Preprint
Concept Paper

Finding an Unique and “Natural” Extension of the Expected Value That Is Finite for All Functions in Non-Shy Subset of the Set of All Measurable Functions

Altmetrics

Downloads

913

Views

269

Comments

0

This version is not peer-reviewed

Submitted:

09 February 2024

Posted:

12 February 2024

Read the latest preprint version here

Alerts
Abstract
Suppose for n∈ ℕ, set A⊆ℝⁿ and function f∶A→ℝⁿ. If set A is Borel; we want to find an unique and “natural” extension of the expected value, w.r.t the Hausdorff measure on bounded functions to unbounded/bounded f, that's finite for all f in a non-shy subset of B*---the set of all Borel measurable functions in ℝᴬ. The issue is current extensions of the expected value are finite for all functions in only a shy subset of B*. The reason this issue wasn't resolved is mathematicians have not thought of the problem, focusing on application rather than generalization. Despite the lack of potential use, we'll attempt to solve the problem by defining a choice function---this shall choose a unique set of equivalent sequences of sets (Fₖ***) , where the set-theoretic limit of Fₖ*** is the graph of f; the measure Hʰ is the ℎ-Hausdorff measure, such for each k∈ℕ, 0 < Hʰ(Fₖ***) < +∞; and (fₖ*) is a sequence of functions where {(x₁,...,xₙ,fₖ*(x₁,...,xₙ))∶x∈ dom(Fₖ***)}=Fₖ***. Thus, if (Fₖ***) converges to A at a rate linear or super-linear to the rate non-equivelant sequences of sets converge, the extended expected value of or E**[f,Fₖ***] is: ∀(ε>0)∃(N∈ℕ)∀(k∈N)(k≥N⇒1/Hʰ(dom(Fₖ***))∫dom(Fₖ*** )fₖ* dHʰ - E**[f,Fₖ***]<ε ) which should be unique and “natural” extension for all f in a non-shy subset of B*. Note we guessed the choice function using computer programming but we don’t use mathematical proofs due to the lack of expertise in the subject matter. Despite this, the biggest use of this research is the extension of the expected value is finite for "almost all" functions: this is easier to use in application when finding the "average" of functions covering an infinite expanse of space.
Keywords: 
Subject: Computer Science and Mathematics  -   Analysis

1. Introduction

Suppose for n N , set A R n and the function f : A R . If A is Borel, note the expected value of f, w.r.t the Hausdorff-dimensionalHausdorff measure, is undefined when the Hausdorff-dimension (of A)-dimensional Hausdorff measure (of also A) is either + , zero or the function f is unbounded. Infact, for a shy subset of B * —the set of Borel measurable functions in R A —the expected value of f is finite, meaning "almost no" functions have a finite expected value. More specifically, we want to manipulate the expected value of f to be finite, for a non-shy (i.e., prevelant or neither prevelant nor shy[2]) subset of B * .
This can be done by taking the expected value of a sequence of bounded functions which converge to bounded/unbounded f. A sequence of bounded functions is chosen using a "choice function", where criteria are added to determine the choice function. Note, we want the expected value from the chosen sequence of bounded functions to be unique and "natural" extension of the original expected value, taking finite values only.
We do this by defining a sequence of sets called ★-sequence of sets (def. 8). Note the -sequences of sets converge to the graph of f rather than A; otherwise, the generalized expected value of f w.r.t to a -sequence (def. 9) cannot be finite for a non-shy subset of Borel measurable functions. Moreover, since there are graphs of functions with multiple -sequences of sets, s.t. the generalized expected values of f w.r.t each -sequence are different and non-unique (depending on the starred-sequence chosen)—we must have a choice function which chooses a unique set of equivalent -sequences with the same, unique expected value.
Therefore, when defining the choice function, we ask a question in §3.4 where with previous sections; we define equivalent & non-equivalent -sequences of sets for §3.1, and "natural" extensions of expected values for §3.3. We attempt to answer the question in §3.4 by redefining linear/super-linear convergence (def. 16) in terms of entropy, samples and "pathways" where the samples are derived by taking a point from each partition of a -sequence of sets, such the partitions have equal Hausdorff measure—§4. Since all samples have finite points; we take a "pathway" of line segments between the nearest point to each start-point of all segments in the pathway (i.e., the pathway should intersect every point once), where in def. 19 we exclude segments with lengths which are outliers [3]. The procedure is similar to the ones used in computers to graph functions [4]. We also take the length of each of the line segments in the "pathway", multiplying all lengths by a constant so they add up to one (i.e. a discrete probability distribution). We take the supremum of the Entropy of the distribution [5] w.r.t all "pathways" to redefine def. 16 as def. 20, where the redefined definition is used to create a choice function in §5.1.

2. Preliminary Definitons/Motivation

Other than integration with filters [6], there are few other constructive approaches to finding a unique and "natural" extension of the average that takes a finite value for additional functions. Before beginning, consider the following mathematical definitions:

2.1. Preliminary Definitions

Let X be a completely metrizable topological vector space.
Definition 1  
(Prevalent Subset of X) . A Borel set E X is said to be prevalent if there exists a Borel measure μ on X such that:
1. 
0 < μ ( C ) < for some compact subset C of X, and
2. 
the set E + x has full μ-measure (that is, the complement of E + x has measure zero) for all x X .
  • More generally, a subset F of X is prevalent if F contains a prevalent Borel Set. Also note:
Definition 2  
(Shy Subset of X). The complement of a prevalent set is called a shy set.
  • such that we define:
Definition 3  
(Non-Shy Subset of X). A subset of X that is prevalent or neither prevalent nor shy.
Furthermore, suppose we define:
Definition 4  
(Hausdorff Measure). Let ( V , d ) be a metric space, α [ 0 , ) . For every C V , define the diameter of C as:
diam ( C ) : = sup d ( x , y ) : x , y C , diam ( ) : = 0
We define:
H δ α ( E ) = inf i = 1 diam ( C i ) α : diam ( C i ) δ , E i = 1 C i .
The Hausdorff Outer Measure is defined by
H α ( E ) = sup δ > 0 H δ α ( E ) = lim δ 0 H δ α ( E )
If i N and δ R such that δ > 0 , where the Euler’s Gamma function is Γ and constant N α is:
N α = π α / 2 2 Γ α 2 + 1
when α N and E is a Borel set we have that
L α ( E ) = 1 2 N α H α ( E )
such that H α ( E ) is related to the α-dimensional Lebesgue Measure.
Definition 5  
(Hausdorff Dimension). The Hausdorff Dimension of E is defined by dim H ( E ) where:
H α ( E ) = if 0 α < dim H ( E ) 0 if dim H ( E ) < α <
Therefore, we can use definitions 1, 2, 4 to prove or disprove:
Theorem 1.  
The set of Borel measurable, unbounded functions forms a prevalent subset of the set of all Borel measurable functions.
Note 2  
(Notes on Theorem 1). By measurable function, we mean the pre-image of any subset of R (under a measurable function) is in the Borel sigma algebra. (Note function f on set A is unbounded when there is no I 0 such that for all x A ):
f ( x ) I
however, we’re unsure if theorem 1 is correct. Despite this, we could prove or disprove theorem 1 using the paper on prevalence in [2].
If the theorem is true, it is easier to show a shy subset of Borel measurable, unbounded function have finite expected values (see the next definition). Hence, a shy subset of all Borel measurable function, including bounded Borel functions, have finite expected values.
We, therefore, define the expected value w.r.t the Hausdorff measure to be the following:
Definition 6  
(Expected Value of f). If n N , where set A R n , the expected value of function f : A R (using def. 4 and 5) is
E [ f ] = 1 H dim H ( A ) ( A ) A f d H dim H ( A )
where we can see there are cases where E [ f ] is undefined or infinite (e.g. H dim H ( A ) ( A ) is zero, + or f is unbounded). In this case, if topological vector space X is R A (see §2.1) where we define B * such that:
Definition 7  
(The set of all measurable functions). B * is the set of all Borel measurable functions in R A .
Then, we must prove:
Theorem 3.  
The expected value E [ f ] is finite for all f in only a shy subset of B * .
Note 4  
(Note on Theorem 3). We’re not sure how to prove theorem 3; however, we refer to an answer from @Mathe at the last page of this citation [7],
"We can follow the argument presented in example 3.6 of [2]:
Because a function can always be represented as f = f + f we only consider whether positive functions have a mean value. We consider the case of a set A with finite positive measure. In this context having a mean means having a finite integral, and not being integrable means having an infinite integral.
Take X : = L 0 ( A ) (measurable functions over A) let P denote the one-dimensional subspace of L 0 ( A ) consisting of constant functions (assuming the Hausdorff measure on A) and let F : = L 0 ( A ) L 1 ( A ) (measurable functions over A with no finite integral)
If λ P denotes the Lebesgue measure over P, for any fixed f F
λ P β R : A ( f + β ) μ < = 0
Meaning P is a 1-dimensional probe of F, so F is a 1-prevalent set. (In other terms, the set of measurable functions over A with no finite integral or mean, forms a prevalent subset of the set of all measurable functions in R A . Therefore, using def. 2, the set of measurable functions with a finite integral or mean forms a shy subset of all Borel measurable functions in R A .)

2.2. Extended Expected Values

Four solutions to getting a finite expected value for the superset of the subset of B * , where E [ f ] is finite (def. 6), follows:
1.
Defining a dimension function ; i.e., h : [ 0 , + ) [ 0 , + ] , that’s monotonically increasing, strictly positive and right continuous, such that when R denotes the radius of a ball in a covering for the definition of the Hausdorff Measure, we replace R dim H ( A ) with h ( R ) so H h ( A ) : the h-Hausdorff measure , is positive and finite. This leads to the extended expected value E [ f ] , where:
E [ f ] = 1 H h ( A ) A f d H h
Note, however, not all A has dimension function h which leads to:
2.
If A is fractal but has no gauge function, we could use this paper [8] which is an extension of the Lebesgue density theorem and this paper [9] which is an extension of the Hausdorff measure using Hyperbolic Cantor sets. Note, however, when A is non-fractal (e.g. countably infinite) or f is unbounded, there is a possibility that the expected value is infinite or undefined. Hence,
3.
In the case f is unbounded and fractal, we could use [10, p.19-47], which applies a Henstock-Kurzweil type integral (i.e., μ -HK integral) on a measure Metric Space. This coincides with unbounded functions with finite improper Riemman integrals, including bounded functions with finite Lebesgue integrals, bounded function with finite integrals w.r.t the Hausdorff measure, or function with finite Henstock-Kurzweil integrals.

2.3. Examples

  • If n N , set A R n and function f : A R , we want to apply the definitions of the next section for the following examples:
(a)
A = R { 0 } and f ( x ) = 1 / x . This function is unbounded and has an undefined expected value since the average of 1 / x , using the improper Riemann integral on R { 0 } :
lim ( x 1 , x 2 , x 3 , x 4 ) ( , 0 , 0 + , + ) 1 ( x 4 x 3 ) + ( x 2 x 1 ) x 1 x 2 1 x d x + x 3 x 4 1 x d x =
lim ( x 1 , x 2 , x 3 , x 4 ) ( , 0 , 0 + , + ) 1 ( x 4 x 3 ) + ( x 2 x 1 ) ln ( | x | ) + C | x 1 x 2 + ln ( | x | ) + C | x 3 x 4 =
lim ( x 1 , x 2 , x 3 , x 4 ) ( , 0 , 0 + , + ) 1 ( x 4 x 3 ) + ( x 2 x 1 ) ln ( | x 2 | ) ln ( | x 1 | ) + ln ( | x 4 | ) ln ( | x 3 | )
is + (when x 2 = 1 / x 1 , x 3 = 1 / x 4 , and x 1 = exp x 4 2 ) or (when x 2 = 1 / x 1 , x 3 = 1 / x 4 , and x 4 = exp x 1 2 ), making the average undefined.
(b)
A = Q , gcd is the greatest common divisor, and f 1 , f 2 : R R where:
f ( x ) = f 1 ( x ) x A 1 : = r / q : r odd Z , q even Z , q 0 , gcd ( r , q ) = 1 f 2 ( x ) x A 2 : = r 1 / ( q 1 ) : r 1 Z , q 1 odd Z , gcd ( r 1 , q 1 ) = 1
For instance, point ( 1 / 4 , f 1 ( 1 / 4 ) ) is a point in the graph of f (since 1 / 4 Q and 1 / 4 A 1 , making f ( 1 / 4 ) = f 1 ( 1 / 4 ) ). Also, point ( 1 / 3 , f 2 ( 1 / 3 ) ) is a point in the graph of f (since 1 / 3 Q and 1 / 3 A 2 , making f ( 1 / 3 ) = f 2 ( 1 / 3 ) ); however, point ( 2 , 1 ) is not in the graph of f (since 2 Q ).
Note the function in eq. 8 is bounded; however, the expected value & extensions are undefined. (Using def. 6, we know dim H ( A ) = 0 but H dim H ( A ) ( A ) = + , which makes E [ f ] :
E [ f ] = 1 H dim H ( A ) ( A ) A f d H dim H ( A )
undefined by division of + .) Further, we assume using §2.2, crit. 1, there is no (exact) dimension function of A nor could A be "fractal" enough for extensions of the Lebesgue Density Theorem [8], extensions of the Hausdorff measure using Hyperbolic Cantor Sets [9], or extension of the Henstock-Kurzweil integral on the Metric Space [10, p.19-47].

3. Attempt to Answer Thesis

  • Suppose for n N , set A R n and function f : A R . Moreover, H h is the h-Hausdorff measure (§2.2, crit. 1) where h is the dimension function, and B * is the set of all Borel measurable functions in R A .
Definition 8  
(-Sequence of Sets). When we define a sequence of sets ( F r ) r N , where h is the dimension function (§2.2, crit. 1), then if:
(a) 
The set theoretic limit of ( F r ) r N is the graph of f (i.e., ( F r ) r N converges to the graph of f) such that
lim sup r F r = r 1 q r F q lim inf r F r = r 1 q r F q
with the graph of f being:
( x 1 , · · · , x n , f ( x 1 , · · · , x n ) ) : ( x 1 , · · · , x n ) A
such that the set-theoretic limit of ( F r ) r N should be:
lim sup r F r = lim inf r F r = ( x 1 , · · · , x n , f ( x 1 , · · · , x n ) ) : ( x 1 , · · · , x n ) A
(b) 
For all r N , where H h is the h-Hausdorff measure (§2.2, crit. 1):
0 < H h ( F r ) < +
(c) 
we define sequence of functions f r r N where f r : dom ( F r ) range ( F r ) such that:
( x 1 , · · · , x n , f r ( x 1 , · · · , x n ) ) : ( x 1 , · · · , x n ) dom ( F r ) = F r
we have ( F r ) is a ★-sequence of sets or starred-sequence of sets .
Example 1.  
One ★-sequence of sets of f ( x ) = 1 / x on R { 0 } (§2.3, crit. 3a) is:
( F r ) r N = ( x , 1 / x ) : x [ r , 1 / r ] [ 1 / r , r ] r N
Example 2.  
Another example of a ★-sequence of sets of f : Q R where:
f ( x ) = 1 x A 1 : = r / q : r odd Z , q even N , q 0 , gcd ( r , q ) = 1 0 x A 2 : = r 1 / ( q 1 ) : r 1 Z , q 1 odd N , gcd ( r 1 , q 1 ) = 1
using (§2.3, crit. 3b) is the following:
F r r N = x , f ( x ) : x c / ( r ! ) : r · r ! c r · r ! r N
another example is:
F r r N = x , f ( x ) : x c / d : d r , d · r c d · r r N
Note this leads to a new extension of the expected value where when there exists at least one starred-sequence of sets where the extended expected value is finite, the extension could be finite for all f in a non-shy subset of B * (def. 7).
Definition 9  
(Generalized Expected Value). If ( F r ) r N is a ★-sequence of sets (def. 8), the generalized expected value of f w.r.t ( F r ) r N is E * * [ f , F r ] (when it exists) where:
( ϵ > 0 ) ( N N ) ( r N ) r N 1 H h dom F r dom ( F r ) f r d H h E * * [ f , F r ] < ϵ
Example 3.  
Using example 1, we find that when ( F r ) r N = ( x , 1 / x ) : x [ r , 1 / r ] [ 1 / r , r ] r N :
(a) 
dom ( F r ) = [ r , 1 / r ] [ 1 / r , r ] r N
(b) 
f r ( x ) = 1 / x for x [ r , 1 / r ] [ 1 / r , r ]
and the generalized expected value is:
lim ( x 1 , x 2 , x 3 , x 4 ) ( , 0 , 0 + , + ) 1 ( x 4 x 3 ) + ( x 2 x 1 ) x 1 x 2 1 x d x + x 3 x 4 1 x d x =
lim r 1 ( r 1 / r ) + ( 1 / r ( r ) ) r 1 / r 1 x d x + 1 / r r 1 x d x =
lim r 1 ( r 1 / r ) + ( 1 / r + r ) ln ( | x | ) + C | r 1 / r + ln ( | x | ) + C | 1 / r r =
lim r 1 ( r 1 / r ) + ( 1 / r + r ) ln ( | r | ) ln ( | 1 / r | ) + ln ( | r | ) ln ( | 1 / r | ) =
lim r 1 2 r 2 / r · 4 ln ( r ) =
0
We can see from example 2.3 crit. 3a, the average was once undefined but now we’ve "chosen" a ★-sequence which gives a finite expected value.

3.1. Equivalent and Non-Equivalent -sequences of Sets

  • Suppose we define the following:
Definition 10  
(Set V ). Set V is the set of all f, where the generalized expected value—w.r.t at least one starred sequence—exists.
  • The following are definitions of equivelant and non-equivelant starred-sequences of sets:
Definition 11  
(Non-Equivalent Starred-Sequences of Sets).  All starred-sequences of sets (in a set of ★-sequences of sets) are non-equivalent, if there exists an f V (def. 10), where the generalized expected values of f (def. 9) w.r.t each starred-sequence of sets has two or more different values (e.g., defined and undefined values are different). See Figure 1.
Definition 12  
(Equivalent Starred-Sequences of Sets). All starred-sequences of sets (in the set of ★-sequences of sets) are equivalent, if we get for all  f V (def. 10); the generalized expected value of f (def. 9) w.r.t each starred-sequence of sets has the same value. See Figure 2.
  • However, proving that two or more starred-sequences of sets are non-equivalent or equivalent (using def. 12 or 11) is tedious, since we constantly compute def. 9. Therefore, we ask the following:

3.1.1. Question 1

  • Is there are a simpler definition of equivalent and non-equivalent -sequences of sets.

3.1.2. Possible Answer

  • For the sake of brevity, suppose starred-sequences F r 1 ( 1 ) = F r 1 (i.e., def. 8), such that F r 2 ( 2 ) = F r 2 , F r 3 ( 3 ) = F r 3 , and F r s ( s ) = F r s - s times
Definition 13  
(Equivalent Starred-Sequences of Sets [Revisited]). Starred-sequence of sets ( F r 1 ) r 1 N and ( F r 2 ) r 2 N are equivalent, if there exists a N N , where for all r 1 N , there exists a r 2 N , where if h 1 is the dimension function (§2.2, crit. 1) of F r 1 ,
H h 1 ( F r 1 Δ F r 2 ) = 0
and also for all r 2 N , there exists a r 1 N , where if h 2 is the dimension function of F r 2 then:
H h 2 ( F r 1 Δ F r 2 ) = 0
Note we denote these equivalent starred-sequence of sets as
( F r 1 ) r 1 N ( F r 2 ) r 2 N
Definition 14  
(Multiple Equivalent Starred-Sequences of Sets [Revisited]). All starred-sequences of sets in:
( F r 1 ) r 1 N , ( F r 2 ) r 1 N , · · · , ( F r j ( j ) ) r 1 N
are equivalent, if for all k , v 1 , · · · , j where k v , ( F r k ( k ) ) r k N and ( F r v ( v ) ) r v N are equivelant (def. 13). We also state the former as:
( F r k ( k ) ) r k N ( F r v ( v ) ) r v N
Theorem 5.  
If starred-sequences of sets in:
( F r 1 ) r 1 N , ( F r 2 ) r 1 N , · · · , ( F r j ( j ) ) r 1 N
are equivalent (def. 14), then for all k , v 1 , · · · , j where k v , the generalized means of f w.r.t the ★-sequences (def. 9) have the same mean value. In other words:
E * * [ f , F r k ( k ) ] = E * * [ f , F r v ( v ) ]
Definition 15  
(Non-Equivalent Starred-Sequences of Sets [Revisited]). All starred-sequences of sets in ( F r s ( s ) ) r s N : s N are non-equivalent, if def. 14 is false.

3.2. Motivation For Sec. 3.4

  • For all f in a non-shy subset of B * , we could choose a -sequence of sets ( F r ) r N where the generalized expected value of f, w.r.t the chosen starred-sequence, is finite. However, consider the following problem:
Theorem 6.  
The set of all f, where the generalized expected values of f w.r.t two or more non-equivalent ★-sequences of sets has different values, form a non-shy subset of all Borel measurable functions in R A .
  • This means "almost all" measurable functions have several generalized expected values depending on the starred-sequence chosen. Therefore, we need to choose a unique -sequence of sets where the new extended expected value is an "natural" extension of the original expected value.

3.3. Essential Definitions for a "Natural" Expected Value

  • Suppose ( F r ) r N and ( F j ) j N are non-equivelant starred-sequences of sets (def. 8 & 15): we have the following is essential for a "natural" extension of the expected value.
Definition 16  
(Linear & Super-linear Convergence of a -Sequence of Sets To That Of Another -Sequence of Sets). If we define function S : R R , where r N and for any linear j 1 : N N , where j = j 1 ( r ) , O is the Big-O notation, and:
H h ( F r ) = O ( S ( H h ( F j ) ) )
where if the following is true:
0 < lim x S ( x ) / x
then ( F r ) r N converges to the graph of f: i.e.,
( x 1 , · · · , x n , f ( x 1 , · · · , x n ) ) : ( x 1 , · · · , x n ) A
at a linear or super-linear rate compared to that of ( F j ) j N .
Now we may combine the previous definitions into a main question with an answer that solves the thesis 1.

3.4. Main Question

  • Does there exist a choice function that chooses a unique set (of equivalent -sequences of sets—def. 14) such that:
(a)
The chosen starred-sequences of sets converge to ( x 1 , · · · , x n , f ( x 1 , · · · , x n ) ) : ( x 1 , · · · , x n ) A at a rate linear or super-linear (def. 16) to the rate non-equivalent -sequences of sets (def. 15) converge to ( x 1 , · · · , x n , f ( x 1 , · · · , x n ) ) : ( x 1 , · · · , x n ) A
(b)
The generalized expected value (def. 9) of f w.r.t the chosen (and equivalent) starred-sequences of sets (def. 14) is finite.
(c)
The choice function chooses a unique set of equivalent -sequences of sets which satisfy (1) and (2), for all f Q such that Q is a non-shy subset (def. 5) of B * (i.e., the set of all Borel measurable functions in R A ).
(d)
Out of all the choice functions which satisfy (1), (2) and (3), we choose the one with the simplest form, meaning for each choice function fully expanded, we take the one with the fewest variables/numbers (excluding those with quantifiers)?
Note 7  
(Notes On Question). Note, the unique set of equivalent and chosen starred-sequences of sets is defined using notation ( F k ) k N , where ( F k ) k N is a starred-sequence in ( F k ) k N . Therefore, after we define the choice function, the answer should be E * * [ f , F k ] —using def. 9 (when it exists):
( ϵ > 0 ) ( N N ) ( k N ) k N 1 H h dom F k dom ( F k ) f k d H h E * * [ f , F k ] < ϵ
Also, consider the following: if the solution to the main question is extraneous, what other criteria can be included to get a unique choice function? (Note if the solution is always extraneous, we want to replace “equivelant starred-sequences of sets” with the following: ”the set of all ★-sequences of sets, where the generalized expected values of f w.r.t each starred-sequence is the same”.)

4. Solution To The Main Question Of Section 2.4

  • Suppose h is the dimension function, H h is the h-Hausdorff measure (§2.2, crit. 1), and ( F r ) r N is the starred-sequence of sets (def. 8). We will use an alternative approach to definition 16 so we can define a choice function which solves the main question. Read from the second sentence of the last paragraph of the intro of §1 for a summary. Also, refer to sec. 3 and 4 of [1] for examples: (the cited paper uses sets instead of functions).

4.1. Preliminary Definitions

Definition 17  
(Uniform ε coverings of each term of a -sequence of sets). We define uniform ε coverings of each term of ( F r ) r N as a group of pair-wise disjoint sets which cover F r (for some r N ), such when taking dimension function h of F r , we want H h of each pair-wise disjoint set to have the same value ε range ( H h ) , where ε > 0 and the total sum of H h of the coverings is minimized. In shorter notation, if
  • The element t N
  • The set T N is arbitrary and uncountable.
and set Ω is defined as:
Ω = 1 , · · · , t if there are t ways of writing uniform ε coverings of F r N if there are countably infinite ways of writing uniform ε coverings of F r T if there are uncountable ways of writing uniform ε coverings of F r
then for every ω Ω , the set of uniform ε coverings is defined using U ( ε , F r , ω ) where ω “enumerates" all possible uniform ε coverings of F r for every r N .
Definition 18  
(Sample of the uniform  ε  coverings of each term of a -sequence of sets). The sample of uniform ε coverings of each term of ( F r ) r N is the set of points where for every ε range ( H h ) and r N , we take a point from each pair-wise disjoint set in the uniform ε coverings of F r (def. 17). In shorter notation, if
  • The element k N
  • The set K N is arbitrary and uncountable.
and set Ψ ω is defined as:
Ψ ω = 1 , · · · , k if there are k ways of writing the sample of uniform ε coverings of F r N if there are countably infinite ways of writing the sample of uniform ε coverings of F r K if there are uncountable ways of writing the sample of uniform ε coverings of F r
then for every ψ Ψ ω , the set of all samples of the set of uniform ε coverings is defined using S ( U ( ε , F r , ω ) , ψ ) , such that ψ “enumerates" all possible samples of U ( ε , F r , ω ) .
Definition 19  
(Entropy on the sample of uniform coverings of each term of -sequence of sets). Since there are finitely many points in the sample of the uniform ε coverings of each term of ( F r ) r N (def. 18), we:
(a) 
Take a "pathway" of line segments between all points in each sample (def. 18), such that if we define the following:
i. 
· is the ceiling function
ii. 
d ( Q , R ) is the Euclidean-distance between points Q R n and R R n
iii. 
The sequence:
{ x i 1 } i = 1 H h ( F r ) / ε 1
contains all points in the "original" sample S ( U ( ε , F r , ω ) , ψ ) where we define a "pathway" for which we:
A.
Choose a point x 0 S ( U ( ε , F r , ω ) , ψ )
B.
Take a point from S ( U ( ε , F r , ω ) , ψ ) (excluding x 0 ) with smallest euclidean distance from point x 0 S ( U ( ε , F r , ω ) , ψ ) . We denote this point x 1 where we take d ( x 0 , x 1 ) . (If more than one point has the smallest Euclidean distance from x 0 , we take either point).
C.
Take a point in S ( U ( ε , F r , ω ) , ψ ) (excluding x 0 and x 1 ) with smallest euclidean distance from x 1 . We denote this point x 2 , where we take d ( x 1 , x 2 ) . (If more than one point has the smallest Euclidean distance from x 1 , we take either point).
D.
Take a point in S ( U ( ε , F r , ω ) , ψ ) (excluding x 0 , x 1 , and x 2 ) with smallest euclidean distance from x 2 . We denote this point x 3 then take d ( x 2 , x 3 ) . (If more than one point has the smallest Euclidean distance from x 2 , we take either point).
E.
Repeat the process excluding points x 0 , x 1 , x 2 , x 3 , etc. until all points in the sample are "denoted". (This should occur H h ( F r ) / ε 1 times.)
iv. 
V is a subset of i N : 1 i H h ( F r * ) / ε 1 with the largest cardinality, where that we take the subset of i-values where x i has the r i -th smallest Euclidean distance from x i 1 (compared to every point in S ( U ( ε , F r , ω ) , ψ ) x i 1 ) such that r i is not an outlier [3] of
r t : t N , 1 t H h ( F r ) / ε 1
In other words:
A.
For all w V , we want V to be the largest subset of i N : 1 i H h ( F r * ) / ε 1 for which w-values are all i-values satisfying criteria 3(a)iv.
iv. 
Combining everything in def. 19, crit. 3a, we ultimately want all lengths between every point in the "pathway" (def. 18) satisfying crit. 3(a)iv. We call this:
D x 0 , { x w 1 } w V , S ( U ( ε , F r , ω ) , ψ ) = d ( x w , x w 1 ) : w V
(b) 
Using def. 19, crit. 3(a)v, normalize D into a discrete probability distribution. This is defined as:
P D x 0 , { x w 1 } w V , S ( U ( ε , F r , ω ) , ψ ) = y / z D x 0 , { x w 1 } w V , S ( U ( ε , F r , ω ) , ψ ) z : y D x 0 , { x w 1 } w V , S ( U ( ϵ , F r , ω ) , ψ )
(c) 
Take the entropy of (2), (for further reading, see [5]). This is defined as:
E D x 0 , { x w 1 } w V , S ( U ( ε , F r , ω ) , ψ ) = x P D x 0 , { x w 1 } w V , S ( U ( ε , F r , ω ) , ψ ) x log 2 x
(d) 
Take x 0 S ( U ( ε , F r , ω ) , ψ ) ) where E D x 0 , { x w 1 } w V , S ( U ( ε , F r , ω ) , ψ ) is maximized. Call this, E ( D ( S ( U ( ε , F r , ω ) , ψ ) ) ) where:
E ( D ( S ( U ( ε , F r , ω ) , ψ ) ) ) = sup x 0 S ( U ( ε , F r , ω ) , ψ ) E D x 0 , { x w 1 } w V , S ( U ( ε , F r , ω ) , ψ )
with eq. 24 the entropy of the sample of uniform ε coverings of F r .
Definition 20  
(Starred-Sequence of sets converging Sublinearly, Linearly, or Superlinearly to A compared to that of another -Sequence).Suppose we define starred-sequences of sets ( F r ) r N and ( F j ) j N , where for a constant ε range ( H h ) greater than zero and variable r N , we say:
  • (a) 
    Using def. 18 and 19, suppose we have:
    | S ( U ( ε , F r , ω ) , ψ ) | ̲ = sup | S ( U ( ε , F j , ω ) , ψ ) | : j N , ω Ω , ψ Ψ ω , E ( D ( S ( U ( ε , F j , ω ) , ψ ) ) ) E ( D ( S ( U ( ε , F r , ω ) , ψ ) ) )
    then (using  | S ( U ( ϵ , F r , ω ) , ψ ) | ̲   ) we get
    α ¯ ε , r , ω , ψ = S ( U ( ε , F r , ω ) , ψ ) ) / sup ω Ω sup ψ Ψ ω S ( U ( ε , F r , ω ) , ψ ) ̲
    (b) 
    From def. 18 and 19, suppose we have:
    | S ( U ( ε , F r , ω ) , ψ ) | ¯ = inf | S ( U ( ε , F j , ω ) , ψ ) | : j N , ω Ω , ψ Ψ ω , E ( D ( S ( U ( ε , F j , ω ) , ψ ) ) ) E ( D ( S ( U ( ε , F r , ω ) , ψ ) ) )
    then (using | S ( U ( ε , F r , ω ) , ψ ) | ¯ ) we have:
    α ̲ ε , r , ω , ψ = S ( U ( ε , F r , ω ) , ψ ) / sup ω Ω sup ψ Ψ ω S ( U ( ε , F r , ω ) , ψ ) ¯
  • If using α ¯ ϵ , r , ω , ψ and α ̲ ϵ , r , ω , ψ we have that:
    sup ω Ω sup ψ Ψ ω lim sup ε 0 lim sup r α ¯ ε , r , ω , ψ = inf ω Ω inf ψ Ψ ω lim inf ε 0 lim inf r α ̲ ε , r , ω , ψ = 0
    we say F r r N converges to A at a rate superlinear to that of ( F j ) j N .
  • If using equations α ¯ ε , j , ω , ψ and α ̲ ε , j , ω , ψ (where we swap ( F r ) r N in α ¯ ϵ , r , ω , ψ and α ̲ ϵ , r , ω , ψ with ( F j ) j N ) we have that:
    sup ω Ω sup ψ Ψ ω lim sup ε 0 lim sup j α ¯ ε , j , ω , ψ = inf ω Ω inf ψ Ψ ω lim inf ε 0 lim inf j α ̲ ε , j , ω , ψ = 0
    we then say ( F r ) r N converges to A at a rate sublinear to that of ( F j ) j N .
  • If using equations α ¯ ε , r , ω , ψ , α ̲ ε , r , ω , ψ , α ¯ ε , j , ω , ψ , and α ̲ ε , j , ω , ψ (such for the two latter, we swap ( F r ) r N in α ¯ ε , r , ω , ψ and α ̲ ε , r , ω , ψ with ( F j ) j N ) we have both :
    (a) 
    sup ω Ω sup ψ Ψ ω lim sup ε 0 lim sup r α ¯ ε , r , ω , ψ or inf ω Ω inf ψ Ψ ω lim inf ε 0 lim inf r α ̲ ε , r , ω , ψ does not equal zero
    (b) 
    sup ω Ω sup ψ Ψ ω lim sup ε 0 lim sup j α ¯ ε , j , ω , ψ or inf ω Ω inf ψ Ψ ω lim inf ε 0 lim inf j α ̲ ε , j , ω , ψ does not equal zero
and say ( F r ) r N converges to A at a rate linear to that of ( F j ) j N .

5. Attempt to Answer Main Question Of Section 2.4

5.1. Choice Function

Suppose we define the following:
1.
( F k ) k N is a starred-sequence of sets (def. 8) which satisfies (1), (2), and (3) of the main question in §3.4
2.
S ( G ) , where G is the graph of f; i.e.,
G = ( x 1 , · · · , x n , f ( x 1 , · · · , x n ) ) : ( x 1 , · · · , x n ) A
is the set of the starred-sequences of sets that have finite generalized mean (def. 9).
3.
( F j ) j N is an element S ( G ) but not an element in the set of equivalent starred-sequences of sets (def. 14) of ( F k ) k N where using note 7, we can represent this criteria as:
( F j ) j N S ( G ) ( F k ) k N
Further note, from def. 20, if we take:
| S ( U ( ε , F k , ω ) , ψ ) | ¯ = inf | S ( U ( ε , F j , ω ) , ψ ) | : j N , ω Ω , ψ Ψ ω , E ( D ( S ( U ( ε , F j , ω ) , ψ ) ) ) E ( D ( S ( U ( ε , F k , ω ) , ψ ) ) )
and from def. 20, we take:
| S ( U ( ε , F k , ω ) , ψ ) | ̲ = sup | S ( U ( ε , F j , ω ) , ψ ) | : j N , ω Ω , ψ Ψ ω , E ( D ( S ( U ( ε , F j , ω ) , ψ ) ) ) E ( D ( S ( U ( ε , F k , ω ) , ψ ) ) )
Then, when we write def. 18, eq. 30 and eq. 31 as:
sup ω Ω sup ψ Ψ ω | S ( U ( ε , F k , ω ) , ψ ) | = | S ( ε , F k ) | = | S |
sup ω Ω sup ψ Ψ ω | S ( U ( ε , F k , ω ) , ψ ) | ¯ = | S ( ε , F k ) | ¯ = | S | ¯
sup ω Ω sup ψ Ψ ω | S ( U ( ε , F k , ω ) , ψ ) | ̲ = | S ( ε , F k ) | ̲ = | S | ̲
the choice function (which we’ll later define on pg. 16, thm. 8) should immediately choose F k when:
1.
For all m 1 , · · · , n when defining the set of all values of the m-th coordinate of ( c 1 , c 2 , · · · , c n + 1 ) F k (i.e., F k , m —where, unlike cit.[1, §4], we focus on the domain of F k to get "n" instead of " n + 1 "), then when z > 0 , we either want:
(a) 
sup ( F k + 1 , m ) sup ( F k , m ) = z and inf ( F k + 1 , m ) inf ( F k , m ) = z .
(b) 
sup ( F k + 1 , m ) sup ( F k , m ) = 0 and inf ( F k + 1 , m ) inf ( F k , m ) = z .
(c) 
sup ( F k + 1 , m ) sup ( F k , m ) = z and inf ( F k + 1 , m ) inf ( F k , m ) = 0 .
(d) 
sup ( F k + 1 , m ) sup ( F k , m ) = 0 and inf ( F k + 1 , m ) inf ( F k , m ) = 0 .
2.
If the center of the universe is a chosen point Z R n + 1 , where:
Z = ( z 1 , z 2 , · · · , z n + 1 )
then for all m 1 , · · · , n , there exists q N , s.t.for all k q , when set F k , m is a collection of all the values of the m-th co-ordinate of ( c 1 , c 2 , · · · , c n + 1 ) F k , such that x 1 F k , m (again, unlike cit.[1, §4], we focus on the domain of F k to get "n" instead of " n + 1 "), we must get:
1 H h ( F k , m ) F k , m x 1 d H h = z m
where, using absolute value function | | · | | and m 1 , 2 , · · · , n , when set F k , m is a collection of all the values of the m-th co-ordinate of ( c 1 , c 2 , · · · , c n + 1 ) F k , for z > 0 , when we define:
S ( z , k , m ) = | | z sup F k + 1 , m sup F k , m inf F k , m inf F k + 1 , m | | inf F k , m inf F k + 1 , m sup F k + 1 , m sup F k , m 1 | | | |
and
T ( z m , k , m ) = sup F k + 1 , m z m inf F k , m z m sup F k , m z m inf F k + 1 , m z m inf F k , m z m inf F k + 1 , m z m + sup F k + 1 , m z m sup F k , m z m 1 inf F k , m z m inf F k + 1 , m z m sup F k + 1 , m z m sup F k , m z m
criteria (1) is achieved, using eq. 37, when:
S ( z , k ) = 1 n m = 1 n S ( z , k , m )
such that, for all k N :
S ( z , k ) = 1
and criteria (2) is achieved, using eq. 35 and 38, when:
T ( Z , k ) = 1 n m = 1 n T ( z m , k , m )
such that, for all k N :
T ( Z , k ) = 0
where we consider the following:

5.2. Question:

How do we create a choice function which solves the question in sec. 3.4 using S , | S | ¯ , | S | ̲ , S ( z , k ) , and T ( Z , k ) or equations 32, 33, 34, 39 and 41 resp.?

5.3. "Attempt" to answer the Question

(Note the attempt might be wrong but could offer hints to how the solution would appear).
Suppose z = 1 and the chosen coordinate for the center of the universe (i.e., eq. 35) is the origin, where z m = 0 for all m 1 , · · · , n :
Z = ( z 1 , z 2 , · · · , z n + 1 ) Z = O = ( 0 , 0 , · · · , 0 n + 1 times )
Using equations S , | S | ¯ , | S | ̲ , S ( z , k ) , and T ( Z , k ) (i.e., eq. 32, 33, 34, 39 and 41) with the absolute value function | | · | | and the nearest integer function · , we define:
K ( ε , F k ) = S ( 1 , k ) 5 ( 5 | 5 | | S | 1 + | S | | S | ̲ + 2 | S | | S | ̲ + | S | | S | ̲ + | S | + | S | ¯ 1 + | S | ̲ / | S | 1 + | S | / | S | ¯ 1 + | S | ̲ / | S | ¯ | S | 5 | 5 | + | S | 5 ) T ( O , k )
where using K ( ε , F k ) , the choice function should be the following:
Theorem 8.  
If we define:
M ( ε , F k ) = | S ( ε , F k ) | K ( ε , F k ) | S ( ε , F k ) |
M ( ε , F j ) = | S ( ε , F j ) | K ( ε , F j ) | S ( ε , F j ) |
where for M ( ε , F k ) , we define M ( ε , F k ) to be the same as M ( ε , F j ) when swapping " j N " with " k N " (for eq. 33 & 34) and sets F k with F j (for eq. 3244), then for constant v > 0 and variable v * > 0 , if:
S ¯ ( ε , k , v * , F j ) = inf | S ( ε , F j ) | : j N , M ( ε , F j ) M ( ε , F k ) v * { v * } + v
and:
S ̲ ( ε , k , v * , F j ) = sup | S ( ε , F j ) | : j N , v * M ( ε , F j ) M ( ε , F k ) { v * } + v
then for all ( F j ) j N S ( G ) ( F k ) k N (§5.1, crit. 3), if:
lim inf ε 0 lim v * lim k | S ( ε , F k ) | + v S ¯ ( ε , k , v * , F j ) = lim sup ε 0 lim v * lim k | S ( ε , F k ) | + v S ̲ ( ε , k , v * , F j ) = 0
we choose ( F k ) k N satisfying eq. 47. (Note, we want sup = , inf = + , and ( F k ) k N to answer the main question of §3.4) where the answer to the focus2 is E * * [ f , F k ] in eq. 48—using def. 9 (when it exists):
( ϵ > 0 ) ( N N ) ( k N ) k N 1 H h dom F k dom ( F k ) f k d H h E * * [ f , F k ] < ϵ
Note 9  
(Explanation of Theorem 8) . The theorem 8 is similar to the methods used in def. 20 crit. 0a and 0b— α ¯ ε , r , ω , ψ and α ̲ ε , r , ω , ψ —and def. 20 crit. 1 and crit. 3—linear/superlinear convergence—where:
sup ω Ω sup ψ Ψ ω lim sup ε 0 lim sup r α ¯ ε , r , ω , ψ = inf ω Ω inf ψ Ψ ω lim inf ε 0 lim inf r α ̲ ε , r , ω , ψ = 0
such that we replace:
E ( D ( S ( U ( ε , F r , ω ) , ψ ) ) ) M ( ε , F k ) E ( D ( S ( U ( ε , F j , ω ) , ψ ) ) ) M ( ε , F j ) S ( U ( ε , F r , ω ) , ψ ) ) | S ( ε , F j ) | sup ω Ω sup ψ Ψ ω S ( U ( ε , F r , ω ) , ψ ) ̲ S ̲ ( ε , k , v * , F j ) sup ω Ω sup ψ Ψ ω S ( U ( ε , F r , ω ) , ψ ) ¯ S ¯ ( ε , k , v * , F j )
note the changes to def. 20, crit. 1 were made, so M ( ε , F k ) is "large enough" compared to M ( ε , F j ) , with ( F j ) j N non-equivalent to ( F k ) k N (e.g., when A = Q , ( F k ) k N should be ( c / k ! : c N , 1 c k ! ) k N and never give M ( ε , F k ) which increases at a smaller rate than that of "small" M ( ε , F j ) , e.g.:
( F j ) j N = u / w : u Z , w N , w j , w · j u w · j j N
or smaller than that of "large" M ( ε , F j ) ; e.g.,
( F j ) j N = ( u 1 / ( 6 ( j ! ) ) : u 1 Z , 6 j · j ! u 1 6 j · j ! ) j N
Moreover, in S ̲ ( ε , k , v * , F j ) and S ¯ ( ε , k , v * , F j ) of thm. 8, we add constant v > 0 and variable v * > 0 so if either
1. 
S ̲ ( ε , k , v * , F j ) v = 0 (i.e., using a related limit to eq. 47, division by zero is undefined).
2. 
S ¯ ( ε , k , v * , F j ) v = 0 (i.e., using a related limit to eq. 47, division by zero is undefined).
3. 
inf | S ( ε , F j ) | : j N , M ( ε , F j ) M ( ε , F k ) = + (i.e., similar to S ̲ ( ε , k , v * , F j ) of eq. 45, with no variable v * such that M ( ε , F k ) = 0 and ( J > 0 ) ( j 1 > 0 ) ( j j 1 ) ( M ( ε , F j ) J ) , where we apply a related limit to eq. 47 that’s undefined due to division by infinity.)
4. 
inf | S ( ε , F j ) | : j N , M ( ε , F j ) M ( ε , F k ) = (i.e., similar to S ̲ ( ε , k , v * , F j ) of eq. 45, with no variable v * and M ( ε , F j ) = 0 , where we apply a related limit to eq. 47 that’s undefined since inf | S ( ε , F j ) | : j N , M ( ε , F j ) M ( ε , F k ) is an undefined empty set.)
5. 
sup | S ( ε , F j ) | : j N , M ( ε , F j ) M ( ε , F k ) = + (i.e., similar to S ¯ ( ε , k , v * , F j ) of eq. 46, with no variable v * and M ( ε , F j ) = 0 , where we apply a related limit to eq. 47 that’s undefined due to division by infinity.)
6. 
sup | S ( ε , F j ) | : j N , M ( ε , F j ) M ( ε , F k ) = (i.e., similar to S ̲ ( ε , k , v * , F j ) of eq. 45, with no variable v * and M ( ε , F k ) = 0 , where we apply a related limit to eq. 47 that’s undefined since inf | S ( ε , F j ) | : j N , M ( ε , F j ) M ( ε , F k ) is an undefined empty set.)
7. 
| { z : j , z N , M ( ε , F j + z ) M ( ε , F j ) } | = + (i.e., infinite number succeeding F j are smaller than original F j , where such F j should be eliminated).
the limit in eq. 47 still exists.

5.4. Question:

How do we use mathematica code to illustrate §4 and 5?

References

  1. Krishnan, B. Mean of Unbounded Sets Using Method Similar To Conditional Expectation. ResearchGate. http://xxx.lanl.gov/abs/https://www.researchgate.net/publication/376597571_Mean_Of_Unbounded_Sets_Using_Conditional_Expectation_No_Programming_Just_Condensed_Version.
  2. Ott, W.; Yorke, J.A. Prevelance. Bulletin of the American Mathematical Society 2005, 42, 263–290. https://www.ams.org/journals/bull/2005-42-03/S0273-0979-05-01060-8/S0273-0979-05-01060-8.pdf. [CrossRef]
  3. John, R. Outlier. https://en.m.wikipedia.org/wiki/Outlier.
  4. Emily. How do computers draw function graphs? Mathematics Stack Exchange. https://math.stackexchange.com/q/634338. https://math.stackexchange.com/q/634338.
  5. M., G., 2nd ed.; Springer New York: New York [America], 2011; pp. 61–95. https://ee.stanford.edu/~gray/it.pdf. [CrossRef]
  6. E., B.; M., E. Integration with Filters. https://arxiv.org/pdf/2004.09103.pdf.
  7. Mathe. Mean of a function in Rn. Matchmaticians. https://matchmaticians.com/storage/answer/101942/matchmaticians-7gatrd-file-1.pdf https://matchmaticians.com/storage/answer/101942/matchmaticians-7gatrd-file-1.pdf.
  8. B., B.; A., F. Analogues Of The Lebesgue Density Theorem For Fractal Sets Of Reals And Integers. https://www.ime.usp.br/~afisher/ps/Analogues.pdf.
  9. B., B.; A., F. Ratio Geometry, Rigidity And The Scenery Process For Hyperbolic Cantor Sets. https://arxiv.org/pdf/math/9405217.pdf.
  10. Corrao, G. AN HENSTOCK-KURZWEIL TYPE INTEGRAL ON A MEASURE METRIC SPACE. https://core.ac.uk/download/pdf/53287889.pdf.
1
We want to find an unique and "natural" extension of the expected value, w.r.t the Hausdorff measure on bounded functions to other bounded/unbounded f, that takes finite values for all f in a non-shy subset of B * (def. 7)
2
We want to find an unique and "natural" extension of the expected value, w.r.t the Hausdorff measure on bounded functions to other bounded/unbounded f, that takes finite values for all f in a non-shy subset of B * (def. 7)
Figure 1. Below F r , F k , F z are non-equivalent starred sequences of sets, where V is all circles and E * * is the generalized expected value of f w.r.t either -sequence of sets (def. 8)
Figure 1. Below F r , F k , F z are non-equivalent starred sequences of sets, where V is all circles and E * * is the generalized expected value of f w.r.t either -sequence of sets (def. 8)
Preprints 98649 g001
Figure 2. Below F r , F k , F z are equivalent starred sequences of sets, where V is the entire circle and E * * is the generalized expected value of f w.r.t either -sequence of sets (def. 8)
Figure 2. Below F r , F k , F z are equivalent starred sequences of sets, where V is the entire circle and E * * is the generalized expected value of f w.r.t either -sequence of sets (def. 8)
Preprints 98649 g002
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.
Copyright: This open access article is published under a Creative Commons CC BY 4.0 license, which permit the free download, distribution, and reuse, provided that the author and preprint are cited in any reuse.
Prerpints.org logo

Preprints.org is a free preprint server supported by MDPI in Basel, Switzerland.

Subscribe

© 2024 MDPI (Basel, Switzerland) unless otherwise stated