Multi-Parametric Geometry of Numbers

• Multi-parametric geometry of numbers is a framework that generalizes classical techniques by using several independent scaling parameters to study the behavior of successive minima in lattices.
• It employs piecewise-linear templates and variational principles to compute Hausdorff and packing dimensions in Diophantine approximation and fractal analysis.
• Its applications span weighted problems, transference inequalities, and extensions to number fields and higher-rank dynamics, advancing metric number theory.

The multi-parametric geometry of numbers is an advanced framework that generalizes classical one-parameter parametric geometry of numbers to settings involving several independent scaling parameters. Its main purpose is to analyze the joint behavior of successive minima of convex bodies attached to lattices under multi-parameter families of linear transformations, with deep implications for metric Diophantine approximation, fractal geometry, and transference principles. This formalism captures a vast array of multi-slope phenomena, especially in the study of matrix and weighted Diophantine approximation, and underpins key results in the computation of Hausdorff and packing dimensions of arithmetic exceptional sets.

1. From One-Parameter to Multi-Parameter Geometry

Classical parametric geometry of numbers (PGoN) as developed by Schmidt and Summerer is based on the study of the successive minima functions $L_i(q)$ associated to a convex body $C(q)$ depending on a single real parameter $q$ (e.g., scaling the body anisotropically) and a fixed lattice $\Lambda \subset \mathbb{R}^n$ (Roy, 2014). Each function $L_i(q) = \log \lambda_i(C(q), \Lambda)$ is continuous and piecewise linear with controlled slopes, and their combined graph encodes the Diophantine exponents of the relevant point. Roy’s rigid $n$-systems provide a combinatorial model for $L_i(q)$: any legitimate system of such piecewise-linear functions, satisfying monotonicity and volume constraints, is within $O(1)$ of an actual minima function for some lattice. This formalism connects the analytic structure of exponents with finite combinatorics.

The multi-parametric generalization introduces higher-dimensional parameter spaces (e.g., vectors $t = (t_1,\ldots,t_d)$ on a hyperplane), allowing simultaneous—and potentially anisotropic—scaling of multiple directions. Typical constructions include families of bodies

$$K(t) = \operatorname{diag}(e^{t_1}, ..., e^{t_d})B,$$

with constraints like $\sum t_i = 0$, and study the minima $L_i(t)$ as $t$ ranges over $\mathbb{R}^{d-1}$ (German, 2018, German, 2020). This approach enables the investigation of weighted, inhomogeneous, or matrix approximation problems, and the analysis of flows on spaces of higher rank lattices.

2. Templates, Systems, and Their Multi-Parameter Extensions

The extension from rigid systems to "templates" is essential in the multi-parameter context. A template for an $m \times n$ matrix is a $d$-tuple $f = (f_1,\dots,f_d):[t_0, \infty) \to \mathbb{R}^d$, $d = m + n$, where each $f_i$ is piecewise linear with slopes in $[-1/m, 1/n]$, the entries are ordered, and summations of the form $F_j(t) = \sum_{i=1}^j f_i(t)$ are convex, with quantized slopes in a finite set determined by the system’s combinatorics (Das et al., 2017, Das et al., 2019).

In the general multi-parameter setting, the key object is a piecewise linear function $F_{H, \ell}(t)$ (e.g., Harder–Narasimhan flag components), with the partition of parameter space into cells where each $F_{H, \ell}$ is linear. The collection of transition data between regions and the sequence of weights ("multisets") describe an abstract template (Solan, 2021). Such templates correspond (up to $O(1)$ error) to actual minima maps arising from lattices under multi-parameter flows:

$$g_{t} \Lambda = \operatorname{diag}(e^{t_1}, ..., e^{t_d}, e^{-(t_1+\cdots+t_d)})\Lambda .$$

These templates serve as universal objects for encoding the possible joint behaviors of all successive minima and underpin reduction of Diophantine and dimension-theoretic problems to problems about families of such piecewise-linear systems.

3. Variational Principle, Dimension Theory, and Schmidt Game Techniques

A fundamental advance is the variational principle, which asserts that the Hausdorff and packing dimensions of subsets of interest (e.g., singular matrices, exceptional sets in approximation) can be computed as suprema over average combinatorial invariants attached to templates (Das et al., 2017, Das et al., 2019). Specifically, for a family of templates $\mathcal{F}$ closed under finite additive perturbations, the Hausdorff (resp. packing) dimension of the set of points whose minima patterns shadow some template in $\mathcal{F}$ corresponds to

$$\sup_{f \in \mathcal{F}} \delta_H(f) \qquad \text{and} \qquad \sup_{f \in \mathcal{F}} \delta_P(f),$$

where $\delta_H(f), \delta_P(f)$ are suitable averages (lim inf / lim sup) of local combinatorial codimensions derived from the template’s partition into expanding and contracting blocks.

The proofs use Schmidt-type games adapted to the parametric/multi-parameter context: Alice and Bob alternately select nested balls or sets, encoding a dynamic strategic contest to realize or protect a given minima function, with the "score" of Alice corresponding to dimension lower bounds and Bob establishing upper bounds (Das et al., 2019).

This links homogeneous dynamics (via orbit divergence/boundedness under diagonal flows), parametric geometry (encoding Diophantine data in the flow), and metric geometry (dimensions of exceptional sets), encapsulating these connections in finite-dimensional optimization problems.

4. Multiplicative and Multi-Parameter Transference: Lattice Exponents

The multi-parametric geometry of numbers also generalizes the classical transference theorems for Diophantine exponents of lattices. Given a unimodular lattice $A \subset \mathbb{R}^n$, and the family $K(t)$ as above, local duality yields two-sided bounds for sums of minima, e.g.,

$L_k(t; A) + L_{n+1-k}(-t; A^*) = O(1),$

with Mahler’s principle providing the link between $A$ and its dual $A^*$. In this formalism, intermediate exponents (indexed by $k$) and their relationships generalize Khintchine and Dyson’s inequalities, yielding "splitting" chains of inequalities between intermediate exponents (German, 2018, German, 2020). The multiplicative or genuinely multi-parametric context tracks these exponents along higher-dimensional rays or subspaces, revealing richer phenomena and providing conceptual clarity to hierarchy and sharpness in classical bounds. This approach is especially powerful in the analysis of weighted approximation and in formulating and proving explicit splitting of transference inequalities for intermediate exponents.

5. Applications: Diophantine Approximation, Dimensions of Exceptional Sets, and Weighted Problems

A central result is the exact computation of the Hausdorff and packing dimensions of sets of matrices with prescribed Diophantine properties:

$$\operatorname{dim}_H \operatorname{Sing}(m, n) = \operatorname{dim}_P \operatorname{Sing}(m, n) = mn \bigg(1 - \frac{1}{m+n}\bigg),$$

answering longstanding conjectures for singular matrices (Das et al., 2017, Das et al., 2019). The framework extends to compute dimensions for sets associated to uniform exponents, gaps in successive minima (Schmidt’s conjectures), and points failing to meet certain quality-of-approximation criteria (e.g., Starkov’s sets), all via the variational optimization over families of templates.

Weighted and inhomogeneous Diophantine approximation is naturally encoded by allowing the parameter space and convex bodies to reflect arbitrary weighting and scaling—such as boxes or parallelepipeds with variable directions and sides. Construction of regular weighted graphs and templates provides extremal examples saturating transference/hierarchy inequalities and enables explicit computations in weighted transference settings (Schmidt et al., 2020, Roy, 1 Feb 2026).

The multi-parametric paradigm is also at the core of effective results in transcendental number theory (via Padé approximation), simultaneous rational approximation, and the determination of best possible irrationality exponents by tracking the growth and decay of explicit families of convex bodies with multiple independent parameters (Kawashima et al., 2022).

6. Extensions: Number Fields, Adelic Settings, and Higher-Rank Flows

The template formalism and parametric geometry of numbers have been extended to number fields $K$. For each place $w$ of $K$, the approximation theory over the local field $K_w$ (archimedean or non-archimedean) is modeled by associating adelic convex bodies scaled anisotropically at $w$ and studying their minima functions (Poëls et al., 2022). The same combinatorial n-system structure and joint spectrum of exponents emerge across all $(K, w)$, revealing deep universality in metric Diophantine phenomena. Scalar extension techniques provide explicit connections between spectra in $K_w^n$ and $\mathbb{Q}_\ell^{nd}$ and enable the construction of algebraic curves with extremal approximation properties. Generalizations to higher-rank diagonal flows, non-diagonal groups, and reductive homogeneous spaces remain an active direction for future work (Solan, 2021).

7. Open Problems and Outlook

While the template and variational approach resolves a wide range of longstanding questions and provides a flexible toolkit for further dimension analysis, several challenges remain. The explicit combinatorial modeling of multidimensional (beyond one-dimensional rays) rigid systems/templates is incomplete, particularly for genuinely multiparametric phenomena (cell decompositions, intricate gradient changes). Full classification and rigidity results in the higher-parameter setting, as well as extensions to flows associated with flag varieties, inhomogeneous problems, and non-Euclidean norms, are open research frontiers (Roy, 2014, German, 2020, Solan, 2021).

---

Summary Table: Principal Multi-Parametric Constructs

Object	Definition/Structure	Role/Purpose
Template / n-system	Piecewise-linear, ordered d-tuple, slopes quantized	Encodes joint behavior of minima, replaces rigid system
Multi-parametric body	$K(t) = \mathrm{diag}(e^{t_1},...,e^{t_d}) B$	Parameterizes simultaneous scaling in multiple directions
Variational principle	Supremum of combinatorial averages over templates	Computes dimension of sets with prescribed Diophantine properties
Schmidt game (multi)	Strategic selection in parameter/lattice space	Realizes extremal/range behavior of minima functions

Further technical details, precise definitions, and explicit constructions are found in (Das et al., 2017, Das et al., 2019, German, 2020, German, 2018, Roy, 2014, Solan, 2021), and related works.