Wiki
Wiki

Research notes on every problem, and a library of the papers behind them. Built from the open erdos repository.

Updated


Source. Alexander Ostrowski, Mathematische Miszellen. IX. Notiz zur Theorie der Diophantischen Approximationen, Jahresbericht der Deutschen Mathematiker-Vereinigung 36 (1927), 178--180. The selected theorem is equation (3), printed p. 179; its complete short proof runs through equation (4) on printed p. 180.

Selected translated-interval theorem

Write {t}=t−⌊t⌋\{t\}=t-\lfloor t\rfloor. Let α∈R\alpha\in\mathbb R and j∈Z∖{0}j\in\mathbb Z\setminus\{0\}, and suppose β={jα}∈(0,1)\beta=\{j\alpha\}\in(0,1). For any u∈Ru\in\mathbb R, let JJ be the circle interval represented by [u,u+β)[u,u+\beta) modulo 11, including its initial endpoint and excluding its final endpoint. Set

NJ(n)=#{1≤m≤n:{mα}∈J}.N_J(n)=\#\{1\le m\le n:\{m\alpha\}\in J\}.

Then, for every positive integer nn,

∣NJ(n)−nβ∣<∣j∣.(O3)|N_J(n)-n\beta|<|j|. \tag{O3}

This is the nonempty proper-interval case of the source's (3), retaining arbitrary translation and the source's integer sampling parameter. The source allows real α\alpha; irrationality is not needed for this bound. A wrapped circle interval may appear as two intervals in [0,1)[0,1).

Complete source proof in modern notation

Suppose first that j>0j>0, and put ξ=1−u−β\xi=1-u-\beta. For any real tt, membership of {t}\{t\} in JJ is equivalent to

{t+ξ}∈[1−β,1).\{t+\xi\}\in[1-\beta,1).

Indeed, adding 1−u−β1-u-\beta sends the half-open arc from uu to u+βu+\beta to the half-open arc from 1−β1-\beta to 11 modulo 11. The chosen endpoint convention is preserved, including points exactly at an endpoint.

Since jαj\alpha differs from β\beta by an integer, if r={t+ξ}r=\{t+\xi\} then

{t+jα+ξ}−{t+ξ}={r+β}−r={β,0≤r<1−β,β−1,1−β≤r<1.\{t+j\alpha+\xi\}-\{t+\xi\} =\{r+\beta\}-r =\begin{cases} \beta,&0\le r<1-\beta,\\ \beta-1,&1-\beta\le r<1. \end{cases}

Summing this identity for t=mαt=m\alpha, 1≤m≤n1\le m\le n, gives

Sξ(n)=∑m=1n({(m+j)α+ξ}−{mα+ξ})=nβ−NJ(n).(O4)\begin{aligned} S_\xi(n) &=\sum_{m=1}^{n} \bigl(\{(m+j)\alpha+\xi\}-\{m\alpha+\xi\}\bigr)\\ &=n\beta-N_J(n). \tag{O4} \end{aligned}

Reindexing finite sums also gives

Sξ(n)=∑h=1j({(n+h)α+ξ}−{hα+ξ}).S_\xi(n)=\sum_{h=1}^{j} \bigl(\{(n+h)\alpha+\xi\}-\{h\alpha+\xi\}\bigr).

For completeness, if A(k)=∑m=1k{mα+ξ}A(k)=\sum_{m=1}^{k}\{m\alpha+\xi\}, each side equals A(n+j)−A(n)−A(j)A(n+j)-A(n)-A(j). Thus this equality also holds when n<jn<j. Each of its jj summands has absolute value strictly below 11, because both fractional parts belong to [0,1)[0,1). The triangle inequality gives ∣Sξ(n)∣<j|S_\xi(n)|<j, proving (O3) for j>0j>0.

If j<0j<0, the complementary circle interval JcJ^c has length 1−β={−jα}1-\beta=\{-j\alpha\} and positive index −j-j. Its half-open convention partitions the circle with JJ, so NJc(n)=n−NJ(n)N_{J^c}(n)=n-N_J(n). Hence

NJc(n)−n(1−β)=−(NJ(n)−nβ).N_{J^c}(n)-n(1-\beta)=-(N_J(n)-n\beta).

Applying the positive-index case to JcJ^c proves the same strict bound ∣j∣|j|. This is the source's complement reduction for the negative index. Every finite-sum and endpoint step needed for the selected result is now included. The proof uses no external equidistribution or Kesten theorem.

The zero-length case is outside the displayed selected statement; if {jα}=0\{j\alpha\}=0 with j≠0j\ne0 and JJ is interpreted as the empty arc, both discrepancy terms vanish and the same bound is immediate. No j=0j=0 strict bound is asserted.

Relation to the endpoint question

The source first discusses intervals whose endpoints are rotation-orbit points and then explicitly extends the bound to arbitrary translations. Kesten's Theorem 4 supplies the separate necessity of the length condition; its proof is not included here.

The complete selected source proof survived whole-proof review by the independent source reviewer. A distinct grader passed the report contract and independence; the [[irrationality/ostrowski_1927_mathematische_miszellen/evidence/verify/translated_interval_review|retained review and grade]] identify the exact frozen subject and its unchanged current mathematics. This coverage grants no full Kesten proof, formal verification or community-acceptance finding.

Bears on. Problem 998: (O3) is the direction of the length criterion opposite to the one the corrected statement asks for, giving bounded discrepancy to every translate of an interval of length {jα}\{j\alpha\}, j≠0j\ne0. It does not decide the corrected statement, whose direction is Kesten's necessity.