---
title: Quantitative Khintchine Theorems
url: https://www.emergentmind.com/topics/quantitative-khintchine-theorem
type: topic
---

# Quantitative Khintchine Theorems

Quantitative refinements of Khintchine’s theorem form a family of results in metric Diophantine approximation rather than a single theorem. In the classical simultaneous setting, for a decreasing approximation function \(\psi:\mathbb N\to \mathbb R_{>0}\), one considers
\[
W(\psi)=\left\{\mathbf x\in\mathbb R^d:\|q\mathbf x\|<\psi(q)\ \text{for infinitely many }q\in\mathbb N\right\},
\]
and Khintchine’s theorem asserts a zero–full law according as \(\sum_{q=1}^\infty \psi(q)^d\) converges or diverges [2002.06221]. Quantitative versions sharpen this template in several directions: by counting approximants asymptotically, by giving effective lower bounds for the measure of uniform non-approximability sets, by proving Hausdorff-measure or exact-dimension analogues, or by supplying explicit probabilistic and overlap estimates strong enough to force the same metric dichotomy in restricted or random models [2011.06360].

## 1. Classical template and the main quantitative interpretations

The classical Khintchine–Groshev framework for matrices \(\vartheta\in \mathrm M_{m\times n}(\mathbb R)\) studies inequalities of the form
\[
\|\vartheta \mathbf q+\mathbf p\|^m<\psi(\|\mathbf q\|^n),
\]
with \((\mathbf p,\mathbf q)\in \mathbb Z^m\times \mathbb Z^n\), and the usual divergence criterion is \(\sum_{t=1}^\infty \psi(t)\) [2011.06360]. Quantitative work begins when the zero–one law is replaced by finer statements about counts, measure loss, local ubiquity, or fractal size.

| Quantitative mode | Typical conclusion | Representative papers |
|---|---|---|
| Asymptotic counting | \(\mathcal N(\vartheta,T)\sim\) explicit main term | [2011.06360], [2205.06425], [2003.02243] |
| Metric threshold and Hausdorff law | full \(\mathcal H^s\)-measure under explicit divergence and Diophantine conditions | [2002.06221], [2505.01227] |
| Effective convergence | explicit \(\kappa\) giving good set of measure at least \(1-\delta\) | [2209.14196], [2106.04806] |
| Probabilistic quantitative inputs | explicit expectation, variance, and overlap bounds | [1909.07256], [1708.02874] |
| Continued-fraction large deviations | explicit deviation bounds for \(\frac1n\log M_n\) | [1903.00255] |

A persistent source of ambiguity is that “quantitative” does not have a unique meaning. Several papers explicitly distinguish between metric or fractal-dimensional quantitative statements and effective counting asymptotics. For affine subspaces, for example, the full-measure implication from a divergent sum is described as a “quantitative metric threshold theorem,” but not as an asymptotic rational-point counting theorem [2002.06221]. Conversely, on congruence-restricted lattices and on spheres, the main theorems are genuine asymptotic counting laws [2011.06360], [2003.02243].

## 2. Asymptotic counting theorems

The strongest quantitative form is an almost-everywhere asymptotic for the number of approximants up to height \(T\). In the congruence-restricted Khintchine–Groshev theorem, for \(d=m+n\ge 3\), continuous non-increasing \(\psi\), residue data \((\mathbf v,N)\), and norms \(\nu_1,\nu_2\), the counting function
\[
\mathcal N(\vartheta,T)
\]
for solutions to
\[
\nu_1(\vartheta \mathbf q+\mathbf p)^m<\psi(\nu_2(\mathbf q)^n),\qquad (\mathbf p,\mathbf q)\equiv \mathbf v \pmod N,\qquad 1\le \nu_2(\mathbf q)^n<T
\]
satisfies
\[
\mathcal N(\vartheta,T)\sim N^{-d}c_{\nu_1}c_{\nu_2}\sum_{1\le t<T}\psi(t)
\]
for almost every \(\vartheta\in \mathrm M_{m\times n}(\mathbb R)\) [2011.06360]. The main term is obtained by encoding approximation as lattice points in a region
\[
E_T=\{(\mathbf x,\mathbf y):\nu_1(\mathbf x)^m<\psi(\nu_2(\mathbf y)^n),\ 1\le \nu_2(\mathbf y)^n<T\},
\]
computing \(|E_T|\), and combining Schmidt-type counting for generic lattices with a transfer argument from generic lattices to the unipotent family \(u(\vartheta)\).

An \(S\)-arithmetic analogue replaces the single approximation function by a collection \(\psi=(\psi_p)_{p\in S}\) and imposes separate local inequalities
\[
\|A_pq+p\|_p^m\le \psi_p(\|q\|_p^n)\qquad (p\in S).
\]
For almost every \(A\in \mathrm{Mat}_{m,n}(\mathbb Q_S)\), the counting function \(N_{\psi,A}(T)\) is asymptotic to
\[
\frac{V_\psi(T)}{N^d},
\]
where
\[
V_\psi(T)=2^m \int_{\{y\in \mathbb Q_S^n:\ \|y\|_p\le T_p,\ \forall p\in S\} \prod_{p\in S}\psi_p(\|y\|_p^n)\,dy
\]
is the volume of the corresponding \(S\)-arithmetic target region [2205.06425]. The paper proves asymptotics along increasing sequences \(T_r\), and under an additional regularity condition on maximal and minimal regions, for all sufficiently large \(T\).

Intrinsic approximation on spheres yields another counting form. For the unit sphere \(\mathbb S^n\), the number \(N_{T,c}(\alpha)\) of primitive rational points \(\frac pq\in \mathbb S^n\) satisfying
\[
\left\|\alpha-\frac pq\right\|<\frac{c}{q},\qquad 1\le q<\cosh T
\]
obeys
\[
\frac{N_{T,c}(\alpha)}{T}\to \eta(c)
\]
for almost every \(\alpha\in \mathbb S^n\), equivalently
\[
N_{T,c}(\alpha)\sim |E_{T,c}|\sim \eta(c)T
\]
[2003.02243]. The same paper proves a spiraling theorem:
\[
N_{T,c,A}(\alpha,k)\sim |E_{T,c,A}|\sim \mathrm{vol}(A)\,\eta(c)\,T,
\]
so the directions of approximation errors become equidistributed in prescribed angular sectors.

These counting theorems are Schmidt-type refinements of Khintchine–Groshev laws. They do not merely assert infinitely many approximants; they identify the main term and, at the lattice level, arise from variance bounds or second-moment estimates on spaces of lattices [2011.06360], [2205.06425].

## 3. Metric threshold theorems, Hausdorff measure, and fractal dimension

A second quantitative interpretation keeps the zero–full structure but makes the threshold explicit on subspaces or manifolds. For affine subspaces
\[
\mathcal L=\{(\mathbf x,\mathbf x\mathfrak a+\mathbf a_0)\in \mathbb R^d:\mathbf x\in \mathbb R^n\},
\]
the tilt matrix \(\mathfrak a\) carries a multiplicative Diophantine exponent \(\omega(\mathfrak a)\). If
\[
\omega(\mathfrak a)<dn,
\]
then \(\mathcal L\) is of Khintchine type for divergence: for every decreasing \(\psi\) with
\[
\sum_{q=1}^\infty \psi(q)^d=\infty,
\]
almost every point of \(\mathcal L\) is \(\psi\)-approximable [2002.06221]. More generally, if
\[
\omega(\mathfrak a)<\frac{n(d-n+s)}{n+1-s},
\]
then
\[
\mathcal H^s(W(\psi)\cap \mathcal L)=\mathcal H^s(\mathcal L)
\quad\text{whenever}\quad
\sum_{q=1}^{\infty}\psi(q)^{d-n+s}q^{n-s}=\infty.
\]
In the power-law case \(\psi(q)=q^{-\tau}\), the same framework yields the exact formula
\[
\dim \bigl(W(\tau)\cap \mathcal L\bigr)=n-\frac{\tau d-1}{\tau+1}
\quad\text{under}\quad
\omega(\mathfrak a)<\frac n\tau
\]
[2002.06221]. The paper explicitly interprets these results as metric and fractal-dimensional quantitative statements rather than effective counting asymptotics.

For arbitrary nondegenerate manifolds, the 2025 resolution of the long-standing divergence problem proves the inhomogeneous manifold analogue of the Khintchine theorem:
\[
L_d\big(f^{-1}(\mathcal S_n^\theta(\psi))\big)=
\begin{cases}
0 & \text{if } \sum_q \psi(q)^n<\infty,\\
L_d(U) & \text{if } \sum_q \psi(q)^n=\infty,
\end{cases}
\]
with \(M=f(U)\subset \mathbb R^n\) nondegenerate and \(\theta\in \mathbb R^n\) [2505.01227]. Quantitative content enters through rational-point counts near the manifold. Writing \(n=d+m\), the paper proves a lower bound
\[
N^\star_\theta(B;\varepsilon,t)\gg \varepsilon^m e^{(d+1)t}L_d(B)
\]
for \(\varepsilon\gg e^{-\eta t}\), and an upper bound
\[
N^\star_\theta(B_0;\varepsilon,t)\ll \varepsilon^m e^{(d+1)t}
+e^{(d+1)t}\left(\varepsilon^{\,n-1/2}e^{3t/2}\right)^{-\alpha},
\]
where
\[
\eta=
\begin{cases}
\frac1{n-1},& d>1,\\[1ex]
\frac{3}{2n-1},& d=1,
\end{cases}
\qquad
\alpha=\frac{1}{d(2l-1)(n+1)}.
\]
This lower–upper pair yields local ubiquity for divergence and summable coverings for convergence, and also feeds a Jarník-type Hausdorff theorem [2505.01227].

A common misconception is that such results are merely qualitative because the final conclusion is “full measure.” The subspace and manifold papers show that explicit divergence criteria, explicit Diophantine thresholds, and exact Hausdorff-dimension formulas are themselves quantitative, even when no asymptotic rational-point count with power-saving error term is claimed [2002.06221], [2505.01227].

## 4. Effective convergence theorems and explicit measure estimates

A third major interpretation of “quantitative Khintchine theorem” is effective control in the convergence regime. In simultaneous approximation on nondegenerate manifolds, Datta proved an effective version of the convergence theorem: for any prescribed loss of measure \(0<\delta<1\), one can choose an explicit \(\kappa\) so that
\[
\|qf(x)\|_{\mathbb Z}>\kappa\,\psi(q)\qquad \forall q\in \mathbb Z\setminus\{0\}
\]
holds on a set of measure at least \(1-\delta\) [2209.14196]. The proof makes major-arc and minor-arc contributions explicit and turns Beresnevich–Yang’s qualitative measure-zero theorem into a uniform lower bound for the measure of the good set
\[
\mathcal B(\kappa,\psi)=\{x\in U:\|qf(x)\|_{\mathbb Z}>\kappa\psi(q)\ \forall q\neq 0\}.
\]
This is effective in the strong sense that \(\kappa\) depends explicitly on \(\delta\), \(\psi\), and local manifold data.

A function-field analogue for affine hyperplanes goes further in the direction of explicit constants. For
\[
F=\mathbb F_q((T^{-1}))
\]
and an affine hyperplane
\[
\mathscr H=\{(\mathbf x,\mathbf x\cdot \mathbf a):\mathbf x\in F^{n-1}\},
\]
the convergence-case theorem states that \(\lambda(\mathcal W(\mathscr H;\psi))=0\) under a Diophantine condition on the coefficients \(a_i\) and the convergence of
\[
\sum_{t=0}^\infty \psi(q^t)q^{nt}
\]
[2106.04806]. The quantitative strengthening defines
\[
\mathcal B(U,\psi,k)=\{x\in U: |(x,x\cdot a)\cdot q+p|\ge k\,\psi(\|q\|)\ \forall q\in A^n\setminus\{0\},\ \forall p\in A\},
\]
and proves the existence of explicitly computable constants \(K_0\) and \(K_1\), depending on \(n\), \(U\), and \(a\) only, such that for any \(\xi\in(0,1)\),
\[
\lambda(\mathcal B(U,\psi,k))\ge (1-\xi)\lambda(U)
\]
whenever
\[
k<\min\left\{1,\ \frac{\xi}{2K_0K_1}\right\}
\]
[2106.04806].

In positive characteristic, the convergence theorem for analytic nonplanar manifolds is again qualitative at the final level but quantitatively structured in the proof. The core estimate is
\[
\mathcal N_{\boldsymbol\Theta}(B\setminus \mathfrak M(s,t);s,t)\ll q^{(d+1)t-ms}\mathcal L_d(B),
\]
together with the special-set measure bound
\[
\mathcal L_d(\mathfrak M(s,t)\cap B)\le C \max\left\{ q^{-\alpha t}, q^{\alpha\left(\frac{2n-1}{2(n+1)}s-\frac{3}{2(n+1)}t\right)} \right\}\mathcal L_d(B),
\]
which together imply summability and hence the convergence statement [2511.01991]. This suggests a broad taxonomy: some quantitative Khintchine theorems are effective theorems about the final exceptional set, while others are convergence theorems whose quantitative content lies in the counting and measure estimates that drive Borel–Cantelli.

## 5. Random, moving-target, and restricted-source variants

Quantitative Khintchine theory also includes models in which the set of admissible rationals is random or structurally constrained. In the random-fractions model with prescribed numbers of numerators, if the average order of \(f\) is positive and bounded, or at least linear in \(n\), then for \(\mathbb P_f\)-almost every realization \(P\), and for every decreasing \(\Psi\),
\[
\lambda(W^P(\Psi))=
\begin{cases}
1 & \text{if }\sum_{n=1}^\infty f(n)\Psi(n)=\infty,\\
0 & \text{if }\sum_{n=1}^\infty f(n)\Psi(n)<\infty.
\end{cases}
\]
The proof uses block sums \(F_t\), explicit expectation and variance bounds, Chebyshev, and local ubiquity [1708.02874]. The same paper proves that if
\[
\frac{f(n)\log\log n}{n}\to\infty,
\]
then monotonicity cannot in general be removed: there exists \(\Psi\) with \(\sum f(n)\Psi(n)=\infty\) but \(\lambda(W^P(\Psi))=0\) for every realization \(P\) [1708.02874].

A Bernoulli random model gives a monotonicity-free theorem under logarithmic sparsity. If
\[
p_n\ll (\log n)^{-\varepsilon}
\quad\text{for some }\varepsilon>0,
\]
then almost surely
\[
\lambda(W^P(\psi))=
\begin{cases}
0 & \text{if } \sum_{n=1}^\infty p_n\psi(n)<\infty,\\
1 & \text{if } \sum_{n=1}^\infty p_n\psi(n)=\infty,
\end{cases}
\]
for every \(\psi:\mathbb N\to[0,1/2]\), and the proof is built from explicit first-moment, variance, and overlap estimates rather than from a separate zero–one law [1909.07256].

Moving-target inhomogeneous approximation introduces another quantitative defect term. For
\[
A_q=\{\alpha\in[0,1]:\|q\alpha-\gamma_q\|<\psi(q)\},
\]
the key overlap estimate is
\[
m(A_q\cap A_r)\le 2\,m(A_q)m(A_r)+\frac{\gcd(q,r)}{q}\,m(A_q),
\]
which explains why extra divergence is needed in the general moving-center problem [2506.04187]. One theorem shows that if \(\psi\) is decreasing and
\[
\sum_{q=1}^\infty \frac{\psi(q)}
{\sqrt{\log q}\,(\log\log q)\,(\log\log\log q)\cdots \bigl(\underbrace{\log\log\cdots\log}_{k\text{ iterates}} q\bigr)^{1+\varepsilon}}
=\infty,
\]
then \(m(W(\psi,\gamma))=1\) for every target sequence \(\gamma_q\); a simpler sufficient condition is
\[
\sum \psi(q)(\log q)^{-1/2-\varepsilon}=\infty \Longrightarrow m(W(\psi,\gamma))=1
\]
[2506.04187]. A second theorem proves the full moving-target statement under the classical divergence condition \(\sum\psi(q)=\infty\) when the centers range over only finitely many values. In the fast-divergence regime, the paper also invokes Sprindžuk’s asymptotic formula
\[
N(Q,\alpha)=\Phi(Q)+ O\!\left((\Phi(Q)\log Q)^{1/2}\bigl[\log(\Phi(Q)\log Q)\bigr]^{3/2+\varepsilon}\right),
\qquad
\Phi(Q)=\sum_{q\le Q}2\psi(q),
\]
which is a direct counting theorem [2506.04187].

A different restricted-source phenomenon appears when rationals are required to lie in a fixed ball in one completion of \(\mathbb Q\) while approximation is measured in another completion. The resulting measure law still has the classical one-dimensional criterion
\[
\mu_{p_2}(W_{p_2}(\psi,B_{p_1}))=
\begin{cases}
1 & \sum_{n=1}^\infty n\psi(n)=\infty,\\
0 & \sum_{n=1}^\infty n\psi(n)<\infty,
\end{cases}
\]
and the Hausdorff refinement is
\[
\mathcal H^f(W_{p_2}(\psi,B_{p_1}))=
\begin{cases}
\mathcal H^f(\mathbb Z_{p_2}) & \sum_{n=1}^\infty n f(\psi(n))=\infty,\\
0 & \sum_{n=1}^\infty n f(\psi(n))<\infty,
\end{cases}
\]
with \(\dim W_{p_2}(\tau,B_{p_1})=2/\tau\) for \(\psi(q)=q^{-\tau}\), \(\tau\ge 2\) [2006.14764]. Here the quantitative feature is the preservation of the same critical sum despite strong local restrictions on the admissible rationals.

## 6. Continued fractions, Khintchine’s constant, and fast spectra

In continued-fraction theory, the phrase “quantitative Khintchine theorem” often refers not to rational approximation by inequalities \(\|qx\|<\psi(q)\), but to quantitative versions of the theorem on Khintchine’s constant. For an irrational \(\omega=[a_1,a_2,\dots]\), define
\[
M_n(\omega)=a_1(\omega)\cdots a_n(\omega),\qquad \log M_n=\sum_{j=1}^n \log a_j.
\]
The classical theorem states that for Lebesgue-almost every \(\omega\),
\[
\frac1n\log M_n(\omega)\to \kappa,\qquad e^\kappa=2.685\ldots
\]
[1903.00255]. The quantitative refinement studies the sets
\[
KL^\pm(T,N)=\{\omega: M_n(\omega)\le e^{(\kappa+T)n}\ \forall n\ge N\},
\]
\[
KL^-(T,N)=\{\omega: e^{(\kappa-T)n}\le M_n(\omega)\ \forall n\ge N\},
\]
and gives explicit lower bounds on their Gauss measure. The central deviation inequality is
\[
\gamma\!\left(\pm\left(\frac1n\log M_n-\kappa\right)\ge T\right)\le \Xi(T)^{\sqrt n},
\]
hence
\[
\gamma\!\left(\left|\frac1n\log M_n-\kappa\right|>T\right)\le 2\Xi(T)^{\sqrt n},
\]
and summation over \(n\ge N\) yields explicit lower bounds for the measure of numbers whose growth remains within \(\kappa\pm T\) for all \(n\ge N\) [1903.00255]. The paper stresses that the estimates are explicit but numerically conservative.

A related but distinct quantitative theory concerns exceptional continued-fraction growth at superlinear speed. If \(\psi(n)/n\to\infty\), define the upper and lower fast Khintchine sets by
\[
\overline E(\psi)=\left\{x\in[0,1]: \limsup_{n\to\infty}\frac{\log a_1(x)+\cdots+\log a_n(x)}{\psi(n)}=1\right\},
\]
\[
\underline E(\psi)=\left\{x\in[0,1]: \liminf_{n\to\infty}\frac{\log a_1(x)+\cdots+\log a_n(x)}{\psi(n)}=1\right\}.
\]
If
\[
\liminf_{n\to\infty}\frac{\log \psi(n)}{n}=\log b,\qquad
\limsup_{n\to\infty}\frac{\log \psi(n)}{n}=\log B,
\]
with \(b,B\in(1,\infty]\), then
\[
\dim_H \overline E(\psi)=\frac{1}{1+b},
\qquad
\dim_H \underline E(\psi)=\frac{1}{1+B}.
\]
The same paper recalls the earlier exact-limit spectrum
\[
\dim_H E(\psi,1)=\frac{1}{1+B(\psi)}
\]
when \(\psi\) is equivalent to a nondecreasing function, where
\[
B(\psi)=\limsup_{n\to\infty}\frac{\psi(n+1)}{\psi(n)}
\]
[1406.1148]. These formulas show that upper, lower, and exact fast Khintchine spectra can differ. Quantitative Khintchine theory in this continued-fraction sense is therefore a theory of explicit large deviations and exact Hausdorff dimensions for exceptional growth sets, not of rational-point counting.

Taken together, these strands show that “quantitative Khintchine theorem” has stabilized as a genuinely plural notion. In one direction it denotes Schmidt-type asymptotic counting theorems for approximants; in another it means explicit convergence-side measure estimates; in another it means Hausdorff-measure and dimension formulas for limsup sets on subspaces, manifolds, or continued-fraction phase spaces. The unifying feature is that the classical zero–one metric law is replaced by explicit asymptotics, explicit thresholds, or explicit exceptional-set geometry [2011.06360], [2209.14196], [2002.06221], [1903.00255].

Source: https://www.emergentmind.com/topics/quantitative-khintchine-theorem