Infinitely often and almost always: Difference between revisions

From Machinelearning
No edit summary
 
(8 intermediate revisions by the same user not shown)
Line 20: Line 20:
==Analogy with sequences of real numbers==
==Analogy with sequences of real numbers==


Let <math>(a_n)_{n=1}^\infty</math> be a sequence of real numbers, and let <math>\epsilon > 0</math> be a real number.
Let <math>(a_n)_{n=1}^\infty</math> be a sequence of real numbers, let <math>\epsilon > 0</math> be a real number, and let <math>x</math> be a real number.


We say <math>(a_n)_{n=1}^\infty</math> is eventually <math>\epsilon</math>-close to <math>x</math> iff there exists some <math>N \geq 1</math> such that for all <math>n \geq N</math> we have <math>|a_n - x| \leq \epsilon</math>.
We say <math>(a_n)_{n=1}^\infty</math> is eventually <math>\epsilon</math>-close to <math>x</math> iff there exists some <math>N \geq 1</math> such that for all <math>n \geq N</math> we have <math>|a_n - x| \leq \epsilon</math>.


We say that <math>(a_n)_{n=1}^\infty</math> is continually <math>\epsilon</math>-adherent iff for every <math>N \geq 1</math> there exists some <math>n \geq N</math> such that <math>|a_n - x| \leq \epsilon</math>.
We say that <math>(a_n)_{n=1}^\infty</math> is continually <math>\epsilon</math>-adherent iff for every <math>N \geq 1</math> there exists some <math>n \geq N</math> such that <math>|a_n - x| \leq \epsilon</math>.
I think we can even define <math>A_n := \{x \in \mathbf R : |a_n-x| \leq \epsilon\}</math>.
<math>\limsup_{n\to\infty} |a_n-x| = \inf_{N \geq 1} \sup_{n \geq N} |a_n - x| \leq \epsilon</math>
<math>\liminf_{n\to\infty} |a_n-x| = \sup_{N \geq 1} \inf_{n \geq N} |a_n - x| \leq \epsilon</math> -- I think this one is equivalent to infinitely often, which is confusing since now the quantifier order has seemingly switched.
but this makes sense in terms of strength of "infinitely often" vs "almost always". We have <math>\liminf_{n\to\infty} |a_n-x| \leq \limsup_{n\to\infty} |a_n-x|</math>, so if <math>\limsup_{n\to\infty} |a_n-x| \leq \epsilon</math> (i.e. <math>a_n</math> is <math>\epsilon</math>-close to <math>x</math> almost always) then <math>\liminf_{n\to\infty} |a_n-x| \leq \epsilon</math> (i.e. <math>a_n</math> is <math>\epsilon</math>-close to <math>x</math> infinitely often).
{| class="wikitable"
|-
! perspective !! infinitely often !! almost always
|-
| Tao's terminology (see his ''Analysis'') || <math>(a_n)_{n=1}^\infty</math> is continually <math>\epsilon</math>-adherent || <math>(a_n)_{n=1}^\infty</math> is eventually <math>\epsilon</math>-close to <math>x</math>
|-
| first-order quantifier || for every <math>N \geq 1</math> there exists some <math>n \geq N</math> such that <math>|a_n - x| \leq \epsilon</math> || there exists some <math>N \geq 1</math> such that for all <math>n \geq N</math> we have <math>|a_n - x| \leq \epsilon</math>
|-
| || <math>|a_n - x| \leq \epsilon</math> for infinitely many <math>n</math> || <math>|a_n - x| \leq \epsilon</math> for all but finitely many <math>n</math>
|-
| || <math>\liminf_{n\to\infty} |a_n-x| \leq \epsilon</math> || <math>\limsup_{n\to\infty} |a_n-x| \leq \epsilon</math>
|-
| || <math>x \in \limsup_{n\to\infty} A_n</math> || <math>x \in \liminf_{n\to\infty} A_n</math>
|}


[[Category:Probability]]
[[Category:Probability]]

Latest revision as of 22:37, 31 July 2019

Let A1,A2,A3, be a sequence of events in some sample space Ω. Let ωΩ be an outcome.

In the following table, all statements in the "infinitely often" column are logically equivalent. Similarly, all statements in the "almost always" column are logically equivalent.

perspective infinitely often almost always
unions and intersections ωN=1n=NAn ωN=1n=NAn
first-order quantifiers N1nN:ωAn N1nN:ωAn
verbal expression ωAn for infinitely many n1 ωAn for almost all n1, i.e. ωAn for all but finitely many n1, i.e. ωAn for finitely many n1
lim sup/lim inf ωlim supnAn ωlim infnAn
limit of sup/inf ωlimNn=NAn ωlimNn=NAn

Analogy with sequences of real numbers

Let (an)n=1 be a sequence of real numbers, let ϵ>0 be a real number, and let x be a real number.

We say (an)n=1 is eventually ϵ-close to x iff there exists some N1 such that for all nN we have |anx|ϵ.

We say that (an)n=1 is continually ϵ-adherent iff for every N1 there exists some nN such that |anx|ϵ.

I think we can even define An:={xR:|anx|ϵ}.

lim supn|anx|=infN1supnN|anx|ϵ

lim infn|anx|=supN1infnN|anx|ϵ -- I think this one is equivalent to infinitely often, which is confusing since now the quantifier order has seemingly switched.

but this makes sense in terms of strength of "infinitely often" vs "almost always". We have lim infn|anx|lim supn|anx|, so if lim supn|anx|ϵ (i.e. an is ϵ-close to x almost always) then lim infn|anx|ϵ (i.e. an is ϵ-close to x infinitely often).

perspective infinitely often almost always
Tao's terminology (see his Analysis) (an)n=1 is continually ϵ-adherent (an)n=1 is eventually ϵ-close to x
first-order quantifier for every N1 there exists some nN such that |anx|ϵ there exists some N1 such that for all nN we have |anx|ϵ
|anx|ϵ for infinitely many n |anx|ϵ for all but finitely many n
lim infn|anx|ϵ lim supn|anx|ϵ
xlim supnAn xlim infnAn