Higher dimensional essential minima and equidistribution of cycles (Q2675326)

In Arakelov theory, one of the problems is to evaluate the \textit{naive Betti numbers}, namely the logarithm of the number of small sections of the line bundle \(L\). This is suppose to mimic the classical Betti number \(h^{0}(L)\), which is the dimension of global sections of \(L\). In the one dimensional case this has been successfully studied by Bost in his paper on theta invariants. However, the analogous study for higher dimensional arithmetic varieties has been difficult. For the two dimensional case the issue is already extremely difficult, though progress has been made by Shou-wu Zhang and Ullmo showing an appropriate lower bound exists. The main methods hinges upon the introduction of the notion of admissible pairing and equidistribution of algebraic points [\textit{E. Ullmo}, Ann. Math. (2) 147, No. 1, 167--179 (1998; Zbl 0934.14013)]. The methods are not effective for attaining effective estimates for upper bounds, however. Faltings suggested a strategy in 2021 (see [\textit{G. Faltings}, Arakelov estimates. Max Planck Institute of Mathematics (2021)]) to circumvent this difficulty by analyzing possible ramification points of the Weil parametrization \(X_{0}(N)\rightarrow E\) of elliptic curves over \(\mathbb{Q}\). But even if this ambitious strategy is successful, most likely we still do not have an effective estimate for the upper bounds. As of 2023 as far as I know no essential new progress has been made on this thorny issue. This does not mean our study of the subject has stopped, but it is a strong indication of deeper understanding is needed to make progress beyond piece-wise technical advances. The study of the Betti numbers thus evolves around various ideas, each of which has been developed but encountered various difficulties: \begin{itemize} \item[(1)] The theory by Arakelov-Faltings still treated \(p\)-adic (finite) places and complex (infinite) places using different language. It is now possible to introduce the language of Berkovich spaces and (more generally) tools related to \(p\)-adic Hodge theory. The main issue is despite extensive development of the adelic theory, as far as I know we still have difficulty to reconcile the methods we used for two places. For example, an extension of the Quillen metric at the infinite place would entail a similar development of \(p\)-adic differential operators. Such a theory has been developed but I am not sure if it is suitable for Arakelov theory purposes. \item[(2)] A second approach try to by-pass the above issues by working with a \(\theta\)-\textit{regularized} version of naive \(h^{0}\). Similar to the one dimensional case discussed by Bost/Borisov, this approach aims to establish an exact arithmetic Riemann-Roch formula for algebraic surfaces. The contribution of finite places is now included in the summands of an appropriate infinite dimensional normal distribution. The main difficulty, however is that such a theory is hard to evaluate in practice and numerical approximation is needed. Therefore it is unclear how it helps to establish \(h_0\)'s upper bounds. While some progress has been made, at the moment we do not have a satisfactory cohomological theory based on \(\theta\)-regularized invariants. Some philosophical questions also arise from such an approach. Bjorn Poonen pointed out that if we successfully extending Tates' thesis to arithmetic surfaces. But we know this is hard. \item[(3)] A third approach, which is arguably less ambitious is to abandon the framework of cohomology theory together. Rather it focuses on establishing certain numerical relationship of the Betti numbers. This approach has the advantage of generating explicit inequalities which can have number theory applications. However, work based on this approach quickly encountered group-theory rhetoric difficulties even in 1D case. For 2D cases the approach requires at the very least, estimates of analytic torsion. While some progress has been made, in general we still do not have a good understanding of secondary invariants near the boundary of \(M_{g}\). As a result the inequalities extracted out from this theory also cannot help gaining upper bounds for \(h_0\). \item[(4)] A fourth approach is to ignore the complexities of two dimensional case altogether. Rather people try to work on the situation of general \textit{arithmetic varieties} and apply the results to 2D cases. We note this has been the approach employed by most people since 1990s. The study of Arakelov theory thus shifted remarkably to the study of \textit{heights} of general arithmetic varieties. In particular the results are often no longer quantitative but only holds true after taking limits when \(n\rightarrow \infty\). The hope is then that some of the statements can be refined to be effective when the general theory is well developed. While this approximate issue is a main drawback, such an approach benefits from the fact that the naive Betti number is well defined for all (nice) arithmetic varieties, for example the ones with semi-positive metric. In particular for `very big' line bundles we can still carry out this scheme and get some arithmetic information. The main difficulty is how to reconcile the contribution from infinity, sometimes manifested in estimate the volume of the unit ball in \(H^{0}(X,L)\). \end{itemize} In the present article the authors try to offer something that can partially compensate the issue at infinity. The main technical result the authors prove is the following: Theorem. Let \(|\cdot |_{v}\) be a semipositive quasi-algebraic metric on \(L\), and \(\mathcal{O}_{X}=(\mathcal{O}_{X}, { (|\cdot |'_{v}))_{v} }\) be ``difference of semi-positive'' quasi-algebraically metrized. For \(t\in \mathbb R\), with \(t\) close to \(0\), and for \(n>0\) large enough, there is a non-zero section \(s\in H^{0}(X_{\overline{K}}, L^{\otimes n}_{\overline{K}})\) satisfying \[ \sum_{v\in \mathcal{M}} n_{v}\sup \log |s|_{v}^{\otimes n} |1|_{v}'^{\otimes t}\le n(-\hat{h}_{L\otimes \overline{O}_{X}^{\otimes t}})(X)+O(t^2) \] We note the proof of the theorem (the authors call this key inequality) is not clear. Roughly speaking, the authors manage to prove a related statement on big line bundles (Proposition 4.2, Proposition 3.2 in the arxiv version). Lemma. Let \(\mathcal{N}\) be a line bundle on \(\mathcal{H}\) that is trivial on the generic fiber, and assume \(e=1\). For every \(t\in \mathbb{Q}\), with \(t\) close to \(0\), and every \(r\in \mathbb{Q}\) such that \[ r> \frac{-h_{{\mathcal{L}\otimes \mathcal{N}}^{\otimes t}}(X)}{(N+1)\deg_{M}(C)\deg_{L}(X)}+O(t^2) \] with the implicit constant in \(O(t^2)\) not depending on \(n\), we have \[ h^{0}(\mathcal{H}, (\mathcal{L}\otimes \pi^{*}\mathcal{M}^{\otimes r} \otimes N^{\otimes t})^{\otimes n})\rightarrow 0 \] The proof of this lemma quite nice, but its relationship with the `key inequality' above is a bit unclear. In particular, I find that the proof of the result for function fields (Corollary~4.3) is not written clearly. The authors somehow discuss issues not relevant (for example \(L\) versus \(L^{e}\), note \(L^e\) did not even show up in the statement of Corollary~4.3) and an important step involving \(\deg_{L}(X)\) is omitted. Thus, I am not entirely sure that the proof of Corollary~4.3 is correct. Assuming that Corollary 4.3 is indeed correct and similarly Theorem 4.1 is also correct, I am not entirely sure why this would lead to an analog of Zhang's inequality. The authors claim the following: Theorem. Let \(X\) be a projective variety over \(K\), and \(\overline{L}\) a big and semiample line bundle on \(X\) equipped with a semipositive quasi-algebraic metric. Then, for every integer \(d=0,\cdots, \dim(X)\), we have \[ e^{d}_{1}(X,\overline{L})\ge \widehat{h}_{\overline{L}}(X) \] The proof given by the authors is somewhat unclear to me. Note the proof has to use Proposition~3.11 and Theorem~4.1 together, but the details are essentially left to the reader. Thus many subtleties are omitted, like how the \(\deg_{L}(Y)\) term disappears in the final evaluation. At the moment I am not entirely following the proof. The issue remaining is whether the results (if correct) are useful for arithmetic purposes. The answer is \textit{Yes}. Note that to establish estimates on \(h_0\) even in the 2D case, it suffices to have estimates on the height. Thus the authors' version of Zhang's theorem (if true) does help us to understand the relationship better. However it is completely unclear to me how to sharpen Theorem 0.3 to get effective estimates.