1. Introduction
Before stating and discussing the main problem of this article we give some brief preliminaries and introduce the notation that is used in the article.
1.0.1 Preliminaries on $\mu$-constant deformations
Let
be the $ {\mathbb {C}}$-algebra of analytic function germs at the origin $o$ of $ {\mathbb {C}}^{n+1}$ and $ {\mathbb {C}}^{n+1}_{o}$ the complex-analytic germ of $ {\mathbb {C}}^{n+1}$. By abuse of notation we denote by $o$ the origin of $ {\mathbb {C}}^{n+1}_{o}$. Let $V$ be a hypersurface of $ {\mathbb {C}}^{n+1}_{o}$, $n\geq 1$, given by an equation $f(x)=0$, where $f$ is irreducible in $ {\mathcal {O}_{n+1}^{x}}$. Assume that $V$ has an isolated singularity at $o$. One of the important topological invariants of the singularity $o\in V$ is the Milnor number $\mu (f)$, defined by
where $J(f):=(\partial _1f,\ldots,\partial _{n+1}f) \subset {\mathcal {O}_{n+1}^{x}}$ is the Jacobian ideal of $f$. In this article, we consider deformations of $f$ that preserve the Milnor number. Let $F\in {\mathbb {C}}\{x_1,\ldots,x_{n+1},s_1,\ldots,s_m\}$ be a deformation of $f$:
where $h_i\in {\mathcal {O}_{m}^{s}}:= {\mathbb {C}\{s_{1},\ldots,s_{m}\}}$, $m\geq 1$, and $g_i\in {\mathcal {O}_{n+1}^{x}}$ satisfy
Take a sufficiently small open set $\Omega \subset {\mathbb {C}}^{m}$ containing $o$, and representatives of the analytic function germs $h_1,\ldots,h_i,\ldots$ in $\Omega$. By a standard abuse of notation we denote these representatives by the same letters $h_1,\ldots,h_l$. We use the notation $F_{s'}(x):=F(x,s')$ when $s'\in \Omega$ is fixed. We say that the deformation $F$ is $\mu$-constant if the open set $\Omega$ can be chosen so that $\mu (F_{s'})=\mu (f)$ for all $s'\in \Omega$.
Let $\mathcal {E}:=\{e_1,e_2,\ldots,e_{n+1}\}\subset {\mathbb {Z}}_{\geq 0}^{n+1}$ be the standard basis of $ {\mathbb {R}}^{n+1}$. Let $g\in {\mathbb {C}}\{x_1,\ldots,x_{n+1}\}$ be a convergent power series. Write
in the multi-index notation. The Newton polyhedron $\Gamma _{+}(g)$ is the convex hull of the set $\bigcup _{\alpha \in \operatorname {Supp}(g)}(\alpha + {\mathbb {R}}_{\geq 0}^n)$, where $\operatorname {Supp}(g)$ (short for ‘the support of $g$’) is defined by
The Newton boundary of $\Gamma _{+}(g)$, denoted by $\Gamma (g)$, is the union of the compact faces of $\Gamma _{+}(g)$. For a face $\gamma$ of $\Gamma _{+}(g)$, the polynomial $g_{\gamma }$ is defined as follows:
We say that $g$, is non-degenerate with respect to its Newton boundary (or Newton non-degenerate) if for every compact face $\gamma$ of the Newton polyhedron $\Gamma _{+}(g)$ the partial derivatives $\partial _{x_1}g_{\gamma },\partial _{x_2}g_{\gamma } ,\ldots,\partial _{x_{n+1}}g_{\gamma }$ have no common zeros in $( {\mathbb {C}}^{\star })^{n+1}$.
We say that a deformation of $F$ of $f$ is non-degenerate if the neighborhood $\Omega$ of $o$ in $ {\mathbb {C}}^m$ can be chosen so that for all $s'\in \Omega$ the germ $F_{s'}$ is non-degenerate with respect to its Newton boundary $\Gamma (F_{s'})$.
We rewrite the deformation $F$ in the form:
and let $\operatorname {Supp}(F):=\{\alpha \, | \, a_{\alpha }(s)\not \equiv 0\}$. Given a sufficiently small open set $\Omega \subset {\mathbb {C}}^m$ containing $o$, we say that $s'\in \Omega$ is a general point of $ {\mathbb {C}}^m_{o}$ if
We remark that
and that $s'$ is general if whenever $s'$ belongs to the non-empty open set
In particular, plenty of general points $s'$ exist.
1.0.2 Preliminaries on simultaneous embedded resolutions
Let us keep the notation from the previous section. We put $S:= {\mathbb {C}}^m_o$, and denote by $W$ the deformation of $V$ given by $F$. Then we have the commutative diagram
where the morphism $\varrho$ is flat. Given a sufficiently small open set $\Omega \subset {\mathbb {C}}^{m}$ containing $o$, by a standard abuse of notation we denote by the same letters the representatives of $\varrho$ and $W$. (We usually use this abuse of notation for any representative of a germ). We use the notation $W_{s'}:=\varrho ^{-1}(s')$, $s'\in \Omega$.
In what follows we define what we mean by simultaneous embedded resolution of $W$. We give the general definition here, even though, as explained in Remark 1.5, the simultaneous embedded resolutions that we construct in the main theorem are of a special type.
We consider a proper bimeromorphic morphism $\varphi :\widetilde { {\mathbb {C}}^{n+1}_o\times S}\rightarrow {\mathbb {C}}^{n+1}_o\times S$ such that $\widetilde { {\mathbb {C}}^{n+1}_o\times S}$ is formally smooth over $S$, and we denote by $\widetilde {W}^{\rm s}$ and $\widetilde {W}^{\rm t}$ the strict and the total transform of $W$ in $\widetilde { {\mathbb {C}}^{n+1}_o\times S}$, respectively.
Denote by ${\rm Exp}(\varphi )$, the exceptional fiber of $\varphi$.
Definition 1.1 The morphism $\widetilde {W}^{\rm s}\rightarrow W$ is a very weak simultaneous resolution if there exists a sufficiently small open set $\Omega \subset {\mathbb {C}}^{m}$ containing $o$ such that $\widetilde {W}^{\rm s}_{s'}\rightarrow W_{s'}$ is a resolution of singularities for each $s'\in \Omega$.
Definition 1.2 We say that $\widetilde {W}^{\rm t}$ is a normal crossing divisor relative to $S$ if $\widetilde {W}^{\rm t}$ is locally embedded trivial, which is to say that for each $p\in \varphi ^{-1}(o,o)$ there exist sufficiently small open sets $o\in \Omega \subset {\mathbb {C}}^{m}$, $o\in \Omega '\subset {\mathbb {C}}^{n+1}$, $o\in \Omega ''\subset {\mathbb {C}}^{n+1}$ and a neighborhood of $p$,
such that there exists a map $\phi$,
biholomorphic onto its image, such that $\widetilde {W}^t\cap U$ is defined by the ideal $\phi ^{\star }\mathcal {I}$, where $\mathcal {I}=(y_{1}^{a_{1}}\cdots y_{n+1}^{a_{n+1}})$, $y_1,\ldots y_{n+1}$ is a coordinate system at $o$ in $\Omega ''$, and the $a_i$ are non-negative integers.
If $p\in \widetilde {W}^{\rm s}_{o}\cap \varphi ^{-1}(o,o)$, we require that $a_{n+1}=1$ and that $\widetilde {W}^s\cap U$ be defined by the ideal $\phi ^{\star }\mathcal {I}'$, where $\mathcal {I}'=(y_{n+1})$.
Remark 1.3 Assume that $\widetilde {W}^{\rm t}$ is a normal crossing divisor relative to $S$. Then $\mathcal {O}_{\widetilde {W}^{\rm t}}$ is a locally free sheaf of $ {\mathcal {O}_{m}^{s}}$-modules. In particular, the morphism $\widetilde {W}^{\rm t}\rightarrow S$ is flat.
Definition 1.4 We say $\varphi$ is a simultaneous embedded resolution if, in the above notation, the morphism $\widetilde {W}^{\rm s}\rightarrow W$ is a very weak simultaneous resolution and $\widetilde {W}^{\rm t}$ is a normal crossing divisor relative to $S$.
Remark 1.5 In the proof of the main result (Theorem 3.2), the construction of a simultaneous embedded resolution $\varphi$ goes as follows: first we construct an adapted toric birational proper morphism $\pi :\widetilde { {\mathbb {C}}^{n+1}_{o}}\rightarrow {\mathbb {C}}^{n+1}_o$ (here $ {\mathbb {C}}^{n+1}_o$ is endowed with the natural toric structure respecting the chosen coordinates) such that ${\rm Exp(\varphi )}=\pi ^{-1}(o)$. Then $\varphi$ is the product morphism which is defined by
Let us recall that $W$ is defined by
where $h_i\in {\mathcal {O}_{m}^{s}}$, $m\geq 1$, and $g_i\in {\mathcal {O}_{n+1}^{x}}$ such that $h_i(o)=g_i(o)=0$.
Let $\epsilon >0$ (respectively, $\epsilon '>0$) be small enough so that $f, g_1,\ldots,g_l$ (respectively, $h_1,\ldots,h_l$) are defined in the open ball $B_{\epsilon '}(o)\subset {\mathbb {C}}^{n+1}$ (respectively, $B_{\epsilon }(o)\subset {\mathbb {C}}^{m}$), and the singular locus of $W$ is $\{o\}\times B_{\epsilon }(o)$. We say that the deformation of $W$ is embedded topologically trivial (in the classical literature, one often says simply that $F$ is topologically trivial) if, in addition, there exists a homeomorphism
such that $\xi (W)=V'\times B_{\epsilon }(o)$, where $V':=\xi (V)$, that is, $\xi$ trivializes $W$.
The following proposition relates simultaneous embedded resolutions, embedded topologically trivial deformations and $\mu$-constant deformations.
Proposition 1.6 Let $V$ and $W$ be as previously. Assume that $W$ admits a simultaneous embedded resolution such that ${\rm Exp}(\varphi )=\varphi ^{-1}(\{o\}\times S)$. Then:
(i) the deformation $W$ is embedded topologically trivial;
(ii) the deformation $W$ is $\mu$-constant.
Proof. The Milnor number $\mu$ is a topological invariant, hence part (i) implies part (ii), see Theorem 1.4 of [Reference TeissierTei73]
We apply Thom's first isotopy lemma (see [Reference MatherMat12, Proposition 2.11]) to a closed neighborhood $C$ (that we describe in the following) of the compact set $\varphi ^{-1}(o,o)$ and to the restriction of $\varphi$ to $C$.
We begin by describing $C$ and then we show that the hypotheses of the lemma are satisfied.
As $W$ admits a simultaneous embedded resolution, there exists a proper bimeromorphic morphism $\varphi :\widetilde { {\mathbb {C}}^{n+1}_o\times S}\rightarrow {\mathbb {C}}^{n+1}_o\times S$ such that $\widetilde { {\mathbb {C}}^{n+1}_o\times S}$ is formally smooth over $S$, and $\widetilde {W}^t$ is a normal crossing divisor relative to $S$. By Definition 1.2, we have that for each $p\in \varphi ^{-1}(o,o)$ there exist sufficiently small $\epsilon, \epsilon ',\epsilon ''>0$, and a map $\phi _p$ biholomorphic onto its image
that trivializes $\widetilde {W}^t\cap U_p$, where $U_p\subset \varphi ^{-1}(B_{\epsilon '}(o)\times B_{\epsilon }(o))$ is a neighborhood of $p$. Without loss of generality, we assume that $\phi _p$ is bijective.
As $\varphi ^{-1}(o,o)$ is a compact set, there exists a finite set of points $\{ p_1,\ldots,p_l\}\subset \varphi ^{-1}(o,o)$ such that $\varphi ^{-1}(o,o)\subset \Omega =\bigcup _{1}^l U_{p_i}$. Moreover, we may assume that $\epsilon$, $\epsilon '$ do not depend on $p_i$, that $\Omega \subset \varphi ^{-1}(B_{\epsilon '}(o)\times B_{\epsilon }(o))$, and using at most a homothetic transformation that $\epsilon ''$ does not depend on $p$.
The open set $\Omega$ is an open neighborhood of $\varphi ^{-1}(o,o)$, and there exist $\epsilon _0'>0$ and $\epsilon _0>0$ such that $\varphi ^{-1}(B_{\epsilon '_0}(o)\times B_{\epsilon _0}(o))\subset \Omega$. Indeed, if this was not true, there would exist a sequence $x_n\in \varphi ^{-1}(B_{\epsilon '/n}(o)\times B_{\epsilon /n}(o))$ such that $x_n\not \in \Omega$ for all $n>0$. The morphism $\varphi$ is proper, hence $\varphi ^{-1}(\overline {B_{\epsilon '/n}(o)}\times \overline {B_{\epsilon /n'}(o)})$ is a compact set. We may assume that the sequence $x_n$ converges to a point $q\in \varphi ^{-1}(o,o)$ (because $\varphi (x_n)$ converge to $(o,o))$. Then there exists $n_0\in \mathbb {N}$ such that $x_n\in \Omega$ for all $n\geq n_0$, which is a contradiction. Note that we can also assume that $\varphi ^{-1}(\overline {B_{\epsilon '_0}(o)}\times B_{\epsilon _0}(o))\subset \Omega$.
Now, it is well known that there exists $\epsilon '_1>0$ small enough such that for all $0<\delta \leq \epsilon '_1$ the hypersurface $V$ intersects the $(2n+1)$-sphere $S_{\delta }(o):=\partial \overline {B_{\delta }(o)}$ transversally (see [Reference MilnorMil68]). There exists $\epsilon _1>0$ small enough so that the hypersurface $W_s$ intersects the $2n+1$-sphere $S_{\delta }(o)$ transversally for all $s\in B_{\epsilon _1}(o)\subset {\mathbb {C}}^m$.
Without loss of generality, we assume that $\epsilon =\epsilon _0=\epsilon _1$ and $\epsilon '_0=\epsilon '_1$ (we can replace $\epsilon$ by $\epsilon _0$ in the definition of $\Omega$). The set $C:=\varphi ^{-1}(\overline {B_{\epsilon '_0}}(o)\times B_{\epsilon }(o))$ is a closed set of $\Omega$.
We now verify the hypotheses of Thom's first isotopy lemma.
(i) The morphisms
\[ \varphi|_C:C\rightarrow \overline{B_{\epsilon'_0}}(o)\times B_{\epsilon}(o) \]and\[ pr_2:\overline{B_{\epsilon'_0}(o)}\times B_{\epsilon}(o)\rightarrow B_{\epsilon}(o) \]are proper, hence so is $\psi :=pr_2\circ \varphi |_C$.(ii) As each intersection of $(C\cap W^t)\cup \varphi ^{-1}(S_{\epsilon '_o}(o)\times B_{\epsilon }(o))$ is transverse, the set $(C\cap W^t)\cup \varphi ^{-1}(S_{\epsilon '_o}(o)\times B_{\epsilon }(o))$ induces a Whitney stratification of $C$ (obtained by first considering the complement in $(C\cap W^t)\cup \varphi ^{-1}(S_{\epsilon '_o}(o)\times B_{\epsilon }(o))$ and then by the natural stratification of $(C\cap W^t)\cup \varphi ^{-1}(S_{\epsilon '_o}(o)\times B_{\epsilon }(o))$ which is a union of manifolds intersecting transversally). Moreover, as $\widetilde { {\mathbb {C}}^{n+1}_0\times S}$ is formally smooth over $S$, on each stratum $X$ of $C$ the morphism $\psi |_X$ is smooth.
(iii) Observe that, by construction, for each stratum $X$ of $C$ and each $q\in X$ there exists a section $r$ of $\psi$ such that $r(\psi |_{X}(q))=q$:
hence, $\psi |_{X}:X\rightarrow B_{\epsilon }(o)$ is a submersive map.
Let $C_o:=C\cap \psi ^{-1}(o)$ and $X_o:=X\cap \psi ^{-1}(o)$, where $X$ is a stratum of $C$. Thom's first isotopy lemma assures us that there exists $\epsilon >0$ small enough and a homeomorphism
such that $\xi _o(X)=X_o\times {B}_{\epsilon }(o)$, see Proposition $11.1$ and Corollary $10.3$ of [Reference MatherMat12]. Then the morphism $\xi _0$ trivializes simultaneously
We denote by $\varphi _o$ the morphism obtained by restricting $\varphi$ to the special fiber.
Consider the morphism
Then for small enough $\epsilon '_0$ and $\epsilon$, the map
is the desired trivialization.
1.0.3 On the main result of the article
Keep the notation of the previous sections. Recall that $W$ is a deformation of $V$ over $S:= {\mathbb {C}}^m_o$ given by $F$. In [Reference OkaOka89], Oka proved that if $W$ is a non-degenerate $\mu$-constant deformation of $V$ that induces a negligible truncation of the Newton boundary then $W$ admits a very weak simultaneous resolution. However, if the method of proof used is observed with detail, what is really proved is that $W$ admits a simultaneous embedded resolution in the special case when
Intuitively one might think that the condition that $W$ admit a simultaneous embedded resolution is more restrictive than the condition that $W$ is a $\mu$-constant deformation. However, this intuition is wrong in the case of Newton non-degenerate $\mu$-constant deformations. More precisely, in this article we prove the following result.
Theorem Assume that $W$ is a Newton non-degenerate deformation. Then the deformation $W$ is $\mu$-constant if and only if $W$ admits a simultaneous embedded resolution.
Observe that if $W$ admits a simultaneous embedded resolution it follows directly from Proposition 1.6 that $W$ is a $\mu$-constant deformation. The converse of this is what needs to be proved.
From this theorem and Proposition 1.6 we obtain the following corollary.
Corollary 1.7 Let $W$ be a Newton non-degenerate $\mu$-constant deformation. Then $W$ is topologically trivial.
The result of the corollary was already obtained in Theorem 1.1 of [Reference AbderrahmaneAbd16].
It was pointed out to us by the referee that Corollary 1.5 follows from the following two known statements:
(i) every small Newton-non-degenerate deformation is a pullback from a linear one (that is, a deformation of type $f(x) + sg(x)$);
(ii) every $\mu$-constant family of isolated hypersurface singularities of type $f(x) + sg(x)$, is topologically trivial; this is a result of Parusinki (Corollary $2.1$ of [Reference ParusinskiPar99]).
In the general case, for $n\neq 2$ it is known that if $W$ is a $\mu$-constant deformation, then the deformation $W$ is topologically trivial (see [Reference Lê Dũng and RamanujamLDR76]). The case $n=2$ is a conjecture (the Lê–Ramanujan conjecture).
Beyond this article, the main result here initiates a new approach to the Lê–Ramanujam conjecture. In characteristic zero every singularity can be embedded in a higher-dimensional affine space in such a way that it is Newton non-degenerate in the sense of Khovanskii or Schön (this is a possible reading of a result of Tevelev, answering a question of Teissier, see [Reference TeissierTei14], [Reference TevelevTev14] and [Reference MourtadaMou17]). Note that Schön (Newton non-degenerate in the sense of Khovanskii) is the notion that generalizes Newton non-degenerate singularities to higher codimensions, and guarantees the existence of embedded toric resolutions for singularities having this property. For example, the plane curve singularity $(\mathcal {C},o)$ embedded in $ {\mathbb {C}}^{2}_{o}$ via the equation $(x_2^2-x_1^3)^2-x_1^5x_2=0$ is degenerate with respect to its Newton polygon; but embedded in $ {\mathbb {C}}^{3}_{o}$ via the equations $x_3-(x_2^2-x_1^3)=0$ and $x_3^2 -x_1^5x_2=0,$ it is non-degenerate in the sense of Khovanskii [Reference Bivia-AusinaBA07, Reference MourtadaMou17, Reference NguyenNgu20]. Now by [Reference NguyenNgu20] (see also [Reference Bivia-AusinaBA07]), we can compute its Milnor number using mixed Newton numbers. Then the idea is to study the monotonicity of the mixed Newton number and to prove a generalization of Theorems 2.4 and 2.25. This should allow us to generalize the main theorem of this article for an adapted embedding and then to apply part (i) of Proposition 1.6. This idea is a research project that, while not developed in the rest of this paper, we nevertheless find important to mention.
The theorem has an interesting implication to spaces of $m$-jets. Let $ {\mathbb {K}}$ be a field and $Y$ a scheme over $ {\mathbb {K}}$. We denote by $Y\!-\!\mathcal {S}ch$ (respectively, $\mathcal {S}et$) the category of schemes over $Y$ (respectively, sets), and let $X$ be a $Y$-scheme. It is known that the functor $Y\!-\!\mathcal {S}ch\rightarrow \mathcal {S}et:Z\mapsto {\operatorname {Hom}}_Y(Z\times _{ {\mathbb {K}}} {\operatorname {Spec}} {\mathbb {K}}[t]/(t^{m+1}), X)$, $m\geq 1$, is representable. More precisely, there exists a $Y$-scheme, denoted by $\operatorname {X(Y)}_{m}$, such that $ {\operatorname {Hom}}_Y(Z\times _{ {\mathbb {K}}} {\operatorname {Spec}} {\mathbb {K}}[t]/(t^{m+1}), X)\cong {\operatorname {Hom}}_{Y}(Z,\operatorname {X(Y)}_{m})$ for all $Z$ in $Y\!-\!\mathcal {S}ch$. The scheme $\operatorname {X(Y)}_{m}$ is called the space of m-jets of $X$ relative to $Y$. For more details see [Reference VojtaVoj07] or [Reference Leyton-ÁlvarezLA18]. Let us assume that $Y$ is a reduced $ {\mathbb {K}}$-scheme, and let $Z$ be a $Y$-scheme. We denote by $Z_{{\rm red}}$ the reduced $Y$-scheme associated to $Z$.
Corollary 1.8 Let $S= {\mathbb {C}}_{0}$ and let $W$ be a non-degenerate $\mu$-constant deformation. The structure morphism $(\operatorname {W(S)}_{m})_{{\rm red}}\rightarrow S$ is flat for all $m\geq 1$.
Proof. By the previous theorem $W$ admits an embedded simultaneous resolution. Hence, the corollary is an immediate consequence of Theorem $3.4$ of [Reference Leyton-ÁlvarezLA18].
Finally, we comment on the organization of the article. In § 2 we study geometric properties of pairs of Newton polyhedra that have the same Newton number. This allows us to construct the desired simultaneous resolution. In this section, we give an affirmative answer to the conjecture presented in [Reference Brzostowski, Krasiński and WalewskaBKW19]. This result together with Theorem 2.4 (see [Reference FuruyaFur04]) is a complete solution to an Arnold problem (No. 1982-16 in his list of problems, see [Reference ArnoldArn05]) in the case of convenient Newton polyhedra. In § 3 we prove the main result of the article. Finally, in § 4 we study properties of degenerate $\mu$-constant deformations. The main result of this section is Proposition 4.2, which is a kind of analogue to the existence of a good apex (see Definition 2.21).
2. Preliminaries on Newton polyhedra
In this section we study geometric properties of pairs of Newton polyhedra having the same Newton number, one contained in the other. In this article we study the deformations of hypersurfaces $ {\mathbb {C}}^{n+1}_{o}$, whereby the natural things would be to study polytopes in $ {\mathbb {R}}^{n+1}$, $n\geq 1$. Nevertheless, in order to avoid complicating the notations unnecessarily, we work with polytopes in $ {\mathbb {R}}^{n}$, $n\geq 2$.
Given an affine subspace $H$ of $ {\mathbb {R}}^n$, a convex polytope in $H$ is a non-empty set $P$ given by the intersection of $H$ with a finite set of half spaces of $ {\mathbb {R}}^n$. In particular, a compact convex polytope can be seen as the convex hull of a finite set of points in $ {\mathbb {R}}^n$. The dimension of a convex polytope is the dimension of the smallest affine subspace of $ {\mathbb {R}}^n$ that contains it. We say that $P$ is a polyhedron (respectively, compact polyhedron) if $P$ can be decomposed into a finite union of convex (respectively, compact convex) polytopes of disjoint interiors. We will say that $P$ is of pure dimension $n$ if $P$ is a finite union of $n$-dimensional convex polytopes. A hyperplane $K$ of $ {\mathbb {R}}^n$ is supporting $P$ if one of the two closed half spaces defined by $K$ contains $P$. A subset $F$ of $P$ is called a face of $P$ if it is either $\emptyset$, $P$ itself, or the intersection of $P$ with a supporting hyperplane. A face $F$ of $P$ of dimension $0\leq d \leq \dim (P)-1$ is called $d$-dimensional face. In the case that $d$ is $0$ or $1$, $F$ is called a vertex or edge, respectively.
An $n$-dimensional simplex $\Delta$ is a compact convex polytope generated by $n+1$ points of $ {\mathbb {R}}^n$ in general position.
Given an $n$-dimensional compact polyhedron $P\subset {\mathbb {R}}^{n}_{\geq 0}$ , the Newton number of $P$ is defined by
where $V_n(P)$ is the volume of $P$, $V_{k}(P)$, $1\leq k\leq n-1$, is the sum of the $k$-dimensional volumes of the intersection of $P$ with the coordinate planes of dimension $k$, and $V_0(P)=1$ (respectively, $V_0(P)=0$) if $o\in P$ (respectively, $o\notin P$), where $o$ is the origin of $ {\mathbb {R}}^n$. In this section, we are interested in studying the monotonicity of the Newton Number, we always consider the case when $P$ is compact.
Let $I\subset \{1,2,\ldots,n\}$. We define the following sets:
Given a polyhedron $P$ in $ {\mathbb {R}}^n$, we write $P^I:=P\cap {\mathbb {R}}^{I}$. Consider an $n$-dimensional simplex $\Delta \subset {\mathbb {R}}^n_{\geq 0}$. A full supporting coordinate subspace of $\Delta$ is a coordinate subspace $ {\mathbb {R}}^{I}\subset {\mathbb {R}}^n$ such that $\dim \Delta ^I=|I|$. In [Reference FuruyaFur04], Furuya proved that there exists a unique full-supporting coordinate subspace of $\Delta$ of minimal dimension. We call this subspace the minimal full-supporting coordinate subspace (m.f.-s.c.s.) of $\Delta$.
We denote by $\operatorname {Ver}(P)$ the set of vertices of $P$.
The next result gives us a way of calculating the Newton number of certain polyhedra using projections.
Proposition 2.1 [Reference FuruyaFur04]
Let $o\notin P\subset {\mathbb {R}}^{n}_{\geq 0}$ be a compact polyhedron that is a finite union of $n$-simplices $\Delta _i$, $1\leq i\leq m$, that satisfy
Assume that there exists $I\subset \{1,2,\ldots,n\}$ such that $ {\mathbb {R}}^I$ is the m.f.-s.c.s. of $\Delta _i$ and $P^I=\Delta _{i}^{I}$ for all $1\leq i\leq m$. Then $\nu (P)= |I|!V_{|I|}(P^I)\nu (\pi _I(P))$ where $\pi _I: {\mathbb {R}}^n\rightarrow {\mathbb {R}}_{I}$ is the projection map.
Let $\mathcal {E}:=\{e_1,e_2,\ldots,e_n\}\subset {\mathbb {Z}}_{\geq 0}^n$ be the standard basis of $ {\mathbb {R}}^n$. Let
be a polyhedron of pure dimension $n$. Consider the following conditions:
(i) $o\in P$;
(ii) $P^J$ is homeomorphic to a $|J|$-dimensional closed disk for each $J\subset \{1,\ldots,n\}$;
(iii) Let $I\subset \{1,\ldots,n\}$ be a non-empty subset; if $(\alpha _1,\ldots,\alpha _n)\in \operatorname {Ver}(P)$, then for each $i\in I$ we must have either $\alpha _i\geq 1$ or $\alpha _i=0$ (recall that the $\alpha _i$ are real numbers that need not be integers).
We say that $P$ is pre-convenient (respectively, $I$-convenient) if it satisfies conditions (i) and (ii) (respectively, conditions (i), (ii), and (iii)). In the case when $I:=\{1,\ldots,n\}$ we simply say that $P$ is convenient instead of $I$-convenient.
Given a closed discrete set $S\subset {\mathbb {R}}^{n}_{\geq 0} {\setminus} \{o\}$, denote by $\Gamma _{+}(S)$ the convex hull of the set $\bigcup _{\alpha \in S}(\alpha + {\mathbb {R}}_{\geq 0}^n)$. The polyhedron $\Gamma _{+}(S)$ is called the Newton polyhedron associated to $S$. The Newton boundary of $\Gamma _{+}(S)$, denoted by $\Gamma (S)$, is the union of the compact faces of $\Gamma _{+}(S)$. Let $\operatorname {Ver}(S):=\operatorname {Ver}(\Gamma (S))$ denote the set of vertices of $\Gamma (S)$
Remark 2.2 Note that when we refer to vertices of $\Gamma (S)$ we are speaking about zero-dimensional faces of $\Gamma (S)$. For example, if
then $\operatorname {Ver}(S)=\{(0,3),(2,1), (4,0)\}$.
We say that a closed discrete set $S\subset {\mathbb {R}}^{n}_{\geq 0} {\setminus} \{o\}$ is pre-convenient (respectively, $I$-convenient) if $\Gamma _{-}(S):=\overline { {\mathbb {R}}_{\geq 0}^n{\setminus} \Gamma _{+}(S)}$ is pre-convenient (respectively, $I$-convenient). The Newton number of a pre-convenient closed discrete set $S\subset {\mathbb {R}}^{n}_{\geq 0} {\setminus} \{o\}$ is
Note that this number can be negative. In the case when $P$ is the polyhedron $\Gamma _-(S)$ associated with a closed discrete set $S$, condition (i) holds automatically and condition (ii) can be replaced by the following:
(ii′) for each $e\in \mathcal {E}$ there exists $m>0$ such that $me\in \operatorname {Ver}(S)$.
Consider a convergent power series $g\in {\mathbb {C}}\{x_1,\ldots,x_n\}$:
We define $\Gamma _{+}(g)=\Gamma _+(\operatorname {Supp}(g))$ and $\Gamma (g)=\Gamma (\operatorname {Supp}(g))$. We say that $g$ is a convenient power series if for all $e\in \mathcal {E}$ there exists $m>0$ such that $me\in \operatorname {Supp}(g)$.
Observe that the closed discrete set $\operatorname {Supp}(g)$ is convenient if and only if the power series $g$ is convenient. We use the following notation: $\operatorname {Ver}(g):=\operatorname {Ver}(\operatorname {Supp}(g))$, and $\nu (g)=\nu (\operatorname {Supp}(g))$.
Remark 2.3 Observe that in the case that $S$ is a closed discrete, pre-convenient set, there exists at least one finite subset $S'\subset S$ such that $\Gamma _{+}(S')=\Gamma _{+}(S)$ (in fact, it suffices to consider $S'=\Gamma (S)\cap S$). Nevertheless, it is more comfortable to work with $S$ than with finite choices, above all because in our proofs we eliminate or move points of $S$.
Theorem 2.4 [Reference FuruyaFur04]
Let $P'\subset P$ be two convenient polyhedra. We have $\nu (P)-\nu (P')=\nu (\overline {P{\setminus} P'})\geq 0$, and $\nu (P')\geq 0$.
Corollary 2.5
(i) Let $S$ and $S'$ be two convenient closed discrete subsets of $ {\mathbb {R}}_{\geq 0}^{n}{\setminus} \{o\}$, and assume that $\Gamma _{+}(S)\subsetneq \Gamma _{+}(S')$. We have
\[ 0\leq\nu(S)-\nu(S')=\nu(\overline{\Gamma_{-}(S){\setminus}\Gamma_{-}(S')}). \](ii) Let $S$, $S'$, and $S''$ be three convenient closed discrete subsets of $ {\mathbb {R}}^{n}_{\geq 0}{\setminus} \{o\}$ such that their Newton polyhedra satisfy
\[ \Gamma_{+}(S)\subset \Gamma_{+}(S')\subset\Gamma_{+}(S'') \]and $\nu (S)=\nu (S'')$. Then $\nu (S)=\nu (S')=\nu (S'')$.
For a set $I\subset \{1,\ldots,n\}$, we write $I^c:=\{1,\ldots,n\}{\setminus} I$. The following result gives us a criterion for the positivity of the Newton number of certain polyhedra.
Proposition 2.6 Let $o\notin P$ be a pure $n$-dimensional compact polyhedron such that there exists $I\subset \{1,\ldots,n\}$ such that $\dim (P^J)<|J|$ (respectively, $P^J$ is homeomorphic to a $|J|$-dimensional closed disk) for all $I\not \subset J$ (respectively, $I\subset J$). Assume that if
then for each $i\in I^c$ we have $\beta _i\geq 1$ or $\beta _i=0$. Then there exists a sequence of sets $I\subset I_1,I_2,\ldots,I_m\subset \{1,\ldots,n\}$, and of polyhedra $Z_{i}$, $1\leq i \leq m$, such that:
(i) $P=\bigcup _{i=1}^{m}Z_i$;
(ii) $\nu (P)=\sum _{i=1}^{m}\nu (Z_i)$;
(iii) $\nu (Z_{i})= |I_i|!V_{|I_i|}(Z_{i}^{I_{i}})\nu (\pi _{I_i}(Z_{i}))\geq 0$.
In particular, $\nu (P)\geq 0$.
Given $S\subset {\mathbb {R}}_{\geq }^{n}{\setminus} \{o\}$ and $R\subset {\mathbb {R}}^n_{\geq 0}$, we denote $S(R):=S\cup R$.
Remark 2.7 Let $S$ be a closed discrete subset of $ {\mathbb {R}}^{n}_{\geq 0}{\setminus} \{o\}$ and $\alpha \in {\mathbb {R}}_{>}^{I}$, $I\subset \{1,\ldots,n\}$.
If $\alpha \not \in \Gamma _{+}(S)$, then $P:=\overline {\Gamma _{+}(S(\alpha )){\setminus} \Gamma _{+}(S)}$, $S(\alpha ):=S\cup \{\alpha \}$, is homeomorphic to an $|n|$-dimensional closed disk. Furthermore, by induction on $n$ we obtain that $\dim (P^{J})=|J|$ for all $J\supset I$ if and only if $P^J$ is topologically equivalent to a $|J|$-dimensional closed disk. In addition, we observe that $\dim (P^{J})<|J|$ for all $J\not \supset I$.
Proof. The method of proof that we use is similar to the proof of Theorem $2.3$ of [Reference FuruyaFur04].
As $P$ is a pure $n$-dimensional compact polyhedron, there exists a finite simplicial subdivision $\Sigma$ of $P$ such that:
(i) if $\Delta \in \Sigma$, then $\dim \Delta =n$;
(ii) for all $\Delta \in \Sigma$, $\operatorname {Ver}(\Delta )\subset \operatorname {Ver}(P)$;
(iii) given $\Delta, \Delta '\in \Sigma$, we have $\dim (\Delta \cap \Delta ')< n$ whenever $\Delta \neq \Delta '$.
Let $\mathcal {S}$ be the set formed by all the subsets $I'\subset \{1,\ldots,n\}$ such that there exists $\Delta \in \Sigma$ such that its m.f.-s.c.s. is $ {\mathbb {R}}^{I'}$.
As $\dim P^{J}<|J|$ for all $J\not \supset I$, we obtain that $I'\supset I$ for all $I'\in \mathcal {S}$. We define
Let us consider the set
Given $\sigma _i\in \Sigma ^{I'}$ , let $C_i:=\{\Delta \in \Sigma (I') : \Delta ^I=\sigma _i\}$. Consider the closed set
Observe that given $\alpha \in \sigma _i^{\circ }$ (where $\sigma _i^{\circ }$ is the relative interior of $\sigma _i$), there exists $\epsilon >0$ such that for each $J\supset I'$, we have $B_{\epsilon }(\alpha )\cap Z_{(i,I')}^J=B_{\epsilon }(\alpha )\cap {\mathbb {R}}_{\geq 0}^J$. Indeed, as $P^J$ is topologically equivalent to a $|J|$-dimensional closed disk for all $J\supset I'$, there exists $\epsilon >0$ such that $B_{\epsilon }(\alpha )\cap {\mathbb {R}}_{\geq 0}^J\subset P^J$. Making $\epsilon$ smaller we may assume that $B_{\epsilon }(\alpha )\cap {\mathbb {R}}_{\geq 0}^J \subset Z_{(i,I')}^{J}$. This implies that $\pi _{I'}\big (Z_{(i,I')}\big )$ is a convenient polyhedron in $ {\mathbb {R}}_{I'}$ (remember that if $(\beta _1,\ldots,\beta _n)\in \operatorname {Ver}(P)$, then for each $i\in I^c$ we have $\beta _i\geq 1$ or $\beta _i=0$), from which it follows that $\nu \big (\pi _{I'}\big (Z_{(i,I')}\big )\big )\geq 0$ (see Theorem 2.4). Now using Proposition 2.1 we obtain $\nu \big (Z_{(i,I')}\big )= |I|!V_{|I|}(\sigma _i)\nu \big (\pi _{I'}\big (Z_{(i,I')}\big )\big )\geq 0$.
By construction, we obtain
and
This implies that
Rearranging the indices, we obtain the desired subdivision.
Let $S$ and $S'$ be two closed discrete subsets of $ {\mathbb {R}}_{\geq 0}^{n}{\setminus} \{o\}$ such that
We define $\operatorname {Ver}(S',S):=\operatorname {Ver}(S') {\setminus} \operatorname {Ver}(S)$. The following result tells us where the vertices $\operatorname {Ver}(S',S)$ are found.
Proposition 2.8 Let $S$, $S'$ be two convenient closed discrete subsets of $ {\mathbb {R}}^{n}_{\geq 0}{\setminus} \{o\}$. Suppose that $\Gamma _{+}(S)\subsetneqq \Gamma _{+}(S')$ and $\nu (S)=\nu (S')$. Then
Proof. Let us suppose that $\operatorname {Ver}(S',S)\not \subset ( {\mathbb {R}}^n_{\geq 0}{\setminus} {\mathbb {R}}^n_{>0})$. Let
and $\alpha \in \operatorname {Ver}(S',S){\setminus} W$. Let us consider $S'':=S\cup \{\alpha \}$. As the closed discrete sets $S$, $S'$, and $S''$ are convenient and
we obtain $\nu (S'')=\nu (S)=\nu (S')$ (see Corollary 2.5). Let us prove that this is a contradiction. In effect, by definition of Newton number we have
where $V_{k}:=V_k(\Gamma _{-}(S))$ and $V''_{k}:=V_k(\Gamma _{-}(S''))$ are the ${k}$-dimensional Newton volumes of $\Gamma _{-}(S)$ and $\Gamma _{-}(S'')$, respectively. By construction, $V''_{n}< V_{n}$ and $V'_k=V_k$, $1\leq k\leq n-1$, which implies that $\nu (S'')<\nu (S)$.
If we suppose that $\nu (S')=\nu (S)$, it is not difficult to verify that this equality is not preserved by homotheties of $ {\mathbb {R}}_{\geq 0}^{n}$. The following result describes certain partial homotheties of $ {\mathbb {R}}_{\geq 0}^{n}$ which preserve the equality of the Newton numbers.
Let us consider $D(S,S')=\{I\subset \{1,2,\ldots,n\}:\Gamma _{-}(S)\cap {\mathbb {R}}^{I}\neq \Gamma _{-}(S')\cap {\mathbb {R}}^{I}\}$ and $I(S,S')=\bigcap _{I\in D(S,S')}I$. It may happen that
or
Proposition 2.9 Let $S, S'\subset {\mathbb {R}}_{\geq 0}^{n}{\setminus} \{o\}$ be two pre-convenient closed discrete sets such that $\Gamma _{+}(S)\subset \Gamma _{+}(S')$. Suppose that $\{1,2,\ldots,k\}\subset I(S,S')$, and consider the map
Then $\nu (\varphi _{\lambda }(S'))-\nu (\varphi _{\lambda }(S))=\lambda ^k(\nu (S')-\nu (S))$.
Proof. We use the notation $V_m(S):=V_m(\Gamma _{-}(S))$. Recall that
where $\operatorname {Vol}_{m}(\cdot )$ is the $m$-dimensional volume.
Let $J=\{1,2,\ldots,k\}$. Observe that if $J\not \subset I$, then
which implies that $\operatorname {Vol}_{|I|}(\Gamma _{-}(\varphi _{\lambda }(S))\cap {\mathbb {R}}^{I})=\operatorname {Vol}_{|I|}(\Gamma _{-}(\varphi _{\lambda }(S'))\cap {\mathbb {R}}^{I})$. In particular, if $m< k$ we have $V_m(\varphi _{\lambda }(S))=V_m(\varphi _{\lambda }(S'))$. Let us suppose that $m\geq k$. Then
From this we obtain that $V_m(\varphi _{\lambda }(S'))-V_m(\varphi _{\lambda }(S))=\lambda ^k(V_m(S')-V_m(S))$ and $\nu (\varphi _{\lambda }(S'))-\nu (\varphi _{\lambda }(S))=\lambda ^k(\nu (S')-\nu (S))$.
The following corollary is an analogue of Proposition 2.8 in the pre-convenient case. Remember that given $S\subset {\mathbb {R}}_{\geq }^{n}{\setminus} \{o\}$ and $R\subset {\mathbb {R}}^n_{\geq 0}$, we use the notation $S(R):=S\cup R$.
Corollary 2.10 Let $S\subset {\mathbb {R}}_{\geq 0}^{n}{\setminus} \{o\}$ be a pre-convenient closed discrete set, and $\alpha \in {\mathbb {R}}^n_{>0}$, such that $\Gamma _{+}(S)\subsetneqq \Gamma _{+}(S(\alpha ))$. Then $\nu (S(\alpha ))<\nu (S)$.
Proof. Observe that there exists $\lambda >0$ such that the closed discrete sets $\varphi _{\lambda }(S)$, $\varphi _{\lambda }(S(\alpha ))$ are convenient where $\varphi _{\lambda }$ is the homothety consisting of multiplication by $\lambda$. As $I(S,S(\alpha ))=\{1,\ldots,n\}$, we have
(see Proposition 2.9). By Theorem 2.4, we have $\nu (\varphi _{\lambda }(S(\alpha )))\leq \nu (\varphi _{\lambda }(S))$, hence $\nu (S(\alpha ))\leq \nu (S)$. If
then $\nu (\varphi _{\lambda }(S))=\nu (\varphi _{\lambda }(S(\alpha ))$. This contradicts Proposition 2.8.
Take a set $I\subset \{1,\dots,n\}$.
Corollary 2.11 Let $S$, $S'$, and $S''$ be three $I^c$-convenient closed discrete sets such that $\Gamma _{+}(S)\subset \Gamma _{+}(S')\subset \Gamma _{+}(S'')$. Suppose that
Then $\nu (S)\geq \nu (S')\geq \nu (S'')$.
Proof. Without loss of generality, we may take $I=\{1,\ldots,k\}$. As $S$, $S'$, and $S''$ are $I^c$-convenient, there exists $\lambda >0$ such that after applying the map $\varphi _{\lambda }$ given by $\varphi _{\lambda }(x_1,\ldots,x_n)=(\lambda x_1,\ldots,\lambda x_k,x_{k+1},\ldots, x_n)$, the closed discrete sets $\varphi _{\lambda }(S)$, $\varphi _{\lambda }(S')$, and $\varphi _{\lambda }(S'')$ are convenient.
As $I\subset I(S,S')\cap I(S',S'')$, we have
and
By Theorem 2.4, we obtain $0\leq \nu (S)-\nu (S')$ and $0\leq \nu (S')-\nu (S'')$.
Convention From now until the end of the paper, whenever we talk about a vertex $\gamma$ of a certain polyhedron and an edge (one-dimensional face) of this polyhedron denoted by $E_\gamma$, it should be understood that $\gamma$ is one of the endpoints of $E_\gamma$.
Given $I\subset \{1,2,\ldots,n\}$, let $ {\mathbb {R}}^{I}_{>0}:=\{(x_1,x_2,\ldots,x_n)\in {\mathbb {R}}^{I}:x_i>0\ \text {if}\ i\in I\}$. Let $S\subset {\mathbb {R}}_{\geq }^{n}{\setminus} \{0\}$ be a closed discrete set, and let $\alpha \in {\mathbb {R}}_{>0}^{I}$ be such that
Let $E_{\alpha }$ be an edge of $\Gamma (S(\alpha ))$ such that $\alpha$ is one of its endpoints. Given a set $J$ with
we say that $E_{\alpha }$ is $(I,J)$-convenient if for all
we have $\beta _{i}\geq 1$ for $i\in J{\setminus} I$ and $\beta _i=0$ for $i\in J^{c}$. We say that $E_{\alpha }$ is strictly $(I,J)$-convenient if $E_{\alpha }$ is $(I,J)$-convenient and whenever
there exists $i\in J{\setminus} I$ such that $\beta _i>1$.
Example 2.12 Let $0< a<1$, $S:=\{(2,0), (0,2), (\frac {3}{2}(1-a), 2a)\}$, and $\alpha := (\frac {3}{2}, 0)$. For the edge $E_{\alpha }$ of $\Gamma _{+}(S(\alpha ))$ (with endpoints $(\frac {3}{2}, 0)$ and $(0,2)$), we have
Then $E_{\alpha }$ is $(\{1\}, \{1,2\})$-convenient (respectively, strictly $(\{1\}, \{1,2\})$-convenient) if and only if $\frac {1}{2}\leq a <1$ (respectively, $\frac {1}{2}< a <1$).
Example 2.13 Let $0< a<1$, $S:=\{(1,0,0), (0,2,0), (\frac {3}{4}(1-a), 2a,0), (0,0,1)\}$, and $\alpha := (\frac {3}{4}, 0,0)$. For the edge $E_{\alpha }$ of $\Gamma _{+}(S(\alpha ))$ (with endpoints $(\frac {3}{4}, 0,0)$ and $(0,2,0)$) we have
Then $E_{\alpha }$ is $(\{1\}, \{1,2\})$-convenient (respectively, strictly $(\{1\}, \{1,2\})$-convenient) if and only if $\frac {1}{2}\leq a <1$ (respectively, $\frac {1}{2}< a <1$).
The following proposition allows us to eliminate certain vertices.
Proposition 2.14 Let $S\subset {\mathbb {R}}_{\geq 0}^{n}{\setminus} \{0\}$ be an $I^{c}$-convenient closed discrete set, $J$ a set such that $I\subsetneqq J\subset \{1,\ldots,n\}$, and $\alpha \in {\mathbb {R}}^{I}_{>0}$ such that
and $\nu (S(\alpha ))=\nu (S)$. Assume that at least one of the following conditions are satisfied:
(i) $\alpha '\in \overline {\Gamma _{+}(S(\alpha )){\setminus} \Gamma _{+}(S)}\cap {\mathbb {R}}^I$;
(ii) $\alpha '\in \overline {\Gamma _{+}(S(\alpha )){\setminus} \Gamma _{+}(S)}\cap {\mathbb {R}}_{>0}^J$ and there exists a strictly $(I,J)$-convenient edge $E_{\alpha }$ of $\,\Gamma (S(\alpha ))$.
Then $\nu (S(\alpha '))=\nu (S)$.
Remark 2.15 In Example 2.12 we have $\nu (S)= \nu (S(\alpha ))$ if and only if
In particular, if $\nu (S)= \nu (S(\alpha ))$ then there are no strictly $(\{1\}, \{1,2\})$-convenient edges for $\alpha$.
Observe that for each $0< a<1$, if $\alpha ' \in ( \Gamma _{+}(S(\alpha )){\setminus} \Gamma _{+}(S))\cap {\mathbb {R}}^2_{>0}$, then $\mu (S)\neq \mu (S(\alpha '))$.
In Example 2.13 we have $\nu (S)= \nu (S(\alpha ))$ for all $0< a<1$. Furthermore, for all $\alpha '\in \overline {\Gamma _{+}(S(\alpha )){\setminus} \Gamma _{+}(S)}\cap {\mathbb {R}}^{\{1,2\}}$ we have $\nu (S(\alpha '))=\nu (S)$, which indicates that the hypotheses of the preceding proposition are just sufficient.
Observe that $\overline {\Gamma _{+}(S(\alpha )){\setminus} \Gamma _{+}(S)}\cap {\mathbb {R}}^{\{1,2\}}$ is the region of the plane bounded by the triangle of the vertices $(\frac {3}{4},0,0)$,$(2,0,0)$ and $(\frac {3}{4}(1-a), 2a,0)$).
Proof. Let us assume that $\alpha '\in \overline {\Gamma _{+}(S(\alpha )){\setminus} \Gamma _{+}(S)}\cap {\mathbb {R}}^{I}$. We may assume that $\Gamma _{+}(S)\subsetneqq \Gamma _{+}(S(\alpha '))\subsetneqq \Gamma _{+}(S(\alpha ))$ (otherwise there is nothing to prove).
Observe that the closed discrete sets $S$, $S(\alpha ')$, and $S(\alpha )$ are $I^c$-convenient and
Using Corollary 2.11, we obtain $\nu (S)=\nu (S(\alpha '))=\nu (S(\alpha ))$. This completes the proof in case (i).
Next, assume that case (ii) holds. Consider a strictly $(I,J)$-convenient edge $E_{\alpha }$ of $\Gamma (S(\alpha ))$. Let $\beta :=(\beta _1,\ldots,\beta _n)\in E_{\alpha }\cap \operatorname {Ver}(S)$. Let $E'\subset E_{\alpha }$ be the line segment with endpoints $\alpha$ and $\beta$. Without loss of generality, we may assume that $E'\cap \operatorname {Ver}(S)=\{\beta \}$. As $E_{\alpha }$ is strictly $(I,J)$-convenient, there exists $i\in J{\setminus} I$ such that $\beta _i>1$. Let $\delta >0$ be sufficiently small so that $\beta _i-\delta \geq 1$ and let $\beta '\in {\mathbb {R}}_{\geq 0}^I$ be such that $\gamma :=\beta -\delta e_i+\beta '\in \Gamma (S(\alpha ))\cap {\mathbb {R}}^J_{>0}$. Then
Observe that the closed discrete sets $S$s, $S(\gamma )$, and $S(\alpha )$ are $I^c$-convenient and
Then $\nu (S(\gamma ))=\nu (S(\alpha ))=\nu (S)$.
If $\alpha '\in \overline { \Gamma _{+}(S(\gamma )){\setminus} \Gamma _{+}(S)}\cap {\mathbb {R}}^{J}$, we have
The closed discrete sets $S$, $S(\alpha ')$, and $S(\gamma )$ are $J^{c}$-convenient and
Then $\nu (S(\alpha '))=\nu (S(\gamma ))=\nu (S)$.
We still need to study the case $\alpha '\in (\Gamma _{+}(S(\alpha )){\setminus} \Gamma _{+}(S(\gamma )))\cap {\mathbb {R}}^J_{>0}$.
Consider the compact set $C:=\overline {(\Gamma _{+}(S(\alpha ')){\setminus} \Gamma _{+}(S))}\cap {\mathbb {R}}^J$, and the map
where $V_m(S(\tau )):=V_m(\Gamma _{-}(S(\tau ))))$. The map $\nu _S$ is continuous in $C$. In effect, recall that $V_m(S(\tau ))=\sum _{|I'|= m}\operatorname {Vol}_{m}(\Gamma _{-}(S(\tau ))\cap {\mathbb {R}}^{I'})$. Hence,
where
The function $V: {\mathbb {R}}^J\rightarrow {\mathbb {R}};\tau \mapsto V(\tau )$ is continuous, because each summand is continuous in $ {\mathbb {R}}^J$. The function $V':C\rightarrow {\mathbb {R}};\tau \mapsto V'(\tau )$ is constant, because $\Gamma _{-}(S(\alpha '))\cap ( {\mathbb {R}}_{\geq 0}^J{\setminus} {\mathbb {R}}_{>0}^J) =\Gamma _{-}(S) \cap ( {\mathbb {R}}_{\geq 0}^J{\setminus} {\mathbb {R}}_{>0}^J)$. Then each $V_{m}(S(\tau ))$ is continuous in $\tau \in C$, which implies that the function $\nu _S$ is continuous in $C$.
Let us assume that $\alpha '\in (\Gamma _{+}(S(\alpha )){\setminus} \Gamma _{+}(S(\gamma )))\cap {\mathbb {R}}^J_{>0}$ and $\alpha '\notin \Gamma (S(\alpha ))$. Let us suppose that $\nu _S(\alpha ')=\nu (S(\alpha '))\neq \nu (S)$. Let us consider the set $\mathcal {C}:=\{ \tau \in C:\nu _S(\tau ))=\nu _S(\alpha '))\}$. The continuity of $\nu _S$ implies that $\mathcal {C}$ is compact. We define the following partial order on $\mathcal {C}$. For $\tau,\tau '\in \mathcal {C}$ we will say that $\tau \leq \tau '$ if $\Gamma _{+}(S(\tau '))\subset \Gamma _{+}(S(\tau ))$. Let us consider an ascending chain
We prove that this chain is bounded above in $\mathcal {C}$. Let us consider the convex closed set
As $\mathcal {C}$ is compact, the sequence $\{\tau _1,\tau _2,\ldots,\tau _n,\ldots \}$ has a convergent subsequence $\{\tau _{i_1}, \tau _{i_2},\ldots,\tau _{i_n},\ldots \}$. Observe that
where $\tau :=\lim _{n\rightarrow \infty }\tau _{i_n}\in \mathcal {C}$. By definition, $\Gamma \subset \Gamma _{+}(S(\tau ))$, and by construction for each $i\geq 1$ there exists $n\geq 1$ such that $\Gamma _{+}(S(\tau _{i_n}))\subset \Gamma _{+}(S(\tau _{i}))$. Then $\Gamma =\Gamma _{+}(S(\tau ))$, which implies that $\tau _i\leq \tau$ for all $i\geq 1$. By Zorn's lemma $\mathcal {C}$ contains at least one maximal element. Let $\tau \in \mathcal {C}$ be a maximal element. Recall that we consider $\alpha '\notin \Gamma (S(\alpha ))$, and we made the assumption that $\nu (S(\alpha '))\neq \nu (S)$. Hence, $\tau \notin (\overline { \Gamma _{+}(S(\gamma )){\setminus} \Gamma _{+}(S)})\cap {\mathbb {R}}^{J}$.
Observe that for all $\alpha ''\in \Gamma _{+}(S(\alpha ))$ we have
As the closed discrete sets $S$, $S(\alpha '')$, and $S(\gamma,\alpha '')$ are $I^{c}$-convenient and
we obtain $\nu (S(\gamma,\alpha ''))=\nu (S(\alpha ''))=\nu (S)$.
As $\tau \notin \overline {\Gamma _{+}(S(\gamma )){\setminus} \Gamma _{+}(S)}\cap {\mathbb {R}}^{J}$ and $\gamma \in \Gamma (S(\alpha ))$, there exists a relatively open subset $\Omega$ of the relative interior of $\overline {\Gamma _{+}(S(\alpha )){\setminus} \Gamma _{+}(S)}\cap {\mathbb {R}}^{I}$ such that $\tau$ belongs to the relative interior of
for all $\alpha ''\in \Omega$. We obtain
The closed discrete sets $S(\alpha '')$, $S(\tau,\alpha '')$, and $S(\gamma,\alpha '')$ are $J^{c}$-convenient, and
Hence, $\nu (S(\tau,\alpha ''))=\nu (S(\alpha ''))=\nu (S)$.
Given an edge $E_{\tau }$ of $\Gamma (S(\tau ))$ that connects $\tau$ with a vertex in $\operatorname {Ver}(\Gamma (S))$, let $E'_{\tau }$ be the subsegment of $E_{\tau }$ containing $\tau$ such that $|E'_{\tau }\cap \operatorname {Ver}(S)|=1$. We choose $\alpha ''\in \Omega '$ such that for each edge $E_{\tau }$ of $\Gamma (S(\tau ))$ connecting $\tau$ with an element of $\operatorname {Ver}(\Gamma (S))$ we have $\dim (E'_{\tau }\cap \Gamma (S(\alpha '')))=0$. In other words, no subsegment of $E'_{\tau }$ is contained in the Newton boundary $\Gamma (S(\alpha ''))$.
Let us consider the compact polyhedron $P:=\overline {\Gamma _{+}(S(\tau,\alpha '')){\setminus} \Gamma _{+}(S(\alpha ''))}$. Observe that $\nu (P)=0$ (see Theorem 2.4).
Given the choice of $\alpha ''$, there exists $\tau '\in P$ such that
and $Q_0:=\overline {(\Gamma _{+}(S(\tau )){\setminus} \Gamma _{+}(S(\tau ')))}\subset P$ (it is for achieving the last inclusion that the choice of $\alpha ''$ is really important).
Let $Q_1:=\overline {P{\setminus} Q_0}$. As $\dim (Q_0^{J'}\cap Q_{1}^{J'})<|J'|$, for all $J'\subset \{1,\ldots,n\}$, we obtain $\nu (P)=\nu (Q_0)+\nu (Q_1)$. The polyhedra $ {Q}_0$ and $ {Q}_1$ satisfy the hypotheses of Proposition 2.6. In effect, we have the following.
(i) By construction $Q_0$ and $Q_1$ are pure $n$-dimensional compact polyhedra and $o\not \in P=Q_0\cup Q_1$.
(ii) Recall that $\tau \in {\mathbb {R}}^J_{>0}$. The polyhedron $P$ satisfies
\[ \dim(P^{J'})<|J'|\quad\text{for all }J'\not \supset J, \]which implies $\dim (Q_0^{J'})<|J'|$ and $\dim (Q_1^{J'})<|J'|$ for all $J'\not \supset J$.(iii) Now we verify that $Q_{0}^{J'}$ is homeomorphic to a $|J'|$-dimensional closed disk for all $J'\supset J$. As $S$ is $I^c$-convenient and $\tau \in {\mathbb {R}}^J_{>0}$, we have $\dim (Q_{0}^{J'})=|J'|$ for each $J'\supset J$. By Remark 2.7 we obtain that $Q_{0}^{J'}$ is homeomorphic to a $|J'|$-dimensional closed disc. The proof for $Q_{1}^{J'}$ is analogous to the proof for $Q_{0}^{J'}$.
(iv) As $S$ is $I^c$-convenient (in particular, $J^c$-convenient), we obtain that if $(\beta _1,\ldots,\beta _n)\in \operatorname {Ver}(P)$, then for each $i\in J^c$ we have $\beta _i\geq 1$ or $\beta _i=0$. This property is inherited by $Q_0$ and $Q_1$.
By Proposition 2.6 we have $\nu (Q_0)\geq 0$, $\nu (Q_1)\geq 0$. As $\nu (P)=0$, we obtain $\nu (Q_0)=\nu (Q_1)\!=\! 0$. We have $\tau <\tau '\in \mathcal {C}$, which contradicts the maximality of $\tau$ in $\mathcal {C}$. As a consequence, we obtain $\nu (S(\alpha '))=\nu (S)$.
Now let us suppose that $\alpha ' \in \Gamma (S(\alpha ))\cap {\mathbb {R}}_{>0}^J$, and let $v\in {\mathbb {R}}^J_{>0}$. For $\epsilon >0$ small enough $\alpha _{\epsilon }:=\alpha '+\epsilon v$ belongs to the relative interior of $\Gamma _{+}(S(\alpha ')){\setminus} \Gamma _{+}(S)$. For the continuity of $\nu _S$ in $C:=\overline {(\Gamma _{+}(S(\alpha ')){\setminus} \Gamma _{+}(S))}\cap {\mathbb {R}}^J$ we obtain
which implies that $\nu (S(\alpha '))=\nu (S)$.
Corollary 2.16 Let $I\subsetneqq J:= \{1,\ldots,n\}$. Let $S, S'\subset {\mathbb {R}}_{\geq 0}^{n}{\setminus} \{o\}$ be two convenient closed discrete sets such that $\Gamma _{+}(S)\subsetneqq \Gamma _{+}(S')$ and $\nu (S)=\nu (S')$. Suppose that there exists $\alpha \in \operatorname {Ver}(S',S)\cap {\mathbb {R}}^I_{>0}$ and an edge $E_{\alpha }$ of $\Gamma (S)$ that is $(I,J)$-convenient. Then there exists
such that $\beta _i=1$ for all $i\in I^c$.
Proof. Let $R:= \operatorname {Ver}(S',S){\setminus} \{\alpha \}$ and $S(R)=S\cup R$. The closed discrete sets $S,$ $S(R)$ and $S'$ are convenient, and $\Gamma _{+}(S)\subset \Gamma _{+}(S(R))\subsetneqq \Gamma _{+}(S')$. Then
We argue by contradiction. If there is no $(\beta _1,\ldots,\beta _n)$ as in the corollary, then the edge $E_{\alpha }$ is strictly $(I,J)$-convenient. By Proposition 2.14, for all $\alpha '\in \overline {\Gamma _{+}(S(\alpha ){\setminus} \Gamma _{+}(S)}\cap {\mathbb {R}}_{>0}^n$ we have
which contradicts Proposition 2.8.
The following proposition allows us to fix a special coordinate hyperplane and gives information about the edges not contained in the hyperplane that contain a vertex of interest belonging to the hyperplane.
Proposition 2.17 Let $S,S'\subset {\mathbb {R}}_{\geq 0}^{n}{\setminus} \{o\}$ be two convenient closed discrete sets such that $\Gamma _{+}(S)\subsetneqq \Gamma _{+}(S')$ and $\nu (S)=\nu (S')$. Let us suppose that
Then there exists $i\in I^c$ such that for all the edges $E_{\alpha }$ of $\Gamma (S')$ not contained in $ {\mathbb {R}}_{\{i\}}$ there exists $(\beta _1,\ldots,\beta _n)\in \operatorname {Ver}(S)\cap E_{\alpha }$ such that $\beta _i=1$.
Proof. First we prove the following lemma.
Lemma 2.18 Let $S\subset {\mathbb {R}}_{\geq 0}^{n} {\setminus} \{o\}$ be an $I^c$-convenient closed discrete set and
such that $\nu (S)=\nu (S(\alpha ))$. Then there exists $i\in I^c$ such that for each edge $E_{\alpha }$ of $\Gamma (S(\alpha ))$ not contained in $ {\mathbb {R}}_{\{i\}}$ there exists $(\beta _1,\ldots,\beta _n)\in \operatorname {Ver}(S)\cap E_{\alpha }$ such that $\beta _i=1$.
Proof of Lemma 2.18. By Corollary 2.10, we have $|I|< n$. Let $k$ be the greatest element of $\{1,\dots,n-1\}$ such that the lemma is false for some $I$ with $|I|=k$. In other words, for all $i\in I^c$ there exists an edge $E_{\alpha }$, not contained in $ {\mathbb {R}}_{\{i\}}$, such that for all $(\beta _1,\ldots,\beta _n)\in \operatorname {Ver}(S)\cap E_{\alpha }$ we have $\beta _i>1$. Let $J\subset \{1,\ldots,n\}$ be a set of the smallest cardinality such that $E_{\alpha }\subset {\mathbb {R}}^J$. Then $E_{\alpha }$ is a strictly $(I,J)$-convenient edge. Using Proposition 2.14 we obtain that for all $\alpha '\in \overline {\Gamma _{+}(S(\alpha ){\setminus} \Gamma _{+}(S)}\cap {\mathbb {R}}_{>0}^J$ we have $\nu (S(\alpha '))=\nu (S)$. Now let us choose $\alpha '$ sufficiently close to $\alpha$ so that for each edge $E_{\alpha '}$ of $\Gamma (S(\alpha '))$, and $\beta \in E_{\alpha '}\cap \operatorname {Ver}(S)$ adjacent to $\alpha '$ in $E_{\alpha '}$, there exists an edge $E_{\alpha }$ of $\Gamma (S(\alpha ))$ such that $\beta \in E_{\alpha }$. Then the closed discrete sets $S$, $S(\alpha ')$ are $J^c$-convenient and do not satisfy the conclusion of the lemma, which is a contradiction, because $|J|>k$.
The proof of the proposition is by induction on the cardinality of $\operatorname {Ver}(S',S)$. Lemma 2.18 says that the proposition is true whenever $|\!\operatorname {Ver}(S',S)|=1$. Let us assume that the proposition is true for all $S$, $S'$ such that
Let $S$, $S'$ with $|\!\operatorname {Ver}(S',S)|= m\geq 2$ be such that the proposition is false. Then there exists $\alpha \in \operatorname {Ver}(S',S)$ such that for each $i\in I^c$ there exists an edge $E_{\alpha }$ of $\Gamma _{+}(S')$, not contained in $ {\mathbb {R}}_{\{i\}}$, that satisfies the following condition:
() for all $\beta =(\beta _1,\ldots,\beta _n)\in \operatorname {Ver}(S)\cap E_{\alpha }$ we have $\beta _i>1$;
note that condition $(*)$ is vacuously true if
Observe that for each $\alpha '\in \operatorname {Ver}(S',S){\setminus} \{\alpha \}$, we have $|\!\operatorname {Ver}(S',S(\alpha '))|=m-1$ and, by Corollary 2.11, $\nu (S(\alpha '))=\nu (S')$.
First, let us suppose that there exists $i\in I^c$ such that (1) does not hold for the corresponding edge $E_\alpha$. Let us fix $\alpha '\in \operatorname {Ver}(S',S){\setminus} \{\alpha \}$. Then $E_\alpha$ connects $\alpha$ with a vertex $\beta$ of $S$, hence $\alpha '\notin E'$, where $E'\subset E_{\alpha }$ is the line segment with endpoints $\alpha$ and $\beta$. We obtain that the polyhedra $\Gamma _{+}(S(\alpha '))\subsetneqq \Gamma _{+}(S')$ do not satisfy the conclusion of the Proposition, which contradicts the induction hypothesis.
Next, let us suppose that there exists $i\in I^c$ such that (1) is satisfied for the corresponding edge $E_\alpha$. Then $|E_{\alpha }\cap \operatorname {Ver}(S',S)|=2$. Now, take
such that $\alpha '\neq \alpha$. If $\alpha '_i>1$, then the Newton polyhedra
do not satisfy the Proposition and (1) does not hold, which is a contradiction. Hence, $\alpha '_i=1$. Let $\epsilon >0$ be such that
Put
Then $\Gamma _{+}(S)\subsetneqq \Gamma _{+}(S(\alpha '_{\epsilon }))) \subsetneqq \Gamma _{+}(S(R)))\subsetneqq \Gamma _{+}(S')$.
The closed discrete sets $S$, $S(\alpha '_{\epsilon })$, $S(R)$, and $S'$ are convenient. We have
Let us assume that $\epsilon$ is small enough so that there exists an edge $E'_{\alpha }\ni \alpha$ of $\Gamma (S(R))$ such that $\alpha '_{\epsilon }\in E'_{\alpha }$. Then the Newton polyhedra $\Gamma _{+}(S) \subsetneqq \Gamma _{+}(S(R))$ satisfy the preceding case (namely, $\alpha '_i>1$). This completes the proof of the proposition.
Corollary 2.19 Assume given two convenient closed discrete sets
such that $\Gamma _{+}(S)\subsetneqq \Gamma _{+}(S')$ and $\nu (S)=\nu (S')$. Assume that
Then, for the $i\in I^c$ of Proposition 2.17 there exists an edge $E_{\alpha }$ of $\Gamma (S')$, and $(\beta _1,\ldots,\beta _n)\in E_{\alpha }\cap \operatorname {Ver}(S)$, such that $\beta _j=\delta _{ij}$, $j\in I^c$, where $\delta _{ij}$ is the Kronecker delta.
Proof of the corollary By Proposition 2.17 there exists $i\in I^c$ such that for all the edges $E_{\alpha }$ of $\Gamma (S')$, not contained in $ {\mathbb {R}}_{\{i\}}$, there exists
such that $\beta _i=1$. As the set $S$ is convenient, there exists $m>1$ such that $me_i\in \operatorname {Ver}(S)$. Let $J=I\cup \{i\}$. As $\alpha,me_i\in {\mathbb {R}}^J$, there exists a chain of edges of $\Gamma (S')$ connecting $\alpha$ with $me_i$, contained in $ {\mathbb {R}}^J$. The edge $E_\alpha$ belonging to this chain and containing $\alpha$ satisfies the conclusion of the corollary.
Remark 2.20 Using the same idea as in Corollary 2.19, but using Lemma 2.18 instead of Proposition 2.17, we can prove the following fact: let $I\subsetneqq \{1,\ldots,n\}$ and let $S\subset {\mathbb {R}}_{\geq 0}^{n}{\setminus} \{o\}$ be an $I^c$-convenient closed discrete set. Let $\alpha \in {\mathbb {R}}^I_{>0}$ be such that $\Gamma _{+}(S)\subsetneqq \Gamma _{+}(S(\alpha ))$, and $\nu (S)=\nu (S')$. Then for the $i\in I^c$ of Lemma 2.18 there exists an edge $E_{\alpha }$ of $\Gamma (S(\alpha ))$, and $(\beta _1,\ldots,\beta _n)\in E_{\alpha }\cap \operatorname {Ver}(S)$, such that $\beta _j=\delta _{ij}$, $j\in I^c$, where $\delta _{ij}$ is the Kronecker delta.
The following theorem generalizes to all dimensions the main theorem of [Reference Brzostowski, Krasiński and WalewskaBKW19]. In [Reference Brzostowski, Krasiński and WalewskaBKW19] this result is conjectured.
Definition 2.21 Let $S, S'\subset {\mathbb {R}}_{\geq 0}^{n}{\setminus} \{o\}$ be two closed discrete sets such that
$I\subsetneqq \{1,\ldots,n\}$ and $\alpha \in \operatorname {Ver}(S',S)\cap {\mathbb {R}}_{>0}^{I}$. We say that $\alpha$ has an apex if there exists $i\in I^c$ and a unique edge $E_{\alpha }$ of $\Gamma (S')$ that contains $\alpha$, and is not contained in $ {\mathbb {R}}_{\{i\}}$.
In this case the point $\beta \in \operatorname {Ver}(S)\cap E_{\alpha }$ adjacent to $\alpha$ in $E_{\alpha }$ is called the apex of $\alpha$. We say that an apex, $\beta :=(\beta _1,\ldots,\beta _n)$, is good if $\beta _j=\delta _{ij}$, $j\in I^c$.
Remark 2.22 Let $S\subset {\mathbb {R}}_{\geq 0}^{n}{\setminus} \{o\}$ be a convenient closed discrete set,
and $\alpha \in {\mathbb {R}}_{>0}^I$ such that $\Gamma _{+}(S)\subsetneqq \Gamma _{+}(S(\alpha ))$. The condition that $\alpha$ has a good apex $\beta \in {\mathbb {R}}^{I\cup \{i\}}_{>0}$, $i\in I^c$, is equivalent to $P:=\overline {\Gamma _{+}(S(\alpha )){\setminus} \Gamma _{+}(S)}$ being a pyramid with apex $\beta$ and base $P\cap {\mathbb {R}}_{\{i\}}$.
Example 2.23 For example, in the case of the $\mu$-constant deformation of Briançon–Speder convenient version (see [Reference Briançon and SpederBS75]),
the pyramid is formed by the base with vertices $(5,0,0)$, $(0,8,0)$, $\alpha =(1,6,0)$, and the good apex $\beta =(0,7,1)$. Note that the vertices $(5,0,0)$, $(0,7,1)$, $(1,6,0)$, $(0,0,15)$ are coplanar.
Remark 2.24 The element $i\in I^{c}$ of condition (ii) of Definition 2.21 may not be unique, and the edge $E_{\alpha }$ is unique only for the chosen $i$. For example, if $S=\{(2,0,0),(0,2,0), (1,0,1), (0,1,1), (0,0,3)\}$ and $\alpha =(0,0,2)$, we have $I=\{3\}$.
If $i=1$, the unique edge $E_{\alpha }$ is the segment between the point $\alpha =(0,0,2)$ and the good apex $\beta =(1,0,1)$.
If $i=2$, the unique edge $E_{\alpha }$ is the segment between the point $\alpha =(0,0,2)$ and the good apex $\beta =(0,1,1)$.
Theorem 2.25 Let $S, S'\subset {\mathbb {R}}_{\geq 0}^{n}{\setminus} \{o\}$ be two convenient closed discrete sets such that $\Gamma _{+}(S)\subsetneqq \Gamma _{+}(S')$. Then $\nu (S)=\nu (S')$ if and only if each $\alpha \in \operatorname {Ver}(S',S)$ has a good apex.
Proof. First we prove the following Lemma.
Lemma 2.26 Let $S\subset {\mathbb {R}}_{\geq 0}^{n}{\setminus} \{o\}$ be a closed discrete set and $\alpha \in {\mathbb {R}}_{>0}^I$, $I\subsetneqq \{1,\ldots,n\}$, such that $\Gamma _{+}(S)\subsetneqq \Gamma _{+}(S(\alpha ))$, and $\alpha$ has a good apex. Then
Proof of Lemma 2.26 Let $\beta$ be a good apex of $\alpha$. Let $i\in I^c$ be such that $\beta \in {\mathbb {R}}_{>0}^{I\cup \{i\}}$.
Given an element $m\in \{1,\dots,n\}$ and $J\subset \{1,\ldots,n\}$ such that $|J|=m$, we use the notation
As $\alpha \in {\mathbb {R}}^{I}_{>0}$, we have
As the apex of $\alpha$ is good, we obtain
which implies that $\nu (S)=\nu (S(\alpha ))$.
Now we prove that if each $\alpha \in \operatorname {Ver}(S',S)$ has a good apex, then
The proof is by induction on the cardinality of $\operatorname {Ver}(S',S)$. Let us assume that the implication is true for all $S$ and $S'$ such that $|\!\operatorname {Ver}(S',S)|< m$. To verify the implication for $|\!\operatorname {Ver}(S',S)|=m$, let $\alpha \in \operatorname {Ver}(S',S)$ and $R=\operatorname {Ver}(S',S){\setminus} \{\alpha \}$. By the induction hypothesis $\nu (S(R))=\nu (S)$ and by Lemma 2.26 we have $\nu (S')=\nu (S(R))$. This proves that $\nu (S')=\nu (S)$.
To finish the proof of the theorem we need the following lemma.
Lemma 2.27 Let $S\subset {\mathbb {R}}_{\geq 0}^{n}{\setminus} \{o\}$ be a closed discrete set and let $\alpha \in {\mathbb {R}}_{>0}^I$, $I\subsetneqq \{1,\ldots,n\}$, be such that $\Gamma _{+}(S)\subsetneqq \Gamma _{+}(S(\alpha ))$. Let us suppose that $S(\alpha )$ is $I^c$-convenient and that $\nu (S(\alpha ))=\nu (S)$. Then $\alpha$ has a good apex.
Proof of Lemma 2.27 Let $i\in I^c$ be as in Remark 2.20. Then there exists $E_{\alpha }$ of $\Gamma (S(\alpha ))$, and $\beta :=(\beta _1,\ldots,\beta _n)\in E_{\alpha }\cap \operatorname {Ver}(S)$, such that $\beta _j=\delta _{ij}$, $j\in I^c$. We want to prove that $\beta$ is a (necessarily good) apex of $\alpha$. Let us assume that $\beta$ is not an apex of $\alpha$, aiming for contradiction. Then there exits another edge $\alpha \in E'_{\alpha }$ de $\Gamma (S(\alpha ))$, and $\beta ':=(\beta '_1,\ldots,\beta '_n)\in E'_{\alpha }\cap \operatorname {Ver}(S)$ adjacent to $\alpha$ in $E'_{\alpha }$ such that $\beta '_i=1$.
Let us consider $\beta '_{\epsilon }:= \beta '+\epsilon e_i$, and the closed discrete set $S^{\epsilon }=(S{\setminus} \{\beta '\})\cup \{\beta '_{\epsilon }\}$, $\epsilon > 0$. Let us assume that $\epsilon$ is small enough so that:
(i) $\operatorname {Ver}(S^{\epsilon })=(\operatorname {Ver}(S){\setminus} \{\beta '\})\cup \{\beta '_{\epsilon }\}$;
(ii) there exists an edge $E_{\alpha }^{\epsilon }$ of $\Gamma (S^{\epsilon }(\alpha ))$ such that $\beta _{\epsilon }\in E_{\alpha }^{\epsilon }\cap \operatorname {Ver}(S^{\epsilon })$ is adjacent to $\alpha$ in $E_{\alpha }^{\epsilon }$.
Let $P^{\epsilon }=\overline {(\Gamma _{+}(S^{\epsilon }(\alpha ){\setminus} \Gamma _{+}(S^{\epsilon }))}$. Let $Q_{0}$ be the convex hull of the set
(observe that $Q_0$ does not depend on $\epsilon$) and $Q^{\epsilon }_1:=\overline {P^{\epsilon }{\setminus} Q_0}$. Recall that $\beta$ satisfies $\beta _j=\delta _{ij}$, $j\in I^c$. Then, using the same idea as in the proof of Lemma 2.26 we obtain $\nu (Q_0)=0$. As $\dim (Q_{0}^{J}\cap (Q^{\epsilon }_1)^{J} )<|J|$ for all $J\subset \{1,\ldots,n\}$, we have
As $S^{\epsilon }(\alpha )$ is $I^c$-convenient, $Q^{\epsilon }_1$ satisfies the hypotheses of Proposition 2.6 (to prove this statement use the same idea as in the proof of Proposition 2.14). Let us consider the sequence
and the polyhedra $Z^{\epsilon }_{j}$, $1\leq j\leq m$, such that:
(i) $Q^{\epsilon }_1=\bigcup _{j=1}^{m}Z^{\epsilon }_{j}$;
(ii) $\nu (Q^{\epsilon }_1)=\sum _{j=1}^{m}\nu (Z^{\epsilon }_{j})$;
(iii) $\nu (Z^{\epsilon }_{j})=|I_{j}|!V_{|I_{j}|}((Z^{\epsilon }_{j})^{I_{j}}) \nu (\pi _{I_{j}}(Z^{\epsilon }_{j}))\geq 0$;
(the existence of these objects is given by Proposition 2.6). For each $j$, $1\le j\le m$, we may choose the family $Z_j^\epsilon$ of polyhedra to vary continuously with $\epsilon$. More precisely, we can choose the $Z_j^\epsilon$ to satisfy the following additional condition: for each $j$, $1\le j\le m$, either $Z_j^\epsilon =Z_j^0$ for all small $\epsilon$ or $\operatorname {Ver}(Z_j^\epsilon )$ differs from $\operatorname {Ver}(Z_j^0)$ in exactly one element, $\beta '_\epsilon \ne \beta '$, for all small $\epsilon >0$. As $i\in I_j$, we have $\pi _{I_j}( \beta '_{\epsilon })=\pi _{I_j}(\beta ')$. This implies that $\nu (\pi _{I_{j}}(Z^{\epsilon }_{j}))$ is independent of $\epsilon$ for all $1\leq {j}\leq m$. For $\epsilon =0$, we have
Hence, $\nu (P^{\epsilon })=\nu (Q_1^{\epsilon })=0$ for $\epsilon$ small enough. Then there exists a set $J$, $\{i\}\cup I\subset J\subset \{1,2,\ldots,n\}$, such that the edge $E^{\epsilon }_{\alpha }$ is strictly $(I,J)$-convenient. By Proposition 2.14, given $\alpha '\in \overline {\Gamma _{+}(S^{\epsilon }(\alpha )){\setminus} \Gamma _{+}(S^\epsilon )}\cap {\mathbb {R}}_{>0}^J$ we have
This proves that $|I|< n-1$: indeed, if $|I|=n-1$, then $\alpha '\in {\mathbb {R}}_{>0}^n$, which contradicts Corollary 2.10.
Let $r$ be the largest element of $\{1,\ldots,n-1\}$ such that the lemma is true for all $I$ such that $|I|>r$. Now let us assume that $|I|=r$. Let us choose $\alpha '$ sufficiently close to $\alpha$ so that for each edge $E_{\alpha '}$ of $\Gamma (S^{\epsilon }(\alpha '))$ and
adjacent to $\alpha '$ in $E_{\alpha '}$, there exists an edge $E_{\alpha }$ of $\Gamma (S^{\epsilon }(\alpha ))$ such that $\beta \in E_{\alpha }$. This implies that $\alpha '$ does not have a good apex, which contradicts the choice of $r$, since $|J|>r$. This completes the proof of the lemma.
Now we can finish the proof of the theorem. We prove that if
then each $\alpha \in \operatorname {Ver}(S',S)$ has a good apex. The proof is by induction on the cardinality of $\operatorname {Ver}(S',S)$. Lemma 2.27 shows that the implication is true for $|\!\operatorname {Ver}(S',S)|=1$. Let us assume that this is true for every pair $(S,S')$ of convenient closed discrete sets such that $|\!\operatorname {Ver}(S',S)|< m$. Let us prove the result for $|\!\operatorname {Ver}(S',S)|=m$. Let $\alpha \in \operatorname {Ver}(S',S)$, $R=\operatorname {Ver}(S',S){\setminus} \{\alpha \}$ and $\alpha _{\epsilon }=(1+\epsilon )\alpha$, where $\epsilon >0$. Then
By Corollary 2.5 we have $\nu (S(\alpha _{\epsilon })(R))=\nu (S(\alpha _{\epsilon }))$. Observe that
By the induction hypothesis, each $\alpha '\in R$ has a good apex $\beta \in \operatorname {Ver}(S(\alpha _{\epsilon }))$ for the inclusion $\Gamma _{+}(S(\alpha _{\epsilon }))\subsetneqq \Gamma _{+}(S(\alpha _{\epsilon })(R))$ of Newton polyhedra. As all the non-zero coordinates of $\alpha _\epsilon$ are strictly greater than one, we have $\beta \neq \alpha _{\epsilon }$, so that $\beta \in \operatorname {Ver}(S)$. We take $\epsilon$ small enough so that for every $\alpha '\in R$ every edge $E_{\alpha '}$ of $\Gamma (S(\alpha _{\epsilon })(R))$ that connects $\alpha '$ with a vertex in $\operatorname {Ver}(S)$ is an edge of $\Gamma (S')$. Thus, every $\alpha '\in R$ has a good apex for the inclusion
of Newton polyhedra.
Now it suffices to verify that $\alpha$ has a good apex for the inclusion
of Newton polyhedra. Let $\epsilon >0$ and put $R_{\epsilon }:=\{(1+\epsilon )\alpha ': \alpha '\in R\}$. Then
By Corollary 2.5 we have $\nu (S(R_{\epsilon }))=\nu (S(R_{\epsilon })(\alpha ))=\nu (S)$. Observe that $\operatorname {Ver}(S(R_{\epsilon })(\alpha ), S(R_{\epsilon }))=\{\alpha \}$. By Lemma 2.27, $\alpha$ has a good apex
As every non-zero coordinate of every element of $R_\epsilon$ is strictly greater than one, we have $\beta \notin R_{\epsilon }$, so that $\beta \in \operatorname {Ver}(S)$. Take $\epsilon$ small enough so that every edge $E_{\alpha }$ of $\Gamma (S(R_{\epsilon })(\alpha ))$ that connects $\alpha$ with a vertex in $\operatorname {Ver}(S)$ is an edge of $\Gamma (S')$. Then $\beta$ is a good apex of $\alpha$ for the inclusion (2), as desired. This completes the proof of the theorem.
We end this section by recalling a result that relates the Milnor number to the Newton number.
If the formal power series $g$ is not convenient, we can define the Newton number $\nu (g)$ of $g$ ($\nu (g)$ could be $\infty$) in the following way. Let $\mathcal {E'}\subset \mathcal {E}$ such that there does not exist $m\in {\mathbb {Z}}_{>0}$, such that $me\in \operatorname {Ver}(g)$. We define the Newton number of $g$ as
where $\mathcal {E'}_m:=\{me:e\in \mathcal {E'}\}$.
Theorem 2.28 [Reference KouchnirenkoKou76]
Let $h\in \mathcal {O}_{n+1}^x$. Then $\mu ( h ) \geq \nu ( h )$. Moreover, $\mu ( h ) = \nu ( h )$ if $h$ is non-degenerate.
Remark 2.29 Let $h\in \mathcal {O}_{n+1}^x$ be non-degenerate and convenient. Then $\mu (h)<\infty$, which implies that $h$ has, at most, an isolated singularity in the origin $o$.
Example 2.30 Consider the following families of non-degenerate deformations:
Observe that $F^1$ is a $\mu$-constant deformation of Briançon–Speder (convenient version), see [Reference Briançon and SpederBS75]. By virtue of Theorem 2.28 and Proposition 2.9, for each $\lambda \geq 1$ the deformation $F^{\lambda }$ is $\mu$-constant.
3. Characterization of Newton non-degenerate $\mu$-constant deformations
First, let us recall some information regarding the Newton fan and toric varieties. Given $S\subset {\mathbb {Z}}_{\geq 0}^{n+1}{\setminus} \{o\}$, consider the support function
where $\Delta := {\mathbb {R}}^{n+1}_{\geq 0}$ is the standard cone and $\langle \cdot,\cdot \rangle$ is the standard scalar product. Let $1\leq i\leq n$ and let $F$ be an $i$-dimensional face of the Newton polyhedron $\Gamma _{+}(S)$. The set $\sigma _F := \{\alpha \in \Delta : \langle \alpha, p\rangle = {\operatorname {h}}_{\Gamma _{+}(S)}(\alpha ),\; \forall p \in F \}$ is a cone, and $\Gamma ^{\star }(S):=\{\sigma _F : F \text { is a face of } \Gamma _{+}(S)\}$ is a subdivision of the fan $\Delta$ (by abuse of notation we denote by $\Delta$ the fan induced by the standard cone $\Delta$). The fan $\Gamma ^{\star }(S)$ is called the Newton fan of $S$. Given a formal power series $g$, we define $\Gamma ^{\star }(g):= \Gamma ^{\star }(\operatorname {Supp}(g))$.
Let be a strict face of the standard cone $\Delta$ and $(\Delta ')^{\circ }$ its interior relative to $\Delta '$. Observe that if there exists $\alpha \in (\Delta ')^{\circ }$ such that $ {\operatorname {h}}_{\Gamma _{+}(\alpha )}=0$, then $\Delta '$ is a cone of the fan $\Gamma ^{\star }(S)$. We say that $\Sigma$ is an admissible subdivision of $\Gamma ^{\star }(S)$ if $\Sigma$ is a subdivision that preserves the above property, which is to say that if there exists $\alpha \in (\Delta ')^{\circ }$ such that $ {\operatorname {h}}_{\Gamma _{+}(S)}(\alpha )=0$, then $\Delta '\in \Sigma$. In the case that the closed discrete set $S$ is convenient, an admissible subdivision of $\Gamma ^{\star }(S)$ is a fan where there are no subdivisions of strict faces of $\Delta$.
Given a fan $\Sigma$, we denote by $X_{\Sigma }$ the toric variety associated to it. Given $\sigma \in \Sigma$, we denote by $X_{\sigma }$ the open affine of $X_{\Sigma }$ associated with the cone $\sigma$. Let $\Sigma '$ be a subdivision of $\Sigma$. It is known that there exists a proper, birational and equivariant morphism $\pi :X_{\Sigma '}\rightarrow X_{\Sigma }$, induced by the subdivision. Given $\sigma '\in \Sigma '$, we write $\pi _{\sigma '}:=\pi |_{X_{\sigma '}}$.
Now we use the notation from § 1.0.1. Let $V$ be a hypersurface of $ {\mathbb {C}}^{n+1}_o$ having a unique isolated singularity at the point $o$. Let us assume that $V$ is given by the equation $f(x)=0$, where $f\in {\mathcal {O}_{n+1}^{x}}$ is irreducible, and let $\varrho :W\rightarrow {\mathbb {C}}^{m}_o$ be a deformation of $V$ given by $F(x,s)\in {\mathbb {C}}\{x_1,\ldots,x_{n+1},s_1,\ldots,s_m\}$.
Without loss of generality we may assume that the germ of analytic function $f$ is convenient. In effect, the Milnor number $\mu (f):=\dim _{ {\mathbb {C}}} {\mathcal {O}_{n+1}^{x}}/ J(f)$ is finite. Hence, for each $e\in \mathcal {E}$ there exists $m \gg 0$ such that $x^{me}$ belongs to the ideal $J(f)$. This implies that the singularities of $f$ and of $f+x^{me}$ have the same analytic type.
Let $s$ be a general point of $ {\mathbb {C}}^m_o$, and let $\Sigma$ be an admissible subdivision of $\Gamma ^{\star }(F_s)$ (not necessarily regular). Denote by $\pi :X_{\Sigma }\rightarrow {\mathbb {C}}^{n+1}$ the morphism given by the subdivision of $\Delta$. Using the morphism $ {\mathbb {C}}_{o}^{n+1}\rightarrow {\mathbb {C}}^{n+1}$ we can consider the base change of $\pi$ and $X_{\Sigma }$ to the base $ {\mathbb {C}}_{o}^{n+1}$. By abuse of notation we denote by $\pi : X_{\Sigma }\rightarrow {\mathbb {C}}_0^{n+1}$ the resulting morphism after the base change.
Let us recall the following known fact. Let $V'$ be a hypersurface of $ {\mathbb {C}}^{n+1}_o$, $n\geq 1$, having a unique isolated singularity at the point $o$. Let us assume that $V'$ is given by the equation $g(x)=0$, where $g\in {\mathcal {O}_{n+1}^{x}}$. Let us assume that $\Sigma$ is a regular admissible subdivision of a Newton fan $\Gamma ^{\star }(g)$. If $g$ is non-degenerate with respect to the Newton boundary, then the morphism $\pi :X_{\Sigma }\rightarrow {\mathbb {C}}^{n+1}_o$ of toric varieties defines an embedded resolution of $V'$ in a neighborhood of $\pi ^{-1}(o)$ (see [Reference VarchenkoVar76], [Reference OkaOka87] or [Reference IshiiIsh07]). This shows that if $\Gamma _{+}(F_s)=\Gamma _{+}(f)$, where $s$ is a general point of $ {\mathbb {C}}^m_o$, and $F$ is a Newton non-degenerate deformation of $f$ (in particular, a $\mu$-constant deformation of $f$ by Theorem 2.28), a regular admissible resolution of the Newton fan defines a simultaneous embedded resolution of $W$. In view of this, for the rest of this section we assume:
(i) $F(x,s)\in {\mathbb {C}}\{x_1,\ldots,x_{n+1},s_1,\ldots,s_m\}$ is a Newton non-degenerate $\mu$-constant deformation of $f$;
(ii) $\Gamma _{+}(F_s)\neq \Gamma _{+}(f)$. In particular, $\operatorname {Ver}(F_s,f):=\operatorname {Ver}(F_s) {\setminus} \operatorname {Ver}(f)\neq \emptyset$.
Let $\varphi :X_{\Sigma }\times {\mathbb {C}}^m_o\rightarrow {\mathbb {C}}^{n+1}_o\times {\mathbb {C}}^{m}_o$ be the morphism induced by $\pi$. Let $s$ be a general point of $ {\mathbb {C}}^{m}_{o}$. Given $\alpha \in \operatorname {Ver}(F_s)$ we denote by $\sigma _{\alpha }$ the $(n+1)$-dimensional cone of $\Gamma ^{\star }(F_s)$ generated by all the non-negative normal vectors to faces of $\Gamma _{+}(F_s)$ which contain $\alpha$. Denote by $\widetilde {W}^{\rm t}$ the total transform of $W$ under $\varphi$.
Proposition 3.1 Let $s$ be a general point of $ {\mathbb {C}}^m_o$, and assume that
There exists an admissible subdivision, $\Sigma$, of $\Gamma ^{\star }(F_s)$ having the following properties.
(i) For each $\alpha \in \operatorname {Ver}(F_s,f)$, the fan $\Sigma$ defines a subdivision, $\{\sigma _{\alpha }^1,\ldots,\sigma _{\alpha }^r\}$, regular to $\sigma _{\alpha }$.
(ii) For each $j\in \{1,\ldots,r\}$, $\widetilde {W}^t\cap (X_{\sigma _{\alpha }^j}\times {\mathbb {C}}^m_o)$ is a normal crossings divisor relative to $ {\mathbb {C}}^m_o$.
Proof. Let us recall that $\mathcal {E}:=\{e_1,e_2,\ldots,e_{n+1}\}\subset {\mathbb {Z}}_{\geq 0}^{n+1}$ is the standard basis of $ {\mathbb {R}}^{n+1}$. First we construct a simplicial subdivision of $\Gamma ^{\star }(F_s)$. Let $\Gamma ^{\star }(F_s)(j)$ be the set of all the $j$-dimensional cones of $\Gamma ^{\star }(F_s)$. Let us consider a compatible simplicial subdivision, $\Sigma S$, of $\bigcup _{j=1}^{n}\Gamma ^{\star }(F_s)(j)$, such that if $\sigma '$ is a simplicial $j$-dimensional cone of $\Gamma ^{\star }(F_s)(j)$, $1\leq j\leq n$, then $\sigma '\in \Sigma S$ and $\Sigma S(1)=\Gamma ^{\star }(F_s)(1)$, where $\Sigma S(1)$ is the set of all the one-dimensional cones of $\Sigma S$.
Let us consider the case
By Theorem 2.25, $\alpha$ has a good apex. Then there exists $I\subsetneqq \{1,\ldots,n+1\}$ such that $\alpha \in {\mathbb {R}}^I_{>0}$ and $i\in I^c$ such that there exists a single edge $E_{\alpha }\ni \alpha$, of $\Gamma (F_s)$ not contained in $ {\mathbb {R}}_{\{i\}}$. Let $\beta =(\beta _1,\ldots,\beta _{n+1})\in \operatorname {Ver}(F_s)\cap E_{\alpha }$ be the good apex, which is to say $\beta _i=\delta _{ij}$, $j\in I^c$.
Observe that $e_{i}\in \mathcal {E}$, is an extremal vector of $\sigma _{\alpha }$. Let us consider the following simplicial subdivision of $\sigma _{\alpha }$:
where cone $\operatorname {Cone}(\{\cdot \})$ is the cone generated by $\{\cdot \}$. Now let us consider the case
let $\Sigma ^s(\sigma _{\alpha })$ be an arbitrary simplicial subdivision of $\sigma _{\alpha }$ that is compatible with $\Sigma S$. Then
is a simplicial subdivision of $\Gamma ^{\star }(F_s)$. As $F_s$ is convenient, the faces of $\sigma _{\alpha }$, $\alpha \in \operatorname {Ver}(F_s)$, contained in a coordinate plane are simplicial cones, then $\Sigma ^s$ is an admissible subdivision.
Now we define a subdivision of $\Sigma ^s$ to obtain the sought after fan.
Let $\alpha \in \operatorname {Ver}(F_s,f)$. By abuse of notation we denote for $\sigma _{\alpha }$ a cone in $\Sigma ^s(\sigma _{\alpha })(n+1)$. Without loss of generality we can suppose $i=n+1$, in this manner we have that $\sigma _{\alpha }=\operatorname {Cone}(e_{n+1},\tau )$ with $\tau \in \Sigma S$. We denote $H_{0}= {\mathbb {R}}_{\{n+1\}}\cap \Gamma _{+}(F_s)$ and $H_1$,…,$H_n$ the $n$-dimensional faces of $\Gamma _{+}(F_s)$ that define $\sigma _{\alpha }$, then $\bigcap _{j=0}^{n}H_j=\{\alpha \}$. Then $E_{\alpha }:=\bigcap _{j=1}^{n}H_j$
Let $p_1,\ldots,p_n$ be non-negative normal vectors to the faces $H_1,\ldots,H_n$. Then
Now we construct a regular subdivision of $\sigma _{\alpha }$. Let us consider the cone
and a regular subdivision $\operatorname {RS}(\tau )$ of $\tau$ that does not subdivide regular faces of $\tau$. Then $\operatorname {RS}(\tau )$ does not subdivide faces . Let $\tau '\in \operatorname {RS}(\tau )$, then there exists $q_1,\ldots,q_n\in \operatorname {Cone}(p_1,\ldots,p_n)$ such that $\tau ':=\operatorname {Cone}(q_1,\ldots,q_n)$. Observe that the cones
define a subdivision of the cone $\sigma _{\alpha }$ that can be extended to a subdivision $\Sigma$ of $\Sigma ^s$ that does not subdivide faces , which implies that $\Sigma$ is admissible.
Now we prove that $\sigma '_{\alpha }:=\operatorname {Cone}(q_1,\ldots,q_n,e_{n+1})$ is regular. Looking at $q_j$ as column vectors, and consider the matrix of the size $(n+1)\times n$:
For each $j\in \{1,\ldots,n+1\}$ let $A_j$ be the matrix of the size $n\times n$ obtained by deleting the row $j$ of the matrix $A$. As $\tau ':=\operatorname {Cone}(q_1,\ldots,q_n)$ is regular, we have that the greatest common divisor, $\gcd (d_1,\ldots,d_{n+1})$, where
is equal to $1$. Let us suppose that the cone
is not regular, then $|\det (q_1,\ldots,q_n,e_{n+1})|=d_{n+1}\geq 2$. For each $H_j$, $1\leq j \leq n$ we have that $\alpha,\beta \in H_j$, then $\langle \alpha, p_j \rangle =\langle \beta, p_j\rangle$ for all $1\leq j \leq n$, which implies that $\langle \alpha, q_j \rangle =\langle \beta, q_j\rangle$ for all $1\leq j \leq n$. From this we obtain that
for all $1\leq i \leq n$ (remember that $\beta$ is the good apex of $\alpha$). Then $d_{n+1}$ divides to $d_j$ for all $1\leq j\leq n$, which contradicts the fact that $\gcd (d_1,\ldots,d_{n+1})=1$. This implies that $\sigma '_{\alpha }$ is regular.
Observe that there exist coordinates $y_1,\ldots,y_{n+1}$ of $X_{\sigma '_{\alpha }}\cong {\mathbb {C}}^{n+1}$ (before the base change) such that the morphism
is defined by
From this we obtain
Let us assume that $r=(r_1,\ldots,r_{n+1})$ is a singular point of $\overline {F}(y,o)$. Then there exists $1\leq j \leq n$ such that $r_j=0$. Without loss of generality, we can suppose that $r_n=0$. We know that for each $\beta '\in E_{\alpha }\cap \operatorname {Ver}(f)$ we have that $\langle \alpha, q_i \rangle =\langle \beta ', q_i\rangle$, for all $1\leq i \leq n$, and as $\alpha$ has a good apex, we obtain
where $\overline {y}=(y_1,\ldots,y_{n-1})$, $c_0(o)=0$, and
If $E_{\alpha }\cap \operatorname {Ver}(f)=\{\beta \}$, then $\overline {K}(y_{n+1},s)=c_1(s)y_{n+1}$. This shows that $r$ cannot be a singular point of $\overline {F}$. If $|E_{\alpha }\cap \operatorname {Ver}(f)|>1$, then the singular point $r=(r_1,\ldots,r_{n+1})$ satisfies
This implies that $r_{n+1}\neq 0$. We prove that this is contradiction.
Let $W=\operatorname {Ver}(F_s,f)\cap {\mathbb {R}}_{n+1}$ and we define
We may assume that $F'$ is a non-degenerate deformation of $f$. As
we have $\nu (F'_s)=\nu (f)$ (see Corollary 2.5). By the definition of $F'$, the point $\alpha$ belongs to $\operatorname {Ver}(F'_s,f)=W$. We note $\sigma _{\alpha }$ the cone of $\Gamma ^{\star }(F'_s)$ associated with $\alpha$. By construction, the cone $\sigma _{\alpha }$ of $\Gamma ^{\star }(F'_s)$ is the cone $\sigma _{\alpha }$ of $\Gamma ^{\star }(F_s)$ defined previously. Using the same regular subdivision of $\sigma _{\alpha }$ we can define a regular admissible subdivision $\Sigma '$ of the fan $\Gamma ^{\star }(F'_s)$.
Let $\sigma '_{\alpha }$ be one of the two regular cones of the subdivision of $\sigma _{\alpha }$ (see $(\star )$). As we previously obtained
Then $r$ is a singular point of $\overline {F}(y,o)$ if and only if $r$ is a singular point of $\overline {F'}(y,o)$ (in fact, $\overline {F}(y,o)=\overline {F'}(y,o)$). We recall that $|E_{\alpha }\cap \operatorname {Ver}(f)|>1$, and that $E_{\alpha }$ is the only edge of $\Gamma _{+}(F_s)$ not contained in $ {\mathbb {R}}_{i}$, which contains $\alpha$ and its good apex. Observe that $E_{\alpha }$ also is the unique edge $\Gamma _{+}(F'_s)$ which satisfies the previous properties. Let $\beta '\neq \alpha$ an end point of $ {\operatorname {E}}_{\alpha }$, and $\sigma _{\beta '}\in \Gamma ^{\star }(F_s)$ the cone associated with $\beta '$. As $|E_{\alpha }\cap \operatorname {Ver}(f)|>1$, and $\operatorname {Ver}(F'_s,f)\subset {\mathbb {R}}_{n+1}$, we have that the cone $\sigma _{\beta '}$ belongs to $\Gamma ^{\star }(f)$. Then the regular subdivision $\sigma ^1_{\beta '}$,…,$\sigma ^t_{\beta '}$ of $\sigma _{\beta '}$ defined by the regular admissible subdivision $\Sigma '$ can be extended to regular admissible subdivision $\Sigma ''$ of $\Gamma _{+}(f)$. By construction, there exists $1\leq j\leq t$ such that $r\in X_{\sigma ^{j}_{\beta '}}\cong {\mathbb {C}}^{n+1}$. But $f$ is non-degenerate, which implies that $\widetilde {V}^{\rm s}\cap X_{\sigma ^{i}_{\beta '}}$ is smooth, from where we obtain the sought after contradiction. This implies that $F(\pi _{\sigma '}(y),s)$, which is a normal crossings divisor relative to $ {\mathbb {C}}^m_o$ around $\pi ^{-1}_{\sigma '_{\alpha }}(o)\times {\mathbb {C}}^m_o$.
The following theorem is the main result of this article. Let $s$ be a general point of $ {\mathbb {C}}^m_{o}$. We construct a regular admissible subdivision, $\Sigma$, of $\Gamma ^{\star }(F_s)$ in the manner that $\rho :X_{\Sigma }\times ( {\mathbb {C}}^m,o)\rightarrow {\mathbb {C}}^{n+1}_o\times {\mathbb {C}}^m_o$ is the sought after simultaneous embedded resolution. Observe that for the result commented upon previously, $\pi :X_{\Sigma }\rightarrow {\mathbb {C}}^{n+1}_o$ defines an embedded resolution of $W_s$.
Theorem 3.2 Assume that $W$ is a Newton non-degenerate deformation. The deformation $W$ is $\mu$-constant if and only if $W$ admits a simultaneous embedded resolution.
Proof. The ‘if’ part is given by Proposition 1.6. We prove ‘only if’.
By Proposition 3.1 there exists an admissible subdivision, $\Sigma$, of $\Gamma ^{\star }(F_s)$ (where $s$ is a general point of $ {\mathbb {C}}^{m}_{o}$) such that for each $\alpha \in \operatorname {Ver}(F_s,f)$, the fan $\Sigma$ defines a subdivision $\sigma _{\alpha }^1,\ldots,\sigma _{\alpha }^r$, regular of $\sigma _{\alpha }$, such that $\widetilde {W}^t\cap X_{\sigma _{\alpha }^i}\times {\mathbb {C}}^m_o$ is a normal crossings divisor relative to $ {\mathbb {C}}^m_o$ for $i\in \{1,\ldots,r\}$. Consider the set, $\Sigma (j)$, of all the cones of dimension $j$ of $\Sigma$. Observe that given a regular admissible subdivision of $\Sigma (j)$, there exists a regular admissible subdivision of $\Sigma (j+1)$ compatible with the given subdivision. Using recurrence we have that there exists a regular admissible subdivision of $\Sigma$ that does not subdivide its regular cones. By abuse of notation we denote for $\Sigma$ the regular admissible subdivision. To finish the proof we still need to consider $\alpha \in \operatorname {Ver}(F_s)\cap \operatorname {Ver}(f)$. Let us consider the cone $\sigma \subset {\mathbb {R}}^{n+1}_{\geq 0}$ generated by all the non-negative normal vectors to faces of $\Gamma _{+}(F_s)$ which contain a $\alpha$, and let $\sigma ^1,\ldots,\sigma ^r$ be the regular subdivision defined by $\Sigma$. Let us suppose that $p_{1}^i,\ldots,p_{n+1}^i$ are the extremal vectors of $\sigma ^i$. As $\sigma ^i$ is regular, we have that $X_{\sigma ^i}\cong {\mathbb {C}}^{n+1}$ (before the base change). Then we can associate the coordinates $y_1,\ldots,y_{n+1}$ to $X_{\sigma ^i}$ such that $\pi _{\sigma ^i}:=\pi |_{X_{\sigma ^i}}$ is defined by
where $x_j:=y_1^{p_{1j}^i}\cdots y_{n+1}^{p_{n+1\;j}^i}$, $p_{j}^i:=(p^i_{j 1},\ldots,p^i_{j \;n+1})$, $1\leq j\leq n+1$. Then
Let $c(s)$ be the degree zero term of $\overline {F}(y,s)$. As $\alpha \not \in \operatorname {Ver}(F_s,f)$, there exits a sufficiently small open set $0\in \Omega \subset {\mathbb {C}}^{m}$ such that $c(s')\neq 0$ for all $s'\in \Omega$. Moreover, for each $s'\in \Omega$, we have $\sigma ^i\subset \sigma _{\alpha,s'}$, $1\leq i\leq r$, where $\sigma _{\alpha, s'}$ is the $(n+1)$-dimensional cone of $\Gamma ^{\star }(F_{s'})$ generated by all the non-negative normal vectors to faces of $\Gamma _{+}(F_{s'})$ which contain $\alpha$. Observe that for each $s'\in \Omega$ we can extend the fan formed by the cones $\sigma ^i$, $1\leq i\leq r$, to a subdivision of the cone $\sigma _{\alpha,s'} \in \Gamma ^{\star }(F_{s'})$ without subdivisions of strict faces of $\Delta$ (this allows us to return to the classical case of non-degenerate hypersurface for each $s'\in \Omega$). Then the property of non-degeneracy of $F$ implies that $F_{s'}(\pi _{\sigma ^i}(y))$ is a normal crossings divisor for each $s'\in \Omega$. This implies that $F(\pi _{\sigma ^i}(y),s)$ is a normal crossings divisor relative to $ {\mathbb {C}}^m_o$ around $\pi ^{-1}_{\sigma ^i}(o)\times {\mathbb {C}}^m_o$.
4. The degenerate case
Let us recall that $V$ is a hypersurface of $C_{o}^{n+1}$, $n\geq 1$, given by $f\in {\mathcal {O}_{n+1}^{x}}$ irreducible such that $V$ has an isolated singularity at $o$. Let $F\in {\mathbb {C}}\{x_1,\ldots,x_{n+1},s_1,\ldots,s_m\}$ be a deformation of $f\in {\mathbb {C}}\{x_1,\ldots,x_{n+1}\}$:
where $h_i\in {\mathcal {O}_{m}^{s}}:= {\mathbb {C}\{s_{1},\ldots,s_{m}\}}$, $m\geq 1$, and $g_i\in {\mathcal {O}_{n+1}^{x}}$ such that $h_i(o)=g_i(o)=0$. Consider the relative Jacobian ideal
The following theorem gives a valuative criterion for the $\mu$-constancy of a deformation.
Theorem 4.1 [Reference GreuelGre86, Reference Lê Dũng and SaitoLDS73, Reference TeissierTei73]
The following are equivalent:
(i) $F$ is a $\mu$-constant deformation of $f$;
(ii) for all $i\in {1,\ldots,m}$ we have that $\partial _{s_{i}}F\in \overline {J_x(F)}$, where $\overline {J_x(F)}$ denotes the integral closure of the ideal $J_x(F)$;
(iii) for all analytic curve $\gamma :( {\mathbb {C}},o)\rightarrow ( {\mathbb {C}}^{n+1}\times {\mathbb {C}}^{m},o)$, $\gamma (o)=o$, and for all $i\in \{1,\ldots,m\}$ we have that $ {\operatorname {Ord}}_t(\partial _{s_i}F\circ \gamma (t))> \min \{ {\operatorname {Ord}}_t(\partial _{x_j}F\circ \gamma (t))\,|\, 1\leq j\leq n+1\}$.
(iv) same statement as in statement (iii) with ‘$>$’ replaced by ‘$\geq$’.
Next, we state and prove an analogue of Corollary 2.19 for deformations that do not satisfy the non-degeneracy assumption. Let $s$ be a general point of $ {\mathbb {C}}_{o}^m$, $I$ a proper subset of $\{1,\dots n+1\}$ and consider
such that $\operatorname {Ver}(F_s,f)\cap {\mathbb {R}}^I_{>0}\neq \emptyset$. If $F$ is a $\mu$-constant non-degenerate deformation of $f$, then by virtue of Corollary 2.19 there exists $i\in I^c$ such that $\beta _i=\delta _{ij}$ for $j\in I^c$, which is analogous to statement (ii) of the following proposition.
In the rest of the section, we use the following notation. Given that $J\subsetneqq \{1,\ldots,n+1\}$, we denote by $ {\mathbb {C}}_o^J$ the complex-analytic germs at the origin of $ {\mathbb {C}}^J:=\{(x_1,\ldots,x_{n+1})\in {\mathbb {C}}^{n+1}:x_i=0\;\text {if}\;i\notin J\}$ and by $f_J$ (respectively, $F_J$) the natural restriction of $f$ (respectively, $F$) to $ {\mathbb {C}}^J_o$ (respectively, $ {\mathbb {C}}_o^J\times {\mathbb {C}}_{o}^{m}$). Let $V_J$ be the subset of $ {\mathbb {C}}_{o}^J$ defined by the equation $f_J(x)=0$.
Let $\operatorname {Supp}(F,f):=\operatorname {Supp}(F){\setminus} \operatorname {Supp}(f)$.
Proposition 4.2 Fix a set $I\subsetneqq \{1,\ldots,n+1\}$. Let us assume that $F$ is a $\mu$-constant deformation of $f$, and that $\operatorname {Supp}(F_s,f)\cap {\mathbb {R}}^{I}_{>0}\neq \emptyset$.
Then given
at least one of the following conditions is satisfied.
(i) The restriction $f_J$ is reduced, $V_J$ is a hypersurface of $ {\mathbb {C}}_{0}^J$ with an isolated singularity at $o$, and $F_J$ is a $\mu$-constant deformation of $f_J$.
(ii) There exists $i\in J^c$ and $\beta :=(\beta _1,\ldots,\beta _{n+1})\in \operatorname {Supp}(F)$ such that $\beta _i=\delta _{ij}$, for $j\in J^c$.
A difference between the degenerate and the non-degenerate cases is that the point $\beta \in \operatorname {Supp}(F_s)$ of the previous proposition need not, in general, belong to the set $\operatorname {Supp}(f)$.
Example 4.3 Consider the following deformation
In [Reference AltmannAlt87] it was shown that $F$ is a $\mu$-constant degenerate deformation of the non-degenerate polynomial $f(x_1,x_2,x_3):=x_1^5+x_2^6+x_3^5+x_2^3x_3^2$ . In this example, we have that $\operatorname {Ver}(F_s,f):=\{(4,1,0)\}\subset {\mathbb {R}}_{>0}^{\{1,2\}}$ and $\beta :=(2,2,1)$. Observe that $\beta \notin \operatorname {Supp}(f)$.
Proof of Proposition 4.2 There is no loss of generality in supposing that $J= \{1,\ldots,k\}$, $k\leq n$. We can always write $F$ in the following manner:
where $s=(s_1,\ldots,s_m)$. Observe that $F_J=G$, and let $g:=f_J=G|_{s=0}$. Let us suppose that condition (ii) is not satisfied, then $G_{i}(x_1,\ldots,x_k,s)\equiv 0$ for all $k< i\leq n+1$, then
Thus, we obtain that:
(i) $\partial _{l}F=\partial _{l}G+ \sum _{k< i\leq j}x_ix_j\partial _lG_{ij},$ for $1\leq l\leq k$;
(ii) $\partial _{l}F=\sum _{k< i\leq l}x_iG_{il}+\sum _{l\leq j}x_jG_{lj}+\sum _{k< i\leq j}x_ix_j\partial _lG_{ij},$ for $k< l$;
(iii) $\partial _{s_{j'}}F=\partial _{s_{j'}}G+ \sum _{k< i\leq j}x_ix_j\partial _{s_{j'}}G_{ij},$ for $1\leq j'\leq m$.
Let us suppose that the singularity of $g(x)=G(x_1,\ldots,x_{k},0)$ is not isolated in the origin $o$, or $g\equiv 0$, or $g$ not reduced. Then for each open set $o\in \Omega \subset {\mathbb {C}}^{k}$ there exists $(p_1,\ldots,p_{k})\in \Omega$ such that:
(a) $g(p_1,\ldots,p_{k})=0,$
(b) $\partial _{l}g(p_1,\ldots,p_{k})=0$, for $1\leq l\leq k$.
Then $(p_1,\ldots,p_{k},0,\ldots,0)\in {\mathbb {C}}^{n+1}$ is a singularity of $f$, which is a contradiction.
Let us suppose that $G(x_1,\ldots,x_{k},s)$ is not a $\mu$-constant deformation of $g$. Then by virtue of Theorem 4.1 there exists $1\leq j\leq m$, and an analytic curve
such that
Let us consider the following analytic curve:
Using (i), (ii), and statement (iii), we observe that we can choose the large enough $r_{k+1},\ldots r_{n+1}$, and the $a_{k+1}(t),\ldots a_{n+1}(t)$, which are general enough in the manner that:
(i) $ {\operatorname {Ord}}_t \partial _{s_j}F\circ \beta (t)= {\operatorname {Ord}}_t\partial _{s_j}G\circ \gamma (t)$ for $1\leq j\leq m$;
(ii) $ {\operatorname {Ord}}_t\partial _{i}F\circ \beta (t)= {\operatorname {Ord}}_t\partial _{i}G\circ \gamma (t)$ for $1\leq i\leq k$;
(iii) $ {\operatorname {Ord}}_t \partial _{l}F\circ \beta (t)\geq \max _{1\leq i\leq k}\{ {\operatorname {Ord}}_t \partial _{i}F\circ \beta (t)\}$ for $k< l$.
This implies that
This contradicts Theorem 4.1 since $F$ defines a $\mu$-constant deformation. Then $G(x_1,\ldots,x_{k},s)$ is a $\mu$-constant deformation of $g$ or there exists at least one non-zero $G_i$.
Acknowledgements
The authors would like to thank the anonymous referees for their valuable comments which helped to improve the manuscript. The first author would also to thank Rachel Rogers for her unconditional support during this time.