The distribution relation and inverse function theorem in arithmetic geometry (Q2031275): Difference between revisions

The paper presents an arithmetic distribution relation as well as two versions of the inverse function theorem in terms of an arithmetic distance function. In sections 3-5, the field \(K\) denotes a field with a complete set of absolute values \(\mathcal{M}_K\) satisfying a product formula. If we fix an algebraic closure \(\bar{K}\) of \(K\), an \(\mathcal{M}_K\)-constant will be a function \(\gamma \colon \mathcal{M}_{\bar{K}} \longrightarrow \mathbb{R}_{\geq 0}\) such that \(\gamma(v)\) depends only on the restriction \(v|_K\) and the set \(\{ v|_K \, | \, \gamma(v) \neq 0 \}\) is finite. The notation \(O(\mathcal{M}_K)\) will be used to denote a relation that holds up to an \(\mathcal{M}_K\)-constant. For instance \(f \leq g +O(h) + O(\mathcal{M}_K)\) means that there exist a \(C>0\) and an \(\mathcal{M}_K\)-constant \(\gamma\) such that \(f \leq g + C|h| + \gamma\). Definition: (Local height functions) Let \(K\) be a field with a set of absolute values \(\mathcal{M}_K\). Let \(V\) be a projective variety (not necessarily irreducible) over \(K\) and \(X \subset V\) a closed subscheme. A local height is a function \(\lambda_X \colon V(\bar{K}) \times \mathcal{M}_K \longrightarrow \mathbb{R} \cup \infty\) determined by the following properties: \begin{itemize} \item[(1)] If \(D\) is an effective divisor, we get the usual local height, i.e., \(\lambda_X = \lambda_D\). \item[(2)] if \(X,X'\) are subschemes \(\lambda_{X \cap X'} = \min{(\lambda_X,\lambda_{X'})}\) (where \(X \cap X'\) denotes the subscheme with ideal sheaf .\(\mathcal{I}_X+\mathcal{I}_{X'}\) ). \end{itemize} and having also many other nice properties: \begin{itemize} \item[(3)] (functoriality) If \(\varphi \colon V \longrightarrow W\) denotes a morphism of varieties and \(X\) is a closed subscheme of \(W\), we have the equality: \[\lambda_{\varphi^{-1} W,V} = \lambda_{X,W}.\] \item[(4)] Local height functions are bounded below, so up a \(\mathcal{M}_K\)-constant, we can assume that \(\lambda_X \geq 0\). \end{itemize} Remark: Using the local height function associated to the boundary divisor, the local height function machinery can be extended to quasi-projective varieties. Definition: Let \(\Delta(V)\) denotes the diagonal subvariety in \(V \times V\). The arithmetic distance function on \(V\) is the local height \[\delta_V = \lambda_{\Delta(V)}.\] It is well-defined up to an \(\mathcal{M}_K\)-bounded function and satisfies many nice properties as well, for example: \begin{itemize} \item[(1)] \(\delta(P,R) \geq \min(\delta(P,Q),\delta(Q,R))\) \item[(2)] \(\lambda_X(Q) \geq \min(\lambda_X(P),\delta(P,Q))\) \end{itemize} Defintion: Let \(\varphi \colon W \longrightarrow V\) be a finite flat morphism between schemes of finite type over a field \(k\). Let \(k'\) be an algebraically closed field containing \(k\). For \(x \in W(k')\), define the multiplicity of \(\varphi\) at \(x\) by the formula \[e_\varphi(x)= \text{length}_{\mathcal{O}_{W_{k'},x}} \mathcal{O}_{W_{k'},x}/\varphi^{-1} m_{\varphi(x)} \mathcal{O}_{W_{k'},x}.\] For finite flat morphisms \(W \longrightarrow V\), the distribution inequality bounds the arithmetic distance in the target variety in terms of the arithmetic distance of the pre-images in \(W\). In good cases, it will give a distribution relation. Theorem: (Arithmetic distribution relation/inequality) Let \(\varphi \colon W \longrightarrow V\) be a generically étale finite flat morphism between quasi-projective geometrically integral varieties over \(K\). For all \((P,q,v) \in W(\bar{K}) \times V(\bar{K}) \times v\), we have \[\delta_V(\varphi(P),q,v) \leq \sum_{Q \in W(\bar{K}),\, \varphi(Q)=q} e_\varphi(Q) \delta_W(P,Q,v)+ O(\lambda_{ \partial(V \times W)}(P,q,v))+ O(\mathcal{M}_K).\] We say that we have an arithmetic distribution relation when the above inequality becomes an equality. For example, assuming that \(V,W\) are both smooth, we have an arithmetic relation in the following two situations: \begin{itemize} \item[(1)] The map \(\varphi \colon W \longrightarrow V\) is étale (where there is not ramification). \item[(2)] The dimensions \(\dim(V)=\dim(W)=1\) (where the ramification divisor is at most zero-dimensional). \end{itemize} Remark: The above inequality does not always became an arithmetic relation. Even when we take a Galois cover \(\varphi \colon W \longrightarrow V\) with Galois group \(\text{Gal}(V/W)=\{\tau_1,\dots,\tau_n\}\), we have an inclusion of associated sheaves of ideals \[\mathcal{I}(\sum_{i=1}^n (1\times \tau_i)^* \Delta(W)) \subset \mathcal{I}((\varphi \times \varphi)^*(\Delta(V))\] that does not always gives an equality of closed schemes. Take for example \(\varphi \colon \mathbb{P}^1 \times \mathbb{P}^1 \longrightarrow \mathbb{P}^1 \times \mathbb{P}^1 \) defined by \(\varphi([x,y],[z,w])=([x^2,y^2],[z,w])\). Remark: The above inequality can be used to obtain a quantitative inverse theorem, namely, given a finite map, how far apart from the ramification locus and the boundary we need to be, to be able to define a local inverse. Also, the inverse obtained can be shown to behave nicely with respect to the distance functions. In some sense, the distance between points is close to the distance between the inverses. Theorem: (Inverse function theorem version \(1\)) Suppose that \(V\) and \(W\) are quasi-projective geometrically integral varieties defined over \(K\). Assume that the map \(\varphi \colon W \longrightarrow V\) is a generically étale finite flat surjective morphism of degree \(d\) also defined over K. Let us denote by \(\text{Ann}(\Omega_{W/V})\), the annihilator ideal sheaf of \(\Omega_{W/V}\) and by \(A(\varphi) \subset W\) the closed subscheme defined by \(\text{Ann}(\Omega_{W/V})\). \begin{itemize} \item[(a)] There exist constants \(C_2,C_3\) and \(\mathcal{M}_K\)-constants \(C_4,C_5\) such that the following holds:\\ If the triple \((P,q,v) \in W(K) \times V(K) \times M(K)\) satisfies \[\delta_V (\varphi(P),q;v) \geq d \lambda_{A(\varphi)}(P;v) + C_2 \lambda_{ \partial_{W \times V} }(P,q;v) + C_4(v)\] then there exists a point \(Q \in W(K)\) satisfying \(\varphi(Q) = q\) and \[\delta_W(P,Q;v) \geq \delta_V (\varphi(P),q;v) -(d-1)\lambda_{A(\varphi)}(P;v)- C_3 \lambda_{\partial(W \times V )}(P, q; v) - C_5(v).\] \item[(b)]If we take \(C_4\) to be an appropriate positive real number, instead of an \(\mathcal{M}_K\) constant, and if we also assume that \(P \notin A(\varphi)\), then the point \(Q\) in (a) is unique. \end{itemize} Remark: The arithmetic distribution relation and the inverse function theorem have been used to study integral points in the following situations: \begin{itemize} \item[(1)] To find uniform height estimates while working with the étale map \([n] \colon A \longrightarrow A\) on an Abelian scheme \(A \longrightarrow T\) over a base variety \(T\). \item[(2)] To find an analogous of Siegel's theorem while working with Iterates \(f^n\) of a rational map \(f \colon \mathbb{P}^1 \longrightarrow \mathbb{P}^1\) of degree at least two. \end{itemize} The first version (version \(1\)) of the inverse function theorem works simultaneously over several places. In the next version the authors present a stronger result working only over a complete field \(K\). By working over a complete field, the exponents or coefficients of the inverse function theorem are improved from \((d,d-1)\) to \((2,1)\). Theorem: (Inverse function theorem version \(2\)). Let \( (K, | . |)\) be a complete field. Let \(W, V\) be smooth quasi-projective varieties over \(K\), and let \(\varphi \colon W \longrightarrow V\) be a generically finite generically étale morphism. Let \(E \subset W\) be the closed subscheme defined by the \(0\)-th fitting ideal sheaf of \(\Omega_{W/V}\). Fix arithmetic distance functions \(\delta_W\), \(\delta_V\), a local height function \(\lambda_E\), and a boundary function \(\lambda_{\partial V}\). Let \(B \subset W(K)\) be a bounded subset. Then there are constants \(C_{36}, C_{37}, C_{38}, C_{39} > 0\) and a bounded subset \(\tilde{B} \subset W (K)\) containing \(B\) such that for all \(P \in B\) and \(q \in V(K)\) satisfying \[P \notin E \quad \text{and} \quad \delta_V(\varphi(P),q) \geq 2 \lambda_E(P) + C_{36} \lambda_{\partial V}(q) + C_{37},\] there is a unique \(Q \in \tilde{B}\) satisfying \[\varphi(Q)=q \quad \text{and} \quad \delta_W(P,Q) \geq \delta_V(\varphi(P),q) - \lambda_E(P)-C_{38} \lambda_{\partial V}(q) - C_{39}.\] Remark: Both versions of the inverse function theorem are suitable to prove results analogous to the continuity of the roots for polynomials. For example from the second version, we could get the following result. Let \( (K, | . |)\) be a complete field. Let \(D \in \mathbb{R}_{>0}\) and \(n \in \mathbb{Z}_{>0}\). Then there are positive constants \(C_{40},C_{41} > 0\) such that the following holds. Suppose that: \begin{itemize} \item \(f,g \in K[t]\) are monic polynomials of degree \(n\); \item The Gauss norms \(|f | \leq D\) and \(|g| \leq D\); \item There is an \(\alpha \in K\) such that \(f(\alpha) = 0\) and \(|f-g| \leq C_{40} |f''\alpha)|\). \end{itemize} Then there is \(\beta \in K\) such that \[g(\beta) = 0 \quad \text{and} \quad |\alpha - \beta||f'(\alpha)| \leq C_{41} |f - g|.\] Remark: The second version of the inverse function theorem is based on a higher dimensional version of Newton's method. The point \(Q\), will be obtained as limit of a Cauchy sequence of points \(Q_0=P, Q_1, Q_2 \dots\).