WorksMathematics
Twisted Diophantine approximation: asymptotic and uniform results
Abstract
Fix an matrix ; twisted Diophantine approximation studies how well we can approximate the affine form by an integer -vector, depending on the choice of -vector . We report on two companion papers in twisted Diophantine approximation, covering two setups: asymptotic and uniform. The asymptotic paper gives an exact Khintchine-type Lebesgue zero–one law, a null–conull criterion for badly approximable shifts, and Hausdorff-measure tests yielding exact power-law dimensions under a critical divergence condition. The uniform paper computes Hausdorff dimensions of sets of uniformly approximable vectors. It also gives Hausdorff zero–full laws, uniform-exponent spectra, and a sharp escape-of-mass bound. In both cases the results are established for arbitrary approximation functions from suitable regularity classes, and the arguments use a homogeneous lattice count involving all successive minima to record clustering in every independent direction.
Introduction
Diophantine approximation: asymptotic and uniform
Let be positive integers, set , and write for the space of real matrices. Classical theory of homogeneous Diophantine approximation asks, for and a positive non-increasing function , where , whether there are integer vectors and satisfying
at scale . Here bounds the size of , while specifies the allowed approximation error; throughout, denotes the supremum norm.
Dirichlet’s theorem (1842) states that (1.1) is solvable for every and every when . For a general , we may ask two types of questions: do solutions exist for an unbounded set of ? Or do solutions exist for all sufficiently large ? In the former case we call -approximable, and in the latter -Dirichlet1; these are the asymptotic and uniform conditions, respectively [20 Section 1.1].
The metric theory asks how large these two classes of matrices are. For the asymptotic condition, the Khintchine–Groshev theorem [11, 7] states that the set of -approximable matrices is null or conull according as
converges or diverges; see [3 Section 1] for this formulation. For the uniform condition, the set of -Dirichlet matrices is likewise null or conull under suitable regularity assumptions on . This zero–one law was established by Kleinbock and Wadleigh for [19], while in arbitrary dimensions the partial result of Kleinbock, Strömbergsson and Yu [18] was completed by Strömbergsson and Yu [27].
Inhomogeneous approximation
In the setup of inhomogeneous Diophantine approximation, we replace the linear form by an affine form , where . We study the system
The same distinction between asymptotic and uniform approximation now applies to pairs: we call -approximable if (3) is solvable for unbounded , and -Dirichlet if it is solvable for all sufficiently large , with allowed here. In the inhomogeneous setting, Dirichlet’s bound is no longer valid for every pair . Indeed, if has integer entries and , then (3) has no solution once is small enough.
The metric results depend on which parameters are allowed to vary. When a nonintegral shift is fixed and varies, the asymptotic problem retains the classical measure law: the set of -approximable matrices is null or conull according as the series in (2) converges or diverges [Szű58, Sch64, Spr79]. With both and varying, the set of -Dirichlet pairs is also null or conull by the zero–one law of Kleinbock and Wadleigh [20]. The first-named author and Kim further studied the complementary non-Dirichlet sets, establishing Hausdorff-measure laws for pairs and for fixed when is decreasing and tends to zero [14]. In these results, the matrix remains a variable. The question addressed in this report is how the two approximation conditions behave when is fixed and only the shift varies.
Twisted approximation
This choice is usually called twisted approximation. From now on, we assume as . Since integer translates of do not affect solvability of (3), shifts naturally belong to ; we write for the image of . For a fixed , let and be the sets of shifts for which (3) is solvable for unbounded and for all sufficiently large , respectively.2
We briefly recall the main metric results for the asymptotic sets . Kurzweil’s theorem [21] states that if is badly approximable, then has zero or full measure according as the series (2) converges or diverges. Moreover, this law holds for every non-increasing only if is badly approximable. Beyond badly approximable matrices, general exact criteria were previously available only in the case : Fuchs and Kim [6] gave an exact criterion based on the continued fraction expansion; later, Kim, Rams and Wang [17] determined the exact Hausdorff dimensions of these sets. For nonsingular matrices, the first-named author [12] gave a sufficient full-measure condition, extended to the weighted setting by Hussain and Ward [9]. The asymptotic problem we consider is to determine when is null or conull for an arbitrary fixed .
We next recall the main metric results for the uniform sets . Transference relates the approximation of shifts to homogeneous approximation of . Jarník [Jar39] gave conditions for every shift to be uniformly approximable, proving an analogue of Dirichlet’s theorem in the twisted setup. Bugeaud and Laurent [4] related almost-everywhere asymptotic exponents to uniform homogeneous exponents, and Moshchevitin and the second-named author [22] obtained a zero–one law for the Lebesgue measure of , up to multiplicative constants in . These results do not in general determine the Hausdorff dimension of when it is Lebesgue null. Kim and Liao [13] obtained Hausdorff-dimension results for irrational rotations when is a power function. Aggarwal [1] obtained dimension bounds for singular affine forms. The uniform problem we consider is to determine dimensions for arbitrary fixed matrices.
The two companion papers
The two companion papers address these questions for arbitrary fixed matrices . For non-increasing satisfying the regularity condition (6), the asymptotic paper [15] gives a convergence–divergence criterion deciding whether is null or conull. It also derives results on twisted badly approximable vectors and Hausdorff measure and dimension.
For continuous, strictly decreasing satisfying the regularity condition (8), the uniform paper [16] computes the Hausdorff dimensions of the sets and , and in many cases of . For powers and near powers, it determines for every , except at finitely many exponents depending only on .
Different denominator vectors can give the same or nearby points in the projected orbit when maps their differences close to integer vectors. Both papers express their formulae through a homogeneous lattice count that measures this clustering. § 2 states the precise hypotheses and main results, and § 3 explains how this count enters the two arguments. Complete proofs are given in the companion papers.
Conventions. All balls in use the quotient supremum metric, and denotes Hausdorff dimension. Write for normalized Lebesgue measure on . The symbols , and allow positive constants depending on , unless additional fixed parameters are indicated.
Use of artificial intelligence. ChatGPT Codex was used in editing, proofreading and creating figures for this article. The authors take full responsibility for all the content of this manuscript.
Acknowledgements. The authors are grateful to Dmitry Kleinbock for useful suggestions and discussion.
Main results
First, we introduce the lattice count which enters both asymptotic and uniform results.
Lattice counts and the diagonal action
To encode the Diophantine properties of , associate to it the unimodular lattice
For , define
Thus counts integer pairs with error at most and denominator size at most . Such pairs describe differences between nearby projected orbit points, so the count records their clustering. It includes the zero vector and is therefore always positive.
We use successive minima relative to the fixed unit box . For a full-rank lattice , let be the least such that contains linearly independent vectors of . Consider the diagonal action
Write . For , put
The map sends the rectangle to , so . Standard lattice-counting estimates therefore give
This comparison is uniform in [8, 15]. Thus the count is determined by the successive minima along , connecting both approximation problems with parametric geometry of numbers [25, 5].
A zero–one law for asymptotic approximation
We call dyadically regular if, for some ,
For instance, if , the power function satisfies this condition. The asymptotic measure law is as follows.
The summand in (7) separates volume from clustering:
The numerator is the -th dyadic term in (2), while the denominator corrects for clustering in the projected orbit. For badly approximable matrices and irrational rotations, this recovers the Kurzweil and Fuchs–Kim criteria, respectively, within the class of dyadically regular functions.
Lagrange constants and badly approximable shifts
Define the inhomogeneous -Lagrange constant by
Under the hypotheses of Theorem 2.1, this constant is almost surely or according as the series in (7) diverges or converges [15 Corollary 2.1]. At the Dirichlet exponent , define . At time , (5) gives
The measure of badly approximable shifts is then determined as follows.
Theorem 2.2 [15 Proposition 2.2]. Let . Then
This answers the question in [12 Remark 1.8(3)] and the Lebesgue-measure case of [MRS24, Question 1.3].
Recall that is singular if, for every , (1) is solvable for all sufficiently large with ; otherwise it is nonsingular. For a nonsingular matrix, the count is bounded for infinitely many , so and is null [15 Corollary 2.4].
Asymptotic Hausdorff measure and dimension
To describe the size of asymptotic approximation sets more finely, we turn to Hausdorff measure. For non-increasing, dyadically regular and , the count gives two sufficient conditions [15 Corollary 3.1]:
The two denominators need not agree, so these implications do not form a single convergence–divergence criterion. The first counts clusters at the original radius, while the second uses the enlarged radius required by mass transference. When the critical series diverges, these tests nevertheless determine the Hausdorff dimension for every power function [15 Corollary 3.2]:
In particular, this formula holds for every nonsingular .
Uniform Hausdorff dimension
The dimension of may depend on . To eliminate this dependence, define the endpoint sets
We refer to the elements of simply as twisted singular vectors; see [1 Section 2.4] for the corresponding fixed-matrix problem for singular affine forms.
We call lower quasimultiplicative if there are and such that
For continuous, strictly decreasing tending to zero, we will assume as , so that is defined on . This assumption does not affect the set .
For , set
and
The count ratio governs how many smaller approximation balls can be retained at each step; measures the total logarithmic shrinkage of the radii. Our main result regarding the uniform setup is as follows.
Theorem 2.3 [16 Theorem 1.2]. Let and let be continuous, strictly decreasing, lower quasimultiplicative and tend to zero.
(i) The limits
exist, and
(ii) The quantity is independent of if and only if . Under this condition,
In particular, lies between and .
Under the hypotheses of Theorem 2.3, both and are Borel subgroups of and each has either zero or locally full -measure for every dimension function [16 Theorem 1.9]. Here is continuous and nondecreasing with as ; locally full means for every nonempty relative ball in .
Power functions and uniform exponents
For power functions, the dimension of is independent of outside the finite set
In this case, we can state the dimension formula in a simpler way than in Theorem 2.3. Put
Thus counts the minima below at time along $a_t\Lambda_A.`
Theorem 2.4 [16 Theorem 1.4]. Let and . For every ,
Write for this common value. It is continuous on each component of . Moreover, if is continuous, strictly decreasing to zero, and
then
No lower quasimultiplicativity is required for this last assertion. For these powers, both and have dimension . The near-power assertion applies, for example, to for every fixed .
The exclusions are necessary: for each , there is a matrix with for small and for large [16 Remark 1.5]. In the case, for irrational , continued fractions give formulae for under the hypotheses of Theorem 2.3 whenever the ratio stays uniformly below or above [16 Theorem 2.2], extending [13].
These formulae also describe the uniform exponent, defined by , with . Write for the one-sided limits of the non-increasing function , taken through exponents outside . For every , including those belonging to the exceptional set , [16 Proposition 1.6] gives
Every nonempty exact level has dimension .
Twisted singular vectors
2.7. Twisted singular vectors. For singular affine forms, a simpler bound uses only the first minimum along . Define the lower escape of mass by
It measures the lower proportion of time spent with an arbitrarily small first minimum.
Theorem 2.5 [16 Propositions 2.4–2.5]. For every ,
For every , there is an with attaining equality.
This improves the coefficient in [1]; the theorem is optimal when only lower escape of mass is used.
The common geometry: clusters and scales
Fix a large . Partition into half-open cubes of side comparable to and at most ; in the supremum norm these are ball-shaped cells. Denote this partition by , and call a cell good if it contains at least one orbit point with denominator bound :
Every shift in a good cell satisfies (3), since it is within of the orbit point in that cell. Conversely, a shift satisfying (3) lies within distance of a good cell. Thus the partition description agrees with the approximation condition up to a multiplicative change of radius, bounded independently of and . Figure 1 shows the question we need to answer: how many cells are good?

Figure 1. Blue orbit centers visit the orange good cells. The partition has cells of side ; the total number of points is , the number of points in one cell is , and .
There are about points with , counted with multiplicity, but this need not give different good cells. Several vectors can visit the same cell, or even give exactly the same torus point. To understand this multiplicity, lift the picture from to the lattice in (4). The extra coordinates keep track of the denominator vectors rather than forgetting them under projection.
First count all visits. For each with , there is exactly one for which . Hence
This is comparable to the volume of the cylinder.
Next look at the visits to a good cell . In a lift of , these correspond to the points of . Choose one such point . Since is a lattice, the translation which moves this point to the origin preserves . The number of points in a good cell is then measured, up to fixed enlargements of the window , by (Figure 2).

Figure 2. The orange window has base centered at . Lattice points in the orange window project to a cluster of orbit centers (, ).
Fixed enlargements are needed because the chosen lattice point need not be the center of its window. If has side length , then contains , where is the concentric triple dilate of . The enlarged window therefore contains at least lattice points. These changes cost only fixed factors, since (5) gives
One short lattice direction gives strings of nearby visits; several short directions give higher-dimensional clusters. This is why the count involves all successive minima, rather than only the shortest vector.
We can now read off the one-scale estimate as
Each cell has volume comparable to , so
Taking gives exactly the summand in Theorem 2.1: the usual block volume, divided by its clustering multiplicity. The same volume estimate holds for the cover by enlarged cells. If these volumes are summable, Borel–Cantelli gives measure zero. For divergence, we must also understand how the good cells at different scales overlap.
Fix and an earlier good cell . At the later time , the cells have diameter comparable to . We ask how many cells meet . The same total/local picture now takes place inside the parent cylinder:
To count the -level cells intersecting an -level cell, center the parent at one of its orbit points and give it radius . For the lower bound, we also assume and . Translation by this lattice point gives
The upper bound for cells meeting the original parent holds for every . Both counts use the later denominator bound (Figure 3).

Figure 3. Inside , later orbit centers are grouped at scale .
The asymptotic problem. The convergence case follows from the volume estimate and Borel–Cantelli, as shown above. For divergence, put , , and
where is chosen so that for , with . One has . We seek the following bound on total pairwise overlaps, which would show quasi-independence of the sets on average:
Suppose . A generalization of the divergent part of the Borel–Cantelli lemma, known as the Kochen–Stone inequality (see [26 Lemma 5]), yields that . A separate statement proved in the asymptotic paper shows that this measure is either zero or one; hence it equals one.
It remains to prove the quasi-independence estimate (13). Counting later good cells near each earlier good cell gives
We split the pairs according to whether
spans . In the full-rank case, (5) shows that the count is comparable to the window volume, giving a total overlap contribution . In case of any smaller rank, the estimates in [15], using the dyadic regularity of , give a total contribution . We add the diagonal contribution and prove (13).
The uniform problem. We use the full approximation targets
Fix and choose the levels from (9), so . For large , lower quasimultiplicativity ensures for all sufficiently large .
Starting at such an index , (12) gives an inner construction with disjoint children inside each parent and an outer covering construction. Enlarging the balls in the outer construction by a fixed factor makes its levels nested. The two intersections of nested finite unions satisfy
In both constructions, the number of chosen children of each level- parent is comparable to
Their radii at level are comparable to . Let be a constant, independent of the level and of sufficiently large , that bounds both the ratio of the number of chosen children to and its reciprocal.
Upper bound. The level- cover of has at most a constant multiple of balls. Its -dimensional covering cost is therefore at most a constant multiple of
Taking logarithms and using shows that this tends to zero along a subsequence when . Hence .
Lower bound. We use the mass distribution principle to obtain a lower bound for . Split each parent’s mass equally among its children. Each level- ball has mass at most . Disjointness bounds how many such balls meet a test ball of comparable radius. The mass distribution estimate for this construction gives .
In both bounds, fixed comparison factors contribute to the logarithm, hence after division by .
These estimates concern the discrete intersection . To pass to all sufficiently large , observe that
Countable stability gives . For the lower bound, if , the denominator chosen at is still at most , and its error is at most
Thus
giving .
The upper and lower bounds apply to and , respectively. The discrepancy in constants disappears for and , since their definitions take intersections and unions over all positive constants. Countable stability gives the dimension of . For , this gives only the upper bound , since an intersection can have smaller dimension. A separate no-dimension-drop argument [16 Proposition 5.4] proves equality, yielding the formula for in Theorem 2.3.
Applications. For asymptotic approximation, the Hausdorff-measure tests in § 2.4 follow from the same geometric considerations. The convergence test uses the -dimensional covering cost of each good cell; the divergence test enlarges the targets and applies mass transference [2].
For uniform approximation, the successive-minima expression (5) yields the rank formula (11) of Theorem 2.4 for powers outside . The critical power , for which the approximation boxes have constant volume , is treated separately: an upper estimate using the small successive minima gives the escape-of-mass bound in Theorem 2.5. DEPARTMENT OF MATHEMATICS, BRANDEIS UNIVERSITY, WALTHAM, MA 02453, USA
Email address: [email protected]
DEPARTMENT OF MATHEMATICS, BRANDEIS UNIVERSITY, WALTHAM, MA 02453, USA
Email address: [email protected]
References
- [1]Gaurav Aggarwal, Dimension bounds for singular affine forms, Mathematika 72 (2026), e70084.DOI
- [2]Victor Beresnevich and Sanju Velani, A mass transference principle and the Duffin–Schaeffer conjecture for Hausdorff measures, Annals of Mathematics 164 (2006), no. 3, 971–992.arxiv.org/abs/math/0412141
- [3]Victor Beresnevich and Sanju Velani, Classical metric Diophantine approximation revisited: The Khintchine–Groshev theorem, International Mathematics Research Notices 2010 (2010), no. 1, 69–86.arxiv.org/abs/0811.0809
- [4]Yann Bugeaud and Michel Laurent, On exponents of homogeneous and inhomogeneous Diophantine approximation, Moscow Mathematical Journal 5 (2005), no. 4, 747–766.DOI
- [5]Tushar Das, Lior Fishman, David Simmons, and Mariusz Urbański, A variational principle in the parametric geometry of numbers, Advances in Mathematics 437 (2024), 109435.DOI
- [6]Michael Fuchs and Dong Han Kim, On Kurzweil’s 0–1 law in inhomogeneous Diophantine approximation, Acta Arithmetica 173 (2016), no. 1, 41–57.arxiv.org/abs/1501.04714
- [7]A. V. Groshev, A theorem on a system of linear forms, Doklady Akademii Nauk SSSR 19 (1938), no. 3, 151–152 (in Russian).
- [8]Martin Henk, Successive minima and lattice points, Rendiconti del Circolo Matematico di Palermo. Serie II. Supplemento (2002), no. 70, part I, 377–384.DOI
- [9]Mumtaz Hussain and Benjamin Ward, Weighted twisted inhomogeneous Diophantine approximation, Nonlinearity 37 (2024), 125004.arxiv.org/abs/2307.13210
- [11]A. Khintchine, Zur metrischen Theorie der diophantischen Approximationen, Mathematische Zeitschrift 24 (1926), 706–714.DOI
- [12]Taehyeong Kim, On a Kurzweil type theorem via ubiquity, Acta Arithmetica 213 (2024), no. 2, 181–191.arxiv.org/abs/2306.00847
- [13]Dong Han Kim and Lingmin Liao, Dirichlet uniformly well-approximated numbers, International Mathematics Research Notices 2019 (2019), no. 24, 7691–7732.arxiv.org/abs/1508.00520
- [14]Taehyeong Kim and Wooyeon Kim, Hausdorff measure of sets of Dirichlet non-improvable affine forms, Advances in Mathematics 403 (2022), 108353.arxiv.org/abs/2006.05727
- [15]Taehyeong Kim and Vasiliy Neckrasov, Twisted Diophantine approximation I: Asymptotic theory, arXiv:2609.36411, 2026.arxiv.org/abs/2609.36411
- [16]———, Twisted Diophantine approximation II: Uniform theory, arXiv:2609.36412, 2026.arxiv.org/abs/2609.36412
- [17]Dong Han Kim, Michal Rams, and Baowei Wang, Hausdorff dimension of the set approximated by irrational rotations, Mathematika 64 (2018), no. 1, 267–283.DOI
- [18]Dmitry Kleinbock, Andreas Strömbergsson, and Shucheng Yu, A measure estimate in geometry of numbers and improvements to Dirichlet’s theorem, Proceedings of the London Mathematical Society (3) 125 (2022), 778–824.arxiv.org/abs/2108.04638
- [19]Dmitry Kleinbock and Nick Wadleigh, A zero-one law for improvements to Dirichlet’s theorem, Proceedings of the American Mathematical Society 146 (2018), no. 5, 1833–1844.arxiv.org/abs/1609.06780
- [20]Dmitry Kleinbock and Nick Wadleigh, An inhomogeneous Dirichlet theorem via shrinking targets, Compositio Mathematica 155 (2019), 1402–1423.DOI
- [21]Jaroslav Kurzweil, On the metric theory of inhomogeneous Diophantine approximations, Studia Mathematica 15 (1955), no. 1, 84–112.DOI
- [22]Nikolay Moshchevitin and Vasiliy Neckrasov, Metric theory of inhomogeneous Diophantine approximations with a fixed matrix, arXiv:2503.21180, 2025.arxiv.org/abs/2503.21180
- [25]Wolfgang M. Schmidt and Leonhard Summerer, Parametric geometry of numbers and applications, Acta Arithmetica 140 (2009), no. 1, 67–91.DOI
- [26]Vladimir G. Sprindžuk, Metric theory of Diophantine approximations, John Wiley & Sons, New York, 1979. English translation.
- [27]Andreas Strömbergsson and Shucheng Yu, A zero-one law for improvements to Dirichlet’s theorem in arbitrary dimension, arXiv:2602.16258, 2026.arxiv.org/abs/2602.16258