Skip to main content
added 111 characters in body
Source Link
Yuval Peres
  • 14.2k
  • 1
  • 28
  • 49

For a nice introductory discussion to the maximal ergodic inequality, see [1]. In particular, inequality (5) there is Wiener's maximal ergodic theorem. See also Lemma 15.3 in [2]. A more advanced account, that in particular includes the $L^p$ maximal inequalities, is in the book [3], see Cor 2.2 page 8 for the Wiener inequality and the proof of Theorem 6.3 page 52 for the $L^p$ inequality

$$\lVert\max_{n \le N} \frac{1}{n} \sum_{i \le n} \phi \circ f^i\rVert_{L^p} \le \frac{p}{p-1} \lVert \phi \rVert_{L^p} \quad (*) $$ valid for $p>1$. Note that Theorem 6.3 in [3] only states the limiting inequality when $n$ tends to infinity on both sides of $(*)$, but the proof gives $(*)$ for every finite $n$ as well, since it is a direct application of Cor. 2.2 and Lemma 6.2. For $p=1$ the inequality (*) does not hold, one needs an extra log term on the RHS, see (6.6)on page 52 in [3].

Finally, note that a general inequality of the form proposed, $$\lVert\max_{n \ge N} \frac{1}{n} \sum_{i \le n} \phi \circ f^i\rVert_{L^p} \precsim \lVert\frac{1}{N} \sum_{i \le N} \phi \circ f^i\rVert_{L^p} \quad (**) $$ cannot be valid for any $p$. Consider, for example, $f$ to be a rotation of angle $2\pi/N$$2\pi(\frac{1}{N}-\epsilon)$ on the unit circle, andwhere $\frac{1}{N}-\epsilon $ is irrational. Let $\phi$ be a function that equals 1 on an arc of length $2\pi/N$$\pi$ on that circle, and equals $-1$ on an adjacentthe complementary arc of the same length, and equals $0$ elsewhere. Then for fixed $N>1$, the RHS of $(**)$ equalstends to zero identicallyas $\epsilon \to 0$, while the LHS is at least $1/(N+1)$.

[1] [1] http://www-stat.wharton.upenn.edu/~steele/Courses/530/Resources/HopfSLLN.pdf

[2] https://people.math.wisc.edu/~roch/teaching_files/275b.1.12w/lect15-web.pdf

[3] Krengel, Ulrich. Ergodic theorems. Vol. 6. Walter de Gruyter, 2011.

https://books.google.ca/books?hl=en&lr=&id=t4_BDh8gt2kC&oi=fnd&pg=PA1&ots=5R-dupukqq&sig=yx6aMMlq93oOvD4eLotHFdk6q7U&redir_esc=y#v=onepage&q&f=false

For a nice introductory discussion to the maximal ergodic inequality, see [1]. In particular, inequality (5) there is Wiener's maximal ergodic theorem. See also Lemma 15.3 in [2]. A more advanced account, that in particular includes the $L^p$ maximal inequalities, is in the book [3], see Cor 2.2 page 8 for the Wiener inequality and the proof of Theorem 6.3 page 52 for the $L^p$ inequality

$$\lVert\max_{n \le N} \frac{1}{n} \sum_{i \le n} \phi \circ f^i\rVert_{L^p} \le \frac{p}{p-1} \lVert \phi \rVert_{L^p} \quad (*) $$ valid for $p>1$. Note that Theorem 6.3 in [3] only states the limiting inequality when $n$ tends to infinity on both sides of $(*)$, but the proof gives $(*)$ for every finite $n$ as well, since it is a direct application of Cor. 2.2 and Lemma 6.2. For $p=1$ the inequality (*) does not hold, one needs an extra log term on the RHS, see (6.6)on page 52 in [3].

Finally, note that a general inequality of the form proposed, $$\lVert\max_{n \ge N} \frac{1}{n} \sum_{i \le n} \phi \circ f^i\rVert_{L^p} \precsim \lVert\frac{1}{N} \sum_{i \le N} \phi \circ f^i\rVert_{L^p} \quad (**) $$ cannot be valid for any $p$. Consider, for example, $f$ to be a rotation of angle $2\pi/N$ on the unit circle, and $\phi$ a function that equals 1 on an arc of length $2\pi/N$ on that circle, equals $-1$ on an adjacent arc of the same length, and equals $0$ elsewhere. Then the RHS of $(**)$ equals zero identically.

[1] http://www-stat.wharton.upenn.edu/~steele/Courses/530/Resources/HopfSLLN.pdf

[2] https://people.math.wisc.edu/~roch/teaching_files/275b.1.12w/lect15-web.pdf

[3] Krengel, Ulrich. Ergodic theorems. Vol. 6. Walter de Gruyter, 2011.

https://books.google.ca/books?hl=en&lr=&id=t4_BDh8gt2kC&oi=fnd&pg=PA1&ots=5R-dupukqq&sig=yx6aMMlq93oOvD4eLotHFdk6q7U&redir_esc=y#v=onepage&q&f=false

For a nice introductory discussion to the maximal ergodic inequality, see [1]. In particular, inequality (5) there is Wiener's maximal ergodic theorem. See also Lemma 15.3 in [2]. A more advanced account, that in particular includes the $L^p$ maximal inequalities, is in the book [3], see Cor 2.2 page 8 for the Wiener inequality and the proof of Theorem 6.3 page 52 for the $L^p$ inequality

$$\lVert\max_{n \le N} \frac{1}{n} \sum_{i \le n} \phi \circ f^i\rVert_{L^p} \le \frac{p}{p-1} \lVert \phi \rVert_{L^p} \quad (*) $$ valid for $p>1$. Note that Theorem 6.3 in [3] only states the limiting inequality when $n$ tends to infinity on both sides of $(*)$, but the proof gives $(*)$ for every finite $n$ as well, since it is a direct application of Cor. 2.2 and Lemma 6.2. For $p=1$ the inequality (*) does not hold, one needs an extra log term on the RHS, see (6.6)on page 52 in [3].

Finally, note that a general inequality of the form proposed, $$\lVert\max_{n \ge N} \frac{1}{n} \sum_{i \le n} \phi \circ f^i\rVert_{L^p} \precsim \lVert\frac{1}{N} \sum_{i \le N} \phi \circ f^i\rVert_{L^p} \quad (**) $$ cannot be valid for any $p$. Consider, for example, $f$ to be a rotation of angle $2\pi(\frac{1}{N}-\epsilon)$ on the unit circle, where $\frac{1}{N}-\epsilon $ is irrational. Let $\phi$ be a function that equals 1 on an arc of length $\pi$ on that circle, and equals $-1$ on the complementary arc of the same length. Then for fixed $N>1$, the RHS of $(**)$ tends to zero as $\epsilon \to 0$, while the LHS is at least $1/(N+1)$. [1] http://www-stat.wharton.upenn.edu/~steele/Courses/530/Resources/HopfSLLN.pdf

[2] https://people.math.wisc.edu/~roch/teaching_files/275b.1.12w/lect15-web.pdf

[3] Krengel, Ulrich. Ergodic theorems. Vol. 6. Walter de Gruyter, 2011.

https://books.google.ca/books?hl=en&lr=&id=t4_BDh8gt2kC&oi=fnd&pg=PA1&ots=5R-dupukqq&sig=yx6aMMlq93oOvD4eLotHFdk6q7U&redir_esc=y#v=onepage&q&f=false

Source Link
Yuval Peres
  • 14.2k
  • 1
  • 28
  • 49

For a nice introductory discussion to the maximal ergodic inequality, see [1]. In particular, inequality (5) there is Wiener's maximal ergodic theorem. See also Lemma 15.3 in [2]. A more advanced account, that in particular includes the $L^p$ maximal inequalities, is in the book [3], see Cor 2.2 page 8 for the Wiener inequality and the proof of Theorem 6.3 page 52 for the $L^p$ inequality

$$\lVert\max_{n \le N} \frac{1}{n} \sum_{i \le n} \phi \circ f^i\rVert_{L^p} \le \frac{p}{p-1} \lVert \phi \rVert_{L^p} \quad (*) $$ valid for $p>1$. Note that Theorem 6.3 in [3] only states the limiting inequality when $n$ tends to infinity on both sides of $(*)$, but the proof gives $(*)$ for every finite $n$ as well, since it is a direct application of Cor. 2.2 and Lemma 6.2. For $p=1$ the inequality (*) does not hold, one needs an extra log term on the RHS, see (6.6)on page 52 in [3].

Finally, note that a general inequality of the form proposed, $$\lVert\max_{n \ge N} \frac{1}{n} \sum_{i \le n} \phi \circ f^i\rVert_{L^p} \precsim \lVert\frac{1}{N} \sum_{i \le N} \phi \circ f^i\rVert_{L^p} \quad (**) $$ cannot be valid for any $p$. Consider, for example, $f$ to be a rotation of angle $2\pi/N$ on the unit circle, and $\phi$ a function that equals 1 on an arc of length $2\pi/N$ on that circle, equals $-1$ on an adjacent arc of the same length, and equals $0$ elsewhere. Then the RHS of $(**)$ equals zero identically.

[1] http://www-stat.wharton.upenn.edu/~steele/Courses/530/Resources/HopfSLLN.pdf

[2] https://people.math.wisc.edu/~roch/teaching_files/275b.1.12w/lect15-web.pdf

[3] Krengel, Ulrich. Ergodic theorems. Vol. 6. Walter de Gruyter, 2011.

https://books.google.ca/books?hl=en&lr=&id=t4_BDh8gt2kC&oi=fnd&pg=PA1&ots=5R-dupukqq&sig=yx6aMMlq93oOvD4eLotHFdk6q7U&redir_esc=y#v=onepage&q&f=false