18

The basis of this problem, and that which allows for the approximations to be made here, can be summarised in one approximation:

$$\Biggl(\frac{n^k -{\lfloor n^{\frac{1}{k}} \rfloor}^{k-1}\gcd({\lfloor n^{\frac{1}{k}} \rfloor}^{k-1},\Bigl\lfloor \frac{p_n^{k-1}}{n^{k-1}} \Bigr\rfloor)}{n^k -{\lfloor n^{\frac{1}{k}} \rfloor}\gcd({\lfloor n^{\frac{1}{k}} \rfloor},\Bigl\lfloor \frac{p_n^{k}}{n^{k}} \Bigr\rfloor)}\Biggr)^{\frac{1}{k}} \approx 1\quad\forall n,k \in \mathbb N\backslash {\{1}\}$$

$$\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad(A0)$$

$$\frac{\Bigl\lfloor \bigl(n^k -{\lfloor n^{\frac{1}{k}} \rfloor}^{k-1}\gcd({\lfloor n^{\frac{1}{k}} \rfloor}^{k-1},\Bigl\lfloor \frac{p_n^{k-1}}{n^{k-1}} \Bigr\rfloor)\bigr)^{\frac{1}{k}}\Bigr\rfloor }{\Bigl\lfloor \bigl(n^k -{\lfloor n^{\frac{1}{k}} \rfloor}\gcd({\lfloor n^{\frac{1}{k}} \rfloor},\Bigl\lfloor \frac{p_n^{k}}{n^{k}} \Bigr\rfloor)\bigr)^{\frac{1}{k}}\Bigr\rfloor} =1\quad\forall n,k \in \mathbb N\backslash {\{1}\}$$

$$\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad(A1)$$

$${\gcd\Bigl({\lfloor n^{\frac{1}{k}} \rfloor},\Bigl\lfloor \frac{p_n^k}{n^k} \Bigr\rfloor\Bigr)}\quad \Biggl|\quad \lfloor n^{\frac{1}{k}} \rfloor \gcd\Bigl({\lfloor n^{\frac{1}{k}} \rfloor}^{k-1},\Bigl\lfloor \frac{p_n^{k-1}}{n^{k-1}} \Bigr\rfloor\Bigr) $$

$$\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad(A2)$$

Defining the above ratio as varsigma: $$\varsigma_{n,k}= \frac{{{\lfloor n^{\frac{1}{k}} \rfloor}^{k-1}\gcd\Bigl({\lfloor n^{\frac{1}{k}} \rfloor}^{k-1},\Bigl\lfloor \frac{p_n^{k-1}}{n^{k-1}} \Bigr\rfloor\Bigr)} }{{\gcd\Bigl({\lfloor n^{\frac{1}{k}} \rfloor},\Bigl\lfloor \frac{p_n^k}{n^k} \Bigr\rfloor\Bigr)}}$$

We have the following:

$n \lt 2^k \Rightarrow \varsigma_{n,k}=1$

$$\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad(A3)$$

The maximum value of the integer remainder of the division of $n^k+m$ by $\gcd(\Bigl\lfloor \frac{p_n^{\, a}}{n^{\,b}} \Bigr\rfloor,n^c)$ is equal to $m$, when $a \gt 1$, $b \gt 1$ and $1 \leq c \leq k$.

This stated in inequalities:

$$a \gt 1\land b \gt 1 \land 1 \leq c \leq k \Rightarrow -m \leq n^k-\Biggl\lfloor\frac{n^k+m}{\gcd(\Bigl\lfloor \frac{p_n^{\, a}}{n^{\,b}} \Bigr\rfloor,n^c)}\Biggr\rfloor\gcd\Biggl(\Bigl\lfloor \frac{p_n^{\, a}}{n^{\,b}} \Bigr\rfloor,n^c\Biggr) \leq 0$$

$$\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad(A7)$$

So the question I am now asking, for that fun person that wants to close this page, is how do I establish a proof for (A5) and (A7) that will be rigorous and indisputable?

note that for now, I will leave the lemma as the restriction

$${\{a,b}\} \subset \mathbb N \land c \in {\{1,2,3,...,k}\} \Rightarrow n^k+m-\Biggl\lfloor\frac{n^k+m}{\gcd(\Bigl\lfloor \frac{p_n^{\, a}}{n^{\,b}} \Bigr\rfloor,n^c)}\Biggr\rfloor\gcd\Biggl(\Bigl\lfloor \frac{p_n^{\, a}}{n^{\,b}} \Bigr\rfloor,n^c\Biggr) \in {\{0,1,2,...,m}\}$$

But there most definitely exists congruence relations that have dependence in $(n,k)$ that allow us to reduce this condition to a specific subset of the least residue system modulo $m+1$ stated on the righthand side of the implicative arrow.

ADDITION NEEDED:

I apologize a lot if making another edit here again cause the website to spiral into chaotic unrest somehow, but the following lemma pertaining to the residues of the floor function of natural numbers raised to unit fraction exponents (as we see playing one of the most significant role in the original formula I found and stated in this discussion page), is very significant to the context as previously stated:

$$y \in{\{j \in \mathbb N : j=k^m+1 \land m \in\mathbb N\backslash {\{1}\} \land k \in\mathbb N }\}\Rightarrow \not\exists\, n \in \mathbb N \backslash {\{1}\}:\lfloor y^{\frac{1}{m}}\rfloor^{m}\equiv y (\operatorname{mod}n)$$

$$\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad(\operatorname{Residue000})$$

Attempting to simplify the above:

$$\exists\,\\ m \in \mathbb N \backslash {\{1}\}\,\,: (y-1)^{\frac{1}{m}}\equiv 0 (\operatorname{mod}1)\Rightarrow \not\exists\, n \in \mathbb N \backslash {\{1}\}:\lfloor y^{\frac{1}{m}}\rfloor^{m}\equiv y (\operatorname{mod}n)$$

$$\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad(\operatorname{Residue001})$$

Yesterday I noticed quite a strong fit for the approximation: $$\vartheta _{{n}}=\min\Biggl(\mathcal D\Bigl(n\cdot\Bigl\lfloor \frac{p_n}{n} \Bigr\rfloor\Bigr) \backslash {\{1}\}\Biggr)$$ $$n-\gcd(\bigl\lfloor \sqrt {n} \bigr\rfloor ,\vartheta _{{n}})) \approx A \cdot (n-1) +B$$

where $A \approx 1$ and $B \approx -1/2$ and $\mathcal D(n)$ denote the set of all divisors of $n$, $p_n$ is the $n^{th}$ prime.

$$\sqrt{\bigl( n^{2}-\bigl\lfloor \sqrt {n} \bigr\rfloor\cdot\gcd(\bigl\lfloor \sqrt {n} \bigr\rfloor ,\vartheta _{{n}})\bigr)}\sim n $$

$$\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad(R1)$$

$$\sqrt{\bigl( n-\gcd(n ,\vartheta _{{n}})\bigr)}+\frac{1}{\sqrt{n}}\sim \sqrt{n}$$

$$\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad(R2)$$

$$\sqrt{{n}^{2}-\max \left( \lfloor \sqrt{n} \rfloor ,n \right) \min \left( \gcd \left( \lfloor \sqrt{n} \rfloor,\vartheta_n \right) ,\gcd \left( n,\vartheta_n \right) \right) }+1+\delta_{{n}}\sim n$$

Where $\delta_n \in {\{-\frac{1}{2},0,\frac{1}{2}}\}$ is a discrete function for which I am unable to determine as yet. $$\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad(R3)$$

So I guess the best idea now would be for me to find either a value on $\mathbb N$ that satisfies neither of the following equalities: $$n- \Biggl(\Bigl\lfloor\sqrt {{n}^{2}- \lfloor \sqrt{n} \rfloor \cdot \gcd \left( \lfloor \sqrt{n} \rfloor ,\vartheta_n \right) } \Bigr\rfloor+1\Biggr) = 0 $$

$$\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad(R4)$$

$$n-\Biggl(\Bigl\lfloor \sqrt{{n}^{2}-\min \left( \lfloor \sqrt{n} \rfloor ,n \right)\cdot\min\left(\gcd ( \lfloor \sqrt{n}\rfloor,\vartheta_n) ,\gcd ( n,\vartheta_n) \right) }\Bigr\rfloor +1\Biggr) = 0$$ $$\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad(R5)$$

Figure 1: enter image description here

Figure 2:

enter image description here

Defining a generalisation of vartheta: $$\vartheta _{{n,k}}=\min\Biggl(\mathcal D\Bigl(n^{k}\cdot\Bigl\lfloor \frac{p_n^{k}}{n^{k}} \Bigr\rfloor\Bigr)\, \backslash\, {\{1}\}\Biggr)$$

$$\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad(R6)$$

Will allow for the following asymptotic relation as we would intuitively expect from the nature of the generalisation and the nature of $(R2)$: $$(n^k -\lfloor n^{\frac{1}{k}} \rfloor\gcd(\lfloor n^{\frac{1}{k}} \rfloor,\vartheta _{{n,k-1}}))^{\frac{1}{k}} \sim n \quad \forall k \in \mathbb N\backslash {\{1}\}$$

$$\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad(R7)$$

Which is based on the following equalities:

$$\lfloor (n^k -\lfloor n^{\frac{1}{k}} \rfloor\gcd(\lfloor n^{\frac{1}{k}} \rfloor,\vartheta _{{n,k-1}}))^{\frac{1}{k}} \rfloor+1=n\quad \forall k \in \mathbb N\backslash {\{1}\} $$

$$\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad(R8)$$

$$\lfloor (n^k -\gcd(\lfloor n^{\frac{1}{k}} \rfloor,\Bigl\lfloor \frac{p_n^{k}}{n^{k}} \Bigr\rfloor))^{\frac{1}{k}} \rfloor+1=n\quad \forall k \in \mathbb N\backslash {\{1}\} $$

$$\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad(R9)$$

$$\lfloor (n^k -{\lfloor n^{\frac{1}{k}} \rfloor}^{k-1}\gcd({\lfloor n^{\frac{1}{k}} \rfloor}^{k-1},\Bigl\lfloor \frac{p_n^{k-1}}{n^{k-1}} \Bigr\rfloor))^{\frac{1}{k}} \rfloor+1=n\quad \forall k \in \mathbb N\backslash {\{1}\} $$

$$\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad\quad(R10)$$

Adam Ledger
  • 1,240
  • We can slightly simplify the left side to $$n-\gcd(\lfloor \sqrt{n} \rfloor,q)$$ where $q$ is the smallest prime factor of $$n\cdot \lfloor \frac{p_n}{n} \rfloor$$ – Peter Jun 10 '18 at 08:28
  • Further we can conclude that the left side is either $n-1$ or $n-q$ – Peter Jun 10 '18 at 08:30
  • True yes. I have been trying to read PNT Wikipedia pages and feel as if there will definitely be a better least square regression choice than linear I just can't see it yet because I have done very little in approximations, always feel cheap about it but they are a discipline in themselves really – Adam Ledger Jun 10 '18 at 08:31
  • what is the criteria for determining that though? – Adam Ledger Jun 10 '18 at 08:32
  • I am guessing something to do with the mobius function – Adam Ledger Jun 10 '18 at 08:33
  • perhaps I should change the question here to what is the exact formula? – Adam Ledger Jun 10 '18 at 08:34
  • The prime $q$ seems to depend pseudorandomly on $n$. I do not think it will be easy to find a good pattern. I would suggest you show the data you already have and the constants $A$ and $B$ you got for larger values. This can be doublechecked by other users. – Peter Jun 10 '18 at 08:42
  • Sure ok no problem, I have a lot of content I have done over the weekend, is it ok if I share it with you in chatrooms or should I just put everything I think do be related? – Adam Ledger Jun 10 '18 at 08:47
  • Another problem : Whether $q|\lfloor \sqrt{n}\rfloor$ also seems to depend pseudorandomly on $n$. In most cases, we have $q=2$ – Peter Jun 10 '18 at 08:47
  • Well I have an interesting question that I believe to be relevant regarding the Euler product of $n$ and this quantity, I am just typing the latex for it now should I just put it in this question seeing relates to all of the same quantities but in relation to the Euler product? – Adam Ledger Jun 10 '18 at 08:50
  • I would suggest to share your work in the question, not in a chat. It is surely interesting also for other users. – Peter Jun 10 '18 at 08:51
  • the denominator of the Euler product of $n$ seems to always divide both the greatest common denominator of $\bigl\lfloor \sqrt {n} \bigr\rfloor$ and $q$, and the gcd of $n$ and $q$ – Adam Ledger Jun 10 '18 at 08:52
  • Well it consistently does, so investigating the criterion of why this is the case when it is, and when it is not is a priority for me – Adam Ledger Jun 10 '18 at 08:58
  • greatest common divisor* of both lol – Adam Ledger Jun 10 '18 at 09:10
  • 1
    Due to Dusart's bound for the $n$ th prime number, for $n\ge 6$, we know $$\frac{p_n}{n}<\ln(n\ln(n))$$ and therefore $$q<\ln(n\ln(n))$$ Hence for large $n$, $q$ is very small compared with $n$ and therefore $n-1$ will be an extremely good approximation (considering the relative error) – Peter Jun 10 '18 at 09:46
  • That makes sense then. How did he or she derive that bound? actually ill look it up. – Adam Ledger Jun 10 '18 at 09:54
  • I don't understand how we can "prove" bounds that are only valid for certain domains of the naturals. For me a proof can only be something that is derived from fundamental lemmas, the Wikipedia page tells me of these approximations and Dusart has proven them, but this is really confusing for me because I have always considered any approximation to be heuristic in nature, therefore a proof does not exist – Adam Ledger Jun 10 '18 at 10:04
  • 1
    https://en.wikipedia.org/wiki/Prime-counting_function shows, among other things, the (proven) bounds from Pierre Dusart. – Peter Jun 10 '18 at 10:21
  • I just noticed that Dusart found the lower bound, not the upper bound, nevertheless it has been proven. Surely a consequence of the prime number theorem, but I do not know the details of the proof. – Peter Jun 10 '18 at 10:22
  • It is correct that often bounds are only heuristics, but in this case, you can trust the inequality. – Peter Jun 10 '18 at 10:25
  • The website is telling me to move the conversation to chat, anyway trusting anything or anyone generally works out horribly I've found, but I can understand it would most definitely come from a more professional person. – Adam Ledger Jun 10 '18 at 10:42
  • The time complexity for the linear approximation must be reasonably good for me to have the LSR coefficients in such a short amount of time, but that doesn't guarantee that I won't be able to find a better curve to approximate to, it only implies that it is a good approximation, as naïve as it may be I will probably keep trying to find an exact formula – Adam Ledger Jun 10 '18 at 10:45
  • The fluctuations won't be predictable. For exact results, you will have to calculate the values directly. Even bounding the ABSOLUTE error for EVERY $n$ will be infeasible. – Peter Jun 10 '18 at 11:16
  • 1
    For comparison : Finding the $10^{100}$ th prime EXACTLY is infeasible at least with the current known methods and computational power. But very good approximations of the $10^{100}$ th prime are possible. But the ABSOLUTE error will still be large. – Peter Jun 10 '18 at 11:18
  • I just can't see how an approximation can be considered very good if there even IS an error, how can a property of the naturals become unpredictable, when the arithmetic remains predictable? – Adam Ledger Jun 10 '18 at 15:44
  • 2
    Your goal is to construct a formula that avoids brute force calculation. If not, you do not need the formula and can calculate the values directly. "Predictible" means that a program can get the correct answer much faster and always correct and needs no cumbersome calculations. – Peter Jun 10 '18 at 15:52
  • 3
    You have two chunks of text, both labeled "final edit and summary". I don't know which final edit and summary is the final final edit and summary. At the end of the top final edit and summary, you ask, "how do I establish a proof that will be rigorous and indisputable?" But it is not clear what you want to prove. There's nothing anywhere labeled "Theorem", or "Conjecture", just a multitude of equations and formulas. I'm very tempted to vote to close as "unclear what you are asking". – Gerry Myerson Nov 14 '18 at 03:42
  • Good. I retract my previous comment. – Gerry Myerson Nov 15 '18 at 01:47
  • Please, stop making so many edits. Excessive edits bump questions for no good reason and are disruptive to the site. – Aloizio Macedo Nov 29 '18 at 14:39
  • Is an addition made based on new information intimately related the problem considered a good reason? Why has this bumping occurred for this one exclusively? Seems like it would be a pretty strange design flaw to have a questions ranking on the display list 'proportional to the number of edits. – Adam Ledger Nov 29 '18 at 15:35
  • I think that it would only inform those that have clicked on the little star on the left hand side that an edit has taken place, thereby informing exclusively those interested that an edit has taken place. – Adam Ledger Nov 29 '18 at 15:37
  • Yeah sorry about that I'm not exactly great at forward planning – Adam Ledger Apr 16 '22 at 19:34

0 Answers0