(X_{\mathrm{reg}})$ near $y$, which is dense in $Y_{\mathrm{reg}}$; since both functions are continuous on $Y_{\mathrm{reg}}$, they agree on $Y_{\mathrm{reg}}$ near $y$. Hence $\widetilde{\mathcal O}_{Y,y}=\mathcal O_{Y,y}$, and $y$ is a normal point. By lemma 9.4.4, all points of $Y$ near $y$ are normal, and by properness the inverse image of a small neighborhood $U$ of $x$ lies in this set of normal points. Replacing $W$ by $U$ and $Y$ by $\nu ^{-1}(U)$, we obtain the assertions of the lemma for $p=1$. They persist when $W$ is replaced by a smaller open neighborhood of $x$, because the closure of the graph over the smaller set is the part of $Y$ over it, and normality, finiteness, bimeromorphy and being an isomorphism over the regular locus are preserved under restriction over open subsets of $X$.

Step 3: several branches. In general, apply Steps 1 and 2 to each irreducible germ $(X_ i,x)$, with $X_ i$ regarded as an analytic subset of a ball around $x$ in $\Omega $, and let $W$ be the intersection of the resulting neighborhoods; by the persistence just noted, the assertions about the individual $Y_ i$ hold over $W$. The disjoint union of the $Y_ i$ is normal, and its map to $X\cap W$ is finite, with $p$ points over $x$. It is an isomorphism over $X_{\mathrm{reg}}\cap W$, because, by corollary 9.3.13(1), $X_{\mathrm{reg}}\cap W$ is the disjoint union of the open sets $X_ i\cap X_{\mathrm{reg}}=(X_ i)_{\mathrm{reg}}\setminus \bigcup _{j\neq i}X_ j$, over which $\nu _ i$ is an isomorphism. It is bimeromorphic, with the nowhere dense analytic subset $X_{\mathrm{sing}}\cap W$ of $X\cap W$. Indeed, by corollary 9.3.13(1), $X_{\mathrm{sing}}\cap X_ i=(X_ i)_{\mathrm{sing}}\cup (X_ i\cap \bigcup _{j\neq i}X_ j)$, which is a nowhere dense analytic subset of $X_ i\cap W$ by corollary 9.3.6; so its inverse image under the bimeromorphic map $\nu _ i$ is nowhere dense in $Y_ i$, by lemma 9.4.1. These properties again persist for smaller neighborhoods of $x$. $\square$

Theorem 9.4.6 (Normalization).12A47DB401E2 Every complex space $X$ has a normalization: a normal complex space $\widetilde X$ with a finite bimeromorphic holomorphic map $\nu \colon \widetilde X\to X$ that is an isomorphism over $X_{\mathrm{reg}}$. It is unique up to a unique isomorphism over $X$. The fiber $\nu ^{-1}(x)$ has one point for each irreducible branch of $(X,x)$, and

\[ \nu _*\mathcal O_{\widetilde X}=\widetilde{\mathcal O}_ X. \]

Proof. Step 1: local normalizations. Let $x\in X$, and choose a local model of a neighborhood of $x$. By lemma 9.4.5, applied to sufficiently small representatives of the branches of $(X,x)$, there is a neighborhood $W$ of $x$ such that $X\cap W$ has a normalization, namely the disjoint union of the normal spaces $Y_ i$ with the maps $\nu _ i$, whose fiber over $x$ has one point for each branch of $(X,x)$.

Step 2: uniqueness. Let $\nu _1\colon Y_1\to X$ and $\nu _2\colon Y_2\to X$ be finite bimeromorphic maps from normal complex spaces that are isomorphisms over $X_{\mathrm{reg}}$. Over $X_{\mathrm{reg}}$ there is a unique isomorphism $\phi _0=\nu _2^{-1}\circ \nu _1$ commuting with the maps to $X$; we show that it extends holomorphically. Let $p\in Y_1$ and $x=\nu _1(p)$. Choose disjoint open neighborhoods $W_ l$ of the finitely many points $q_ l$ of $\nu _2^{-1}(x)$, each relatively compact in the domain $W'_ l$ of a local model of $Y_2$, so that the coordinate functions of these models are bounded on $W_ l$. By properness of $\nu _2$, there is a neighborhood $U$ of $x$ with $\nu _2^{-1}(U)\subset \bigcup _ lW_ l$. The analytic subset $E=\nu _1^{-1}(X_{\mathrm{sing}})$ of $Y_1$ is nowhere dense, because $\nu _1$ is bimeromorphic and $X_{\mathrm{sing}}$ is nowhere dense in $X$ (lemma 9.4.1).

Since $Y_1$ is normal, the germ $(Y_1,p)$ is irreducible, and a sufficiently small parametrization neighborhood $V$ of $p$ with $\nu _1(V)\subset U$ has the property that $V\setminus E$ is connected: its connected covering locus remains connected after removing the analytic subset $E$, which is nowhere dense in it, by lemma 9.2.9, and the result is dense in $V\setminus E$. Hence $\phi _0(V\setminus E)$ lies in a single $W_ l$. The coordinate functions of $W'_ l$ composed with $\phi _0$ are bounded holomorphic functions on $V\setminus E$, and they extend holomorphically to $V$ by lemma 9.4.4. The extended map takes values in $\overline{W_ l}\subset W'_ l$, and it satisfies the equations of the local model, which hold on the dense set $V\setminus E$; so it is a holomorphic map $V\to Y_2$ extending $\phi _0$. These local extensions are unique, by density of $V\setminus E$, and therefore glue to a holomorphic map $\phi \colon Y_1\to Y_2$ with $\nu _2\circ \phi =\nu _1$. Extending $\phi _0^{-1}$ in the same way, we obtain the inverse map, since the compositions are the identity on dense open sets. Any isomorphism over $X$ agrees with $\phi _0$ over $X_{\mathrm{reg}}$, so it equals $\phi $. This proves uniqueness.

Step 3: gluing. By the uniqueness just proved, applied over the overlaps of their domains, the local normalizations of Step 1 are canonically isomorphic there, and these isomorphisms are compatible on triple overlaps, because an isomorphism over $X$ between two normalizations is unique; so they glue to a space $\widetilde X$ with a map $\nu \colon \widetilde X\to X$. The space $\widetilde X$ is Hausdorff, since points with different images are separated by inverse images of disjoint neighborhoods and points with the same image lie in one local normalization. Since $X$ has a countable basis, countably many of the open sets over which local normalizations were constructed cover $X$, and each of these local normalizations, a finite disjoint union of analytic subsets of open subsets of the spaces $\mathbb C^{N+m_ i}$, has a countable basis; hence so has $\widetilde X$. Being locally compact and Hausdorff with a countable basis, $\widetilde X$ is paracompact. Being finite, normal and bimeromorphic are local properties over $X$, so $\nu $ is a normalization. The description of the fibers follows from Step 1.

Step 4: the direct image. Let $U\subset X$ be open. If $g$ is holomorphic on $\nu ^{-1}(U)$, then $g\circ \nu ^{-1}$ is holomorphic on $U\cap X_{\mathrm{reg}}$ and locally bounded near $U\cap X_{\mathrm{sing}}$, because $\nu $ is proper; so it is weakly holomorphic. Conversely, if $f$ is weakly holomorphic on $U$, then $f\circ \nu $ is holomorphic on $\nu ^{-1}(U\cap X_{\mathrm{reg}})$ and locally bounded near the nowhere dense analytic subset $\nu ^{-1}(U\cap X_{\mathrm{sing}})$; by lemma 9.4.4, applied on the normal space $\nu ^{-1}(U)$, it extends holomorphically to $\nu ^{-1}(U)$. These two maps are inverse to each other, which proves $\nu _*\mathcal O_{\widetilde X}=\widetilde{\mathcal O}_ X$. $\square$

Lemma 9.4.5 therefore describes the normalization locally: the normalization of each branch is the closure of the graph of finitely many weakly holomorphic functions satisfying monic equations.

Corollary 9.4.7.8D2ABD4CAC6A Let $X$ be an analytic subset of a domain $\Omega \subset \mathbb C^ N$, with coordinates $z$, let $\nu \colon \widetilde X\to X$ be its normalization, and let $x\in X$. Let $W$ and $\nu _ i\colon Y_ i\to X_ i\cap W$, $1\leq i\leq p$, be as in lemma 9.4.5, for sufficiently small representatives $X_1,\ldots ,X_ p$ of the branches of $(X,x)$. Then each $\nu _ i$ is a normalization of $X_ i\cap W$ whose fiber over $x$ is a single point, and there is an isomorphism of $\nu ^{-1}(W)$ onto the disjoint union of $Y_1,\ldots ,Y_ p$ under which $\nu $ corresponds to the maps $\nu _ i$. In particular, every holomorphic function on $\widetilde X$ near a point of $\nu ^{-1}(x)$ corresponds, near that point, to the restriction to some $Y_ i$ of a holomorphic function of $(z,w)$.

Proof. By lemma 9.4.5, each $Y_ i$ is normal, and $\nu _ i$ is finite, bimeromorphic and an isomorphism over $(X_ i)_{\mathrm{reg}}\cap W$, with a single point over $x$; so $\nu _ i$ is a normalization of $X_ i\cap W$. Likewise, the disjoint union of the $Y_ i$ is a normalization of $X\cap W$. So is $\nu ^{-1}(W)\to X\cap W$, because being finite, normal, bimeromorphic and an isomorphism over the regular locus are local properties over $X$. By the uniqueness in theorem 9.4.6, there is an isomorphism over $X\cap W$ between them. The last assertion holds because the holomorphic functions on the analytic set $Y_ i$ are, locally, restrictions of holomorphic functions of the ambient coordinates $(z,w)$. $\square$

Definition 9.4.8.0E1E539882AD A complex space is unibranch if all its germs $(X,x)$ are irreducible. Equivalently, each local ring $\mathcal O_{X,x}$ is a domain whose integral closure in its field of fractions is a local ring. Indeed, in a local model $\mathcal O_{X,x}=\mathcal O_ N/I_{X,x}$, so by proposition 9.3.1 the germ $(X,x)$ is irreducible if and only if $\mathcal O_{X,x}$ is a domain. In this case, by proposition 9.4.3, the integral closure is $\widetilde{\mathcal O}_{X,x}$, and taking the limit at $x$, which exists by the same proposition, is a ring homomorphism $\widetilde{\mathcal O}_{X,x}\to \mathbb C$. An element with nonzero limit is a unit, because its reciprocal is holomorphic and bounded on $X_{\mathrm{reg}}$ near $x$, hence weakly holomorphic; an element with limit zero is not a unit. So the non-units form the ideal of elements with limit zero, and $\widetilde{\mathcal O}_{X,x}$ is a local ring. Thus this definition agrees with the definition of unibranch local rings in commutative algebra.

Corollary 9.4.9.EE1633042D3D The normalization of a unibranch complex space is a homeomorphism. Every point $x$ of a complex space with $(X,x)$ irreducible, in particular every point of a unibranch space, has arbitrarily small neighborhoods $U$ such that $U\setminus A$ is connected for every nowhere dense analytic subset $A$ of $U$.

Proof. On a unibranch space the normalization has one-point fibers, so it is a continuous bijection; being proper, it is closed, hence a homeomorphism. For the second assertion, let $U$ be the intersection of $X$ with a parametrization polydisc of $(X,x)$. Its covering locus is a connected manifold, dense in $U$; it remains connected after deleting $A$, by lemma 9.2.9, because $A$ is nowhere dense, and the result is dense in $U\setminus A$. Hence $U\setminus A$ is connected. $\square$

Example 9.4.10.60A14C7A1845 The curve $\{ zw=0\} \subset \mathbb C^2$ has two branches at the origin, the coordinate axes, whose germs are irreducible, being germs of submanifolds (lemma 9.3.12). The disjoint union of the two axes, mapped to the curve by the inclusions, is a manifold, hence normal, and the map is finite, bimeromorphic and an isomorphism over the regular locus $\{ zw=0\} \setminus \{ 0\} $; by the uniqueness in theorem 9.4.6 it is the normalization. Its fiber over the origin has two points, and the curve is not unibranch.

The cusp $C=\{ w^2=z^3\} $ is normalized by $\gamma (t)=(t^2,t^3)$. Indeed, $\gamma $ is a holomorphic bijection of $\mathbb C$ onto $C$: for $(z,w)\in C$ with $z\neq 0$, the equations $t^2=z$, $t^3=w$ force $t=w/z$, and $t=w/z$ satisfies them, since $(w/z)^2=z$ and $(w/z)^3=(w/z)z=w$; and $\gamma (t)=0$ only for $t=0$. The points of $C\setminus \{ 0\} $ are regular: at such a point $z\neq 0$, so the map $(z,w)\mapsto (w^2-z^3,w)$ has the nonzero Jacobian determinant $-3z^2$ there, and by proposition 5.2.4(1) the functions $w^2-z^3$ and $w$ form holomorphic coordinates near such a point, in which $C$ is a coordinate line; so $C$ is a submanifold there. On $C\setminus \{ 0\} $ the inverse of $\gamma $ is the holomorphic function $t=w/z$. This function satisfies $t^2=z$, so it tends to $0$ at the origin; hence $\gamma $ is a homeomorphism.

The function $t$ is not the restriction of a holomorphic function $F(z,w)$ near the origin, because $F(t^2,t^3)$ has no term of degree one in $t$. Hence the origin is a singular point of $C$. Otherwise $t$, which is bounded near the origin, would extend holomorphically across it on the manifold $C$, by lemma 9.2.9, and would therefore be the restriction of a holomorphic function of $(z,w)$: in coordinates in which a submanifold is a coordinate plane, a holomorphic function on it extends independently of the other coordinates. So $t$ is weakly holomorphic on $C$, and integral, since $t^2=z$, but it is not holomorphic at the origin, so $C$ is not normal there. Thus $\gamma $ is a finite bimeromorphic map from a manifold that is an isomorphism over $C_{\mathrm{reg}}=C\setminus \{ 0\} $, and by the uniqueness in theorem 9.4.6 it is the normalization. The cusp is therefore unibranch but not normal, and it distinguishes unibranch spaces from normal spaces.

Cite this page

Comments

Corrections, questions, and mathematical discussion are welcome. Sign in with GitHub to comment. Comments are public.