Googology Wiki
Register
Advertisement
Googology Wiki

Circle notation[]

Here, it uses what looks like circle notation - can circle notation accept parameters? If it can, this should be mentioned on Circle notation. {{subst:1x|~~}}{{subst:1x|~~}}

Circle notation, an extension to Steinhaus-Moser Notation, is entirely different from Rob Munafo's hyper-operators. Followed by 100 zeroes (talk | contribs) 21:01, 7 March 2009 (UTC)

Geisler's site[]

I've been trying for a while to decode Geisler's site, but the math is giving me migraines -.- Is anyone interested in perusing that? FB100Ztalkcontribs 03:09, April 29, 2012 (UTC)

Small examples only?[]

Well, I think that only giving small examples doesn't tell you how large the larger numbers are generated by tetration, so it won't be helpful. Large examples tell you how this function is important in googology. Cloudy176 (talk) 10:38, November 18, 2012 (UTC)

Okay, you're right. Some examples reaching into scientific notation and beyond would be nice, but don't let's fill the page with seas of digits. FB100Ztalkcontribs 20:56, November 18, 2012 (UTC)

Real tetration[]

It's don't seem to me that \(^{1/x}a\) = \(\sqrt[x]{a}_s\).

Just see that the infinith-super root of 2 = \(\sqrt{2}\), but \(^{0}a\) = 1, and 1/0 = infinity. Ikosarakt1 (talk) 13:11, January 13, 2013 (UTC)

What's about defining \(a \uparrow\uparrow (1/b)\) as \(\sqrt[a \uparrow\uparrow (b-1)](a)\). That's reasonable, since for b=1 it gives a, and as \(b \rightarrow \infty\) it tends to 0.

Last digits convergence[]

We know very well that for integers \(n\), the last digits of power towers of \(n\) converge, making it easy to predict the last digits of a very large power tower (such as a chained arrow or BEAF array). For example, a power tower of threes converges to ...04575627262464195387, the final digits of Graham's number.

There's a page on the InterNet about "oology," which explores infinities recreationally. (Perhaps coincidentally, the page is from the André Joyce Fan Club, from which the term "googology" first originated.) It has silly things like turning fractions backwards: \(1/3 = 0.333\ldots\), so \(3\backslash 1\) ("3 conquered by 1," as in "divide and conquer") is \(\ldots 333\). Although clearly this is recreational mathematics and somewhat tongue-in-cheek, these look a lot like the notation for digit convergence in tetration. This can give us funny-looking equations such \(^\infty3 = \ldots 04575627262464195387\). FB100Ztalkcontribs 02:08, February 17, 2013 (UTC)

If I take the conquer operator for granted, \(x/1\) is the reversal of \(1\backslash x\). So I guess we have \(1\backslash ^\infty3 = 0.78359146426272657540\ldots\) okay, what the hell did I just write FB100Ztalkcontribs 02:12, February 17, 2013 (UTC)
I've typed a simple-math book about integer tetration, exploring its p-adic convergence properties. There are also  pseudo-convergence properties involving the figures next to the rightmost ones. The caos on the left fight against the order on the right of the number and there are some rules to point out this feature. In short, it's easy to predict the digit on the left of the leftmost frozen digit (without calculating tetration or using modular arithmetics and cycles), but it's very complex to predict other digits, even if you have already found their mutual constraints.
SPIqr (talk) 02:14, February 19, 2013 (UTC)

In popular culture[]

I found a rare occurrence of tetration humor: http://www.smbc-comics.com/?id=1666 (click on the red button below the comic) FB100Ztalkcontribs 06:04, September 12, 2013 (UTC)

What about Knuth Paper-Stack Notation? -- I want more clouds! 07:17, September 12, 2013 (UTC)
True, true. Still, the SMBC comic may be the world's first sex joke with tetration in it. FB100Ztalkcontribs 07:57, September 12, 2013 (UTC)
And then they went on to make one with pentation and hexation in it. ArtismScrub (talk) 02:38, December 1, 2017 (UTC) 

"natural" tetration[]

is there any tetrational analog to the function \(f(x) = e^x\)? that is, is there a constant \(\nu\) such that \(f(x) = {}^x\nu\) is a "natural" function? FB100Ztalkcontribs 08:58, September 24, 2013 (UTC)

We have to define tetration for real heights before answering this question. Ikosarakt1 (talk ^ contribs) 15:22, September 24, 2013 (UTC)
Since e is the infinite sum of the factorial reciprocals, I would say that your \(\nu\) might possibly be the infinite sum of the expofactorial reciprocals. Then again, since ba is not defined for non-integer b, this is meaningless. —Preceding unsigned comment added by ArtismScrub (talkcontribs) 03:52, January 24, 2018 (UTC)

Definition for ordinals[]

I think it's natural to define w^^(1+a) = w^(w^^a), but not w^^(a+1) = w^(w^^a). For finite "a", it doesn't matter, but for a = w there is the fact that w^^(1+w) = w^(w^^w) = w^(e(0)) = e(0) = w^^w. That's consistent with the fact that 1+w = w.

For w^^(a+1), we should use Saibian's definition I believe. So w^^(a+1) = {w^^a}^{w}. Don't be confused with something like (w^^a)^w, look how the proper variant works for a = 4: w^^(4+1) = {w^^4}^{w} = {w^w^w^w}^{w} = w^w^w^w^w, but not (w^w^w^w)^w. Ikosarakt1 (talk ^ contribs) 07:56, July 27, 2014 (UTC)

Nontrivial solutions[]

in the natural numbers to a^^b = c^^d? Are there numbers expressible as a tetration in more than one way? —Preceding unsigned comment added by 160.94.192.193‎ (talkcontribs)

I think I have proven that there are none, but I'm not enirely sure. Full proof with all details would be quite lengthly. LittlePeng9 (talk) 15:59, August 6, 2014 (UTC)

Could you say what your approach was? —Preceding unsigned comment added by 160.94.192.193‎ (talkcontribs)

Suppose that we had a solution. Then we would have a^(a^^(b-1))=c^(c^^(d-1)). Every solution of p^q=r^s has p and r being integer powers of some number k. So we have a=k^l, c=k^m. From non-triviality k isn't 1. Now by laws of exponentiation we can get l*(k^l)^^(b-2)=m*(k^m)^^(d-2). If we assume l>=m then we get l/m is an integer and a power of k. By further manipulation we get that l must be divisible by m*(k^m)^^(d-2), which leads to the contradiction. I used induction to show this, and I'm not entirely sure if this will work. LittlePeng9 (talk) 18:21, August 6, 2014 (UTC)

Actually, it is l*(k^l)^^(b-1) = m*(k^m)^^(d-1). Continuing, let l = m*k^n. Then (m*k^n)((k^(m*k^n))^^(b-1)) = m*(k^m)^^(d-1), and therefore n + (m*k^n)*((k^(m*k^n))^^(b-2)) = m*((k^m)^^(d-2)). Thus n = m*((k^m)^^(d-2)) - (m*k^n)*((k^(m*k^n))^^(b-2)) = m*k^u - m*k^v, where we have v >= n and therefore u >= n. So k^n divides n, which is impossible.

This is of course assuming a,b,c,d > 1 and a != c. Deedlit11 (talk) 03:23, August 7, 2014 (UTC)

***[]

I attempted to create a redirect there, but this was blocked by a spam filter. --84.61.216.30 21:36, February 24, 2018 (UTC)

Can somebody else create the redirect from *** to tetration? --84.61.216.30 21:49, February 24, 2018 (UTC)

I can't create the redirect as well. I suggest reporting this to w:c:vstf:Report:Spam filter problems. -- ☁ I want more clouds! ⛅ 08:01, February 25, 2018 (UTC)

And what about the redirect from / to Extended Cascading-E Notation#Further extensions? --84.61.216.30 13:37, February 25, 2018 (UTC)

Last digits 2[]

In December, I was at the center of a whole debate on here about the computation of last digits of tetrations. It turns out that the original description on this article was indeed wrong, as the method works in base 10 but fails in most other bases. Observe:

\(7 \uparrow\uparrow 2\) = 823,543 == 985 mod 11^3

\(7 \uparrow\uparrow 3\) = \(7^{823543}\) == \(7^{985}\) mod \(11^3\) = 857

\(7 \uparrow\uparrow 4\) = \(7^{7^{823543}}\) == \(7^{857}\) mod \(11^3\) = 378

but this is wrong, because \(\phi(1331)\) (using Euler's totient function) is 1,210, which doesn't evenly divide into 1331. Using the correct method, we have:

\(7 \uparrow\uparrow 2\) = 823,543 == 985 mod 11^3

\(7 \uparrow\uparrow 3\) = \(7^{823543}\) == \(7^{823543 mod 1210}\) or \(7^{743}\) mod \(11^3\) = 332

\(7 \uparrow\uparrow 4\) = \(7^{7^{823543}}\) == \(7^{7^{823543} mod 1210}\) or \(7^{453}\) mod \(11^3\) = 24

However, using Mathematica, I observed that 3^15597484987 ends with the same last 10 digits as 3^5597484987 (...6100739387) and thus \(3 \uparrow\uparrow 4\) also ends in ...6100739387. Similarly, 3^16100739387 ends with the same last 10 digits as 3^6100739387, confirming that the last 10 digits of \(3 \uparrow\uparrow 5\) are indeed ...9660355387. This is because the period of the last d decimal digits of a^b for integers a and b is always a divisor of 10^d for d >= 2 (but not for d = 1).

\(2^{36}\) = 68,719,476,736

\(2^{136}\) = 87,112,285,931,760,246,646,623,899,502,532,662,132,736

This means that 2^...36 = ...736 regardless of the digits before the first ...36, so the last 2 digits of n indeed determine the last 3 digits of 2^n (the same goes with 3^n, 4^n, 5^n, etc). Allam948736 (talk) 20:47, January 27, 2020 (UTC)

No. It has a counterexample. Finding a pattern in specific value can never be a proof, as I pointed out so many time. Please understand the meaning even if you do not find a counterexample.
p-adic 23:30, January 27, 2020 (UTC)
In the case of "2^...36 always ends with ...736", multiplying by \(2^{100}\) in the example I gave clearly has no effect on the last 3 digits, meaning that the last digits will always be ...736 no matter how many times you multiply by \(2^{100}\). I was talking specifically about the last decimal digits of \(^ba\) when a is not a multiple of 10.  Allam948736 (talk) 01:15, January 28, 2020 (UTC)
I am talking about your statement "This means that 2^...36 = ...736 regardless of the digits before the first ...36, so the last 2 digits of n indeed determine the last 3 digits of 2^n (the same goes with 3^n, 4^n, 5^n, etc)." The first sentence doe snot imply the second sentence. In order to avoid shifting of the goalpost, could I ask you to fix the precise statement? You are talking about, say, 3^n for any n, which does not necessarily appear in the computation of a↑↑b. What you stated is "for any natural numbers n and m such that m is not divisible by 10. the last 3 digits of m^n in base 10 coincides with the last 3 digits of m^{n+100}"? Otherwise, please fix a precise statement with full quantifications.
Also, could you understand that even if you are talking only about a↑↑b, your observation of specific values of 2^n does not imply something general? I recommend you the following two:
  1. To understand that observing specific values does not imply a general fact.
  2. To write a statement in a precise formula with full quantifications.
p-adic 01:53, January 28, 2020 (UTC)


> What you stated is "for any natural numbers n and m such that m is not divisible by 10. the last 3 digits of m^n in base 10 coincides with the last 3 digits of m^{n+100}"?

Yes, although this only holds for n >= 3 if m = 2. In fact, for any m, there can only be four possibilities for the last digit of m^n, (1, 3, 7, and 9 for m coprime to 10, 2, 4, 6, and 8 for m even and not divisible by 5, and the last digit is fixed for m divisible by 5 or 1 mod 5). m^n can only end in 2, 3, 7, or 8 if n is odd, which means that if m^n ends in 3 or 7 and m is 3 mod 4, then m^n is also 3 mod 4 and so the tens digit must be even. If m is 1 mod 4, then m^n is also 1 mod 4 regardless of n, meaning the tens digit must be odd in those cases. On the other hand, if m^n ends in 1 or 9 and neither m+1 or m-1 is divisible by 5, then n must be even, meaning that m^n must be 1 mod 4 since -1^2 = 1, thus the tens place digit must be even. If m-1 is divisible by 5, however, the tens place digit of m^n as n increases cycles through all 10 digits if it doesn't repeat sooner, and so does the hundreds place digit each time you increase n by 10. If m+1 is divisible by 5 however then n must be odd if m^n ends in 9 and thus the tens digit of m^n must be odd. For an even value of m, m^n will always be divisible by 4 for n >= 2, which means the tens place digit of m^n must be even if the final digit is 4 or 8 and odd if the final digit is 2 or 6. In each case, there are possibilities for the last 2 digits that never actually occur. With an even value of m (which is the simplest case, read the previous paragraph), there are 4 possibilities for the units digit and only 5 possibilities for the tens place digit, meaning that the last 2 digits repeat after 20 powers, and there are also 5 possibilities for the hundreds place digit given the last 2 digits, thus the period of the last 3 digits of m^n for all even values of m not divisible by 10 is 20*5 = 100. We can similarly show that the period for the last 4 digits is 500, 2500 for 5 digits, 12500 for 6 digits, and so on.

Proving this for odd values of m is a bit harder. If m == 3 mod 4 and m =/= 1 or -1 mod 5, then m^n == 3 mod 4 if n is odd and 1 mod 4 if n is even. Since 3^2 = 9 and 7^2 = 49, m^n will end in a 9 if n is even and not divisible by 4, or 1 if n is divisible by 4. In both cases, this means that the tens digit must be even, which already rules out a power of m ever ending in 11, 19, 31, 39, 51, 59, 71, 79, 91, or 99. If n is odd, then m^n must be 3 mod 4 and end in either 3 or 7, meaning that the tens digit again must be even, leaving only 20 possibilities for the last 2 digits. m^4 will also be 1 mod 8, since 3^4 = 81 == 1 mod 8 and 7^4 = 2401 == 1 mod 8. This means that m^n mod 8 repeats with a period of 4, which evenly divides 20, thus the first value of n for which the last 2 digits of m^n repeat is also 1 mod 8, leaving only five choices for the hundreds digit given the two digits to the right, giving 100 possibilities for 3 digits. In the case where m == 1 mod 5, this is already proven since there are 10 possibilities for the last 2 digits, and the period can be multiplied by no more than 10 if we add another digit. Lastly, if m == 1 mod 4, m^n is always 1 mod 4 regardless of n, meaning the tens digit must be odd if the last digit is 3 or 7, and must be even if the last digit is 1 or 9. 1^2 is just 1, and 5^2 = 25 == 1 mod 8, meaning if m is 1 mod 4, then m^2 and thus m^n for all even values of n must be 1 mod 8. We have already proved that the last 2 digits repeat after 20 and that m^20 == 1 mod 8. There are only 5 possibilities for the last 3 digits that are both congruent to 1 mod 8 and 25, meaning the last 3 digits once again repeat after 100 powers. Note that simply proving that the last 3 digits repeat after 100 powers is already enough to prove that the recursive modular exponentiation method for tetration works in base 10, since the period can be multiplied by no more than 10 if we add another digit. Allam948736 (talk) 02:21, January 28, 2020 (UTC)

> Yes, although this only holds for n >= 3 if m = 2.
Well, what are you going to verify then? You understand that what you stated is wrong, while you are continuing explaining the classification of the possibilities.
> Note that simply proving that the last 3 digits repeat after 100 powers is already enough to prove that the recursive modular exponentiation method for tetration works in base 10, since the period can be multiplied by no more than 10 if we add another
As you already noticed that it does not hold for the case n < 3, it is not provable... What precise statement are you going to show? Please write down full statements before trying to show them.
p-adic 09:39, January 28, 2020 (UTC)


It only doesn't hold for n < 3 if m is congruent to 2 mod 4, because m^n is always divisible by 8 for n >= 3 if m is even, but if m is congruent to 2 mod 4, then neither m nor m^2 are divisible by 8. If m == 4 mod 8 then m is obviously not divisible by 8, but m^2 is a multiple of 8, meaning in those cases it starts working at n = 2. Allam948736 (talk) 15:38, January 28, 2020 (UTC)
It might be related (a link from the last digits of Graham's number in Wikipedia.) Triakula (talk) 15:50, January 28, 2020 (UTC)
@Allam948736
> It only doesn't hold ... because'
Again, you are wrong in this logic. What you explained is that the statement does not hold in that case. It does not imply that it holds otherwise. What you are stating is for any natural numbers n and m satisfying the following, the last 3 decimal digits of m^n coincides with those of m^{n+100}, right? Please fix a statement by yourself, instead of saying "it does not hold if..." or "I assume ...".
  1. m is not divisible by 10.
  2. n≧3 if m is congruent to 2 mod 4.
  3. n≧2 if m is congruent to 4 mod 8.
@Triakula
We know it, because we were talking about it so many times. Allam948736 wrote wrong arguments again and again on last digits of powers without fixing precise statements or showing explicit proofs. There were so many wrong descriptions on last digits in this wiki written by Allam948736 without sources. That is why I require Allam948736 to fix a precise statement before explaining how it is believed to be correct.
p-adic 22:48, January 28, 2020 (UTC)


By the recursive modular exponentiation method to compute the last \(d\) decimal digits of \(^yx\), I mean this:

\(a_1 = x\) \(a_n = x^{a_{n-1}} mod 10^d\) repeated until we reach \(a_y\)

I have proven that it doesn't work if we replace the 10 with another base (say, 11), but it works in base 10! The last 3 decimal digits of m^n always coincide with those of m^(n+100) if m is odd or divisible by 8, and for even numbers not divisible by 8, the values of 1 and 2 are just special cases. This is already enough to prove that the above method works (already explained why). Allam948736 (talk) 03:25, January 30, 2020 (UTC)

> I have proven that it doesn't work if we replace the 10 with another base (say, 11), but it works in base 10!
Although you are talking as if it were your own great discovery, I told you so many times that it does not necessarily hold when the base is not 10 (while you first stated by yourself that it worked for any non-prime bases) and I have explicitly written down the proof for cases including base 10.
The last 3 decimal digits of m^n always coincide with those of m^(n+100) if m is odd or divisible by 8, and for even numbers not divisible by 8, the values of 1 and 2 are just special cases.
You fixed your statement. Why couldn't you fix your statements before you are aked so many times?
>This is already enough to prove that the above method works (already explained why).
No, the observation of the behaviour of the last 3 digits does not imply the validity of the recursive modular exponentiation method. Should I repeat to say "observing lower values itself does not gives a proof" again? How many time should I say? (I conjecture that you will change your statement by saying "it is obvious that d is always assumed to be 3" or something like that, although you intensionally skipped the quantification of d, which I required so many times.)
p-adic 10:58, January 30, 2020 (UTC)


The reason why proving that the period of the last 3 digits is 100 is enough to verify that the recursive modular exponentiation method works in base 10 is because the period can be multiplied by no more than 10 if we add another digit. Allam948736 (talk) 15:50, January 30, 2020 (UTC)
> because the period can be multiplied by no more than 10 if we add another digit.
You have not verified it. Moreover, there is more general a method applicable to wider bases in the article. Why are you trying to prove it in the case where the base is 10? Just you could not understand the article, right?
p-adic 22:50, January 30, 2020 (UTC)


No, I understood the article. Also, could someone actually calculate the last 20 decimal digits of sufficiently high power towers of 2, 3, 4, and 5 using the method described in the article and see if the results are the same as those found here ? Allam948736 (talk) 02:00, January 31, 2020 (UTC)
If you understood the articke, then why would you try to "prove" the statement for the special case as if it were a novel idea?
p-adic 10:52, January 31, 2020 (UTC)
The reason I'm trying to prove the statement for the special case b = 10 is simply because base 10 is the most commonly used. My earlier results found here  were in base 10. Allam948736 (talk) 15:46, January 31, 2020 (UTC)


I used the following method in Wolfram Mathematica: \(s_1 = a\), \(s_n = a^{s_{n-1} \mod \phi(10^d)} \mod 2\times10^d\), and I obtained ...98,615,075,353,432,948,736 as the last 20 decimal digits of a sufficiently high power tower of 2s, and ...04,575,627,262,464,195,387 as the last 20 decimal digits of a sufficiently high power tower of 3s, which are the same results I got using the basic recursive modular exponentiation method which were found hereAllam948736 (talk) 16:15, February 10, 2020 (UTC)

It does not mean that your method generally works for any tetration... Why couldn't you understand it although I repeat to tell you that coincidence at a specific value does not mean that the algorithm always works?
p-adic 23:37, February 10, 2020 (UTC)


I tried it for other values too, and got the same results: for a base of 4, ...22,302,555,290,411,728,896, for a base of 5, ...17,493,152,618,408,203,125. I already explained why the correct method should coincide with the basic recursive modular exponentiation method. This is because the period of the last d decimal digits of \(m^n\) evenly divides 10^d iff d >= 2 for all values of m not divisible by 10. Allam948736 (talk) 01:34, February 11, 2020 (UTC)
I said that checking finitely many specific values does not give a proof and I explained why your explanation is not a proof. How many time could I repeat the same comments?
p-adic 01:47, February 11, 2020 (UTC)
But can it even be proven generally? What if in "computing" many last digits we're always relying only on guesses? Triakula (talk) 08:59, February 11, 2020 (UTC)
It is not necessarily provable in a reasonable length. If it is not proved, then it is just good not to write down it in the article as if it were a verified fact. I am requiring him proofs because he added so many guesses without proofs to articles and stated that they were correct without showing proofs or explicit algorithms. However, once I disproved that segmental informations of his secret method yield contradiction, he changed the statement and started to state as if he had noticed the issue by himself. Now this is the time when he states as if he verified that his method works without showing explicit proofs. He just lists correct equalities, which are not proofs. It is something like writing 1+1=2, 2+3=5, and so on and stating "therefore Fermat's last theorem was proved by me". Listing correct equalities does nothing bad, but does not yield a proof. I explained this so many times, but he always ignores the requirement. Nevertheless, he added his guess without proofs again. I would like to know when he realises the importance of a proof.
p-adic 09:40, February 11, 2020 (UTC)
If it is not provable in a reasonable length, then I don't think that we should asking him to actually write down a proof. Instead, we just can point him that it is an unprovable sentence in the current mathematical equipment. In my opinion, it looks less challenging. Triakula (talk) 09:59, February 11, 2020 (UTC)
> If it is not provable in a reasonable length, then I don't think that we should asking him to actually write down a proof.
I am not saying that this topic is not provable. The problem is that he is stating as if he has verified it. We should distinguish guesses and known facts. I have already asked him to distinguish his guess without proofs from facts here.
> Instead, we just can point him that it is an unprovable sentence in the current mathematical equipment. In my opinion, it looks less challenging.
I am not saying that it is currently unprovable, but am saying that what he explains is not a proof at all, although he believes that it is a proof. I have pointed out so many times the following points:
  • Write down an explicit proof applicable to any cases instead of showing the result of observation of finitely many smaller values. Otherwise, it can never be a proof.
  • Fix a precise statement with full quantification instead of saying something like "I am only considering the case where a=10, b is greater than 10^100, c is non-prime number coprime with 5(a^2+1)" after I point out counterexamples.
He never learns above two issues, and continues to state that he is right until I show him a counterexample. Should I construct an explicit counterexample whenever he writes his guess without a proof as if it were a fact? No way. He should write down a proof, if he insists that it is a fact.
p-adic 11:01, February 11, 2020 (UTC)
Maybe he doesn't know how the proof about this topic must look. I see that the proof-by-examples is wrong, but can't propose a good proof either. So, if I would be in the same situation as him, I might be not able to give a reasonable response here. So, maybe you can write a proof yourself, if you got the idea of his method? I think that he would appreciate it. Triakula (talk) 12:17, February 11, 2020 (UTC)
You are completely right. I think so, too. Therefore I actually have adviced him how to write a proof on a statement including a universal quantifier on natural numbers in the paragraph "You need to learn how to write mathematical equations~" here. Moreover, I have written many counterexamples on his statements and many proofs for statements corrected by myself here. While he just states that he is right, I spend so much time to write down counterexamples, alternative correct statements, full proofs with full sources for the corrected statements, and advices on how to avoid wrong arguments. If he does not understand them, he can say so and ask me the meanings. On the other hand, when I asked him what points in the article he could not understand, he clarified that he understood the descriptions in the article. What should I do more?
p-adic 12:51, February 11, 2020 (UTC)
In that case, I don't know what should we do. It might be just a difference in style of thinking - mathematically educated view is that everything must be proven using some axioms while commonly people doing many things just by-examples. This wiki may be edited by any people though. There's not much we can do here other than just correcting things. Triakula (talk) 13:11, February 11, 2020 (UTC)
Sure. That is why I am repeating to point out his incorrectness. I hope that someday he will understand that listing finitely many examples is not a proof even if he is not given a counterexample.
p-adic 13:16, February 11, 2020 (UTC)


The new method that I provided in my message yesterday works because \(a^b \mod c\) can repeat after no more than \(\phi(c)\), which in the case of \(10^d\) is \(4 \times 10^{d-1}\). The reason why we take the value mod \(2 \times 10^d\) is because \(\phi(10^d)\) doesn't evenly divide \(10^d\), but it does evenly divide \(2 \times 10^d\). Allam948736 (talk) 17:14, February 11, 2020 (UTC)
Also, could someone try to calculate the last digits of \(^yx\) for sufficient values of y using the methods described in the article and settle this whole issue once and for all? Allam948736 (talk) 21:24, February 11, 2020 (UTC)
I told you how to write a proof.
p-adic 22:52, February 11, 2020 (UTC)


See the proof that my new method works that I just wrote here. Allam948736 (talk) 01:55, February 12, 2020 (UTC)

@Allam

> See the proof that my new method works that I just wrote here.

First of all, it is irrelevant to the topic which we are talking about, right? Are you giving up to write down the proof for it?

Anyway, ok, I can check your argument in your blog post. I strongly hope that you will learn many from your failure this time instead of ignoring feedbacks.

EDIT: I think that it is better to directly comment to your blog post, and hence I deleted the original comment here. p-adic 03:08, February 12, 2020 (UTC)


I edited the blog post and removed a few redundant statements, and added clarification for some estimations.

The reason that there are \(y_1 - 1\) convergent digits in \(3 \uparrow\uparrow n\) is because the period of the last \(d\) digits of \(3^n\) is equal to \(5 \times 10^{d-2}\) for d >= 4 (100 for d=3 and 20 for d=2), which always divides into \(10^{d-1}\) but is greater than \(10^{d-2}\), and 27 (\(^23\)) has one convergent digit. 2 convergent digits are added instead of just 1 each time the n in \(^nm\) is incremented by one if and only if either \(m \equiv 0 \mod 5\) or \(m^4 \equiv 1 \mod 25\). Allam(2^^n mod 10^6 for n >= 8) (talk) 00:43, February 17, 2020 (UTC)

The intuitive explanation is not a proof, as I pointed out in my comment to the blog post. How about solving the issues instead of just ignoring the feedbacks?
p-adic 01:06, February 17, 2020 (UTC)


I have fixed numbers 1, 2, 3, 5, 6, 9, 10, and 12 since I wrote the blog post. The x, y, b, and d from the first equality are equivalent to the a, b, c, and d in the proof. By the period I meant the number of powers after which the last base-c digit will repeat. Allam(2^^n mod 10^6 for n >= 8) (talk) 03:13, February 18, 2020 (UTC)

Could you please reply the comment above to my comment, because it is not suitable for this talk page anymore? (This is the talk page for tetration, but not for teaching you.) Also, why could you regard your explanation as a proof, even though you understand that you have not fixed all the issues yet?
p-adic 04:20, February 18, 2020 (UTC)

Congruence speed[]

Let me introduce a peculiar property of the integer tetration (which holds for any base that does not end with zero), which is related to the number of new frozen digits at the end of the tetration a^^b for a unitary increment of the hyperexponent. It has been indicated as the constancy of the "congruence speed" of a^^b in NNTDM, 26(3), pp. 245-260 and in NNTDM, 27(4), pp.43-61, where the complete formula of the congruence speed has been published. Basically, for any positive integer base which is not congruent to 0 (mod 10), I have proved the existence of a function from N\{0} in NU{0} which let us know more about the number of stable digits for any tetration: for any given value of this "congruence speed" V(a), we can find a period of length 25 iff V(a)=1 or 10^(V(a)+1) iff V(a)>1. Moreover (in the aforementioned last paper), it has been published the formula for all the bases characterized by any given congruence speed value.

—Preceding unsigned comment added by SPIqr (talkcontribs)


Number of stable digits of integer tetration[]

A paper providing the direct map of the congruence speed of tetration has just been published on NNTDM 28(3), pp. 441-457. Since it fully describes the relation between the number of stable digits of any integer tetration and the congruence class modulo 20 of the base, it would be important to add it to the references in order to provide both the inverse map (NNTDM, 27(4), pp.43-61) and the direct map (NNTDM, 28(3), pp. 441-457) of the constant congruence speed of tetration. Unfortunately, since I am not able to do it by myself (I've just tried and failed), it would be possible to consider the aforementioned paper for the references, adding the hyperlink next to Ref. 4, since it completes and extends its main result (new theorems are proved thanks to the previous results from NNTDM 28(3), pp. 441-457, so we need to include both in order to prove that Equation 16 of NNTDM, 28(3), pp. 441-457 holds for any tetration base which is not a multiple of 10)?

Thanks in advance for your consideration.

SPIqr (talk) 01:43, 17 August 2022 (UTC)

I added the reference. You edit failed because the name tag was not unique. 🐟 Fish fish fish ... 🐠 09:02, 17 August 2022 (UTC)

Perfect! SPIqr (talk) 03:25, 18 August 2022 (UTC)

First digits?[]

As stated in the article, it is unlikely to be possible to get it down to O(log* n), let alone constant time. It is mentioned that the best known way to extract arbitrary digits of the logarithm is a BBP formula, which unfortunately still runs in linear time. However, is there a better way than the currently known method to calculate first digits? Obviously, it would involve a series for the base-10 logarithm of a positive number that could be used to extract digits beginning at a given place in any base. However, I only know of series for the natural logarithm, and there seems to be no connection between digits of the natural logarithm and the base-10 logarithm at a given position because we have to divide by the irrational ln 10. Allam(2^^n mod 10^6 for n >= 8) (talk) 03:11, 29 November 2023 (UTC)

Actually, it turns out that the series for ln x and ln 10 can be combined to produce a series for log10(n) with rational coefficients. With some tricks using modular arithmetic or similar, it may then be possible to find the fractional part of \(^{b-1}a \times \log_{10}(a)\) without having to calculate the entire integer part, but we need it to be very efficient. Allam(2^^n mod 10^6 for n >= 8) (talk) 02:17, 8 December 2023 (UTC)

Revisiting the last digits issue[]

Can we re-add the statement that the last digits of tetrations converge as \(y \mapsto \infty\), at least in base 10? Additionally, can we re-add the digits I tried to add in December of 2019 during my previous time on here? The information in the article already suggests that the method works in base 10 when d is at least 2 and x is not a multiple of 10 (While the method with b=10 is only guaranteed to work correctly if x is coprime to 10, values of x that are not coprime to 10 but not divisible by 10 are only problematic when the output at the previous step is less than d). Additionally, do we still need the warning about unsourced (and possibly wrong) statements? I tried to change that part recently but it got reverted. The article should still state that the recursive formula at the beginning is wrong, as it is supposedly for any base and not just decimal digits, and it is known not to work in many other bases. You may also be interested in my recent blog post entitled "Last digits of tetrations again". Allam(2^^n mod 10^6 for n >= 8) (talk) 00:01, 20 December 2023 (UTC)

We can add statements as long as we cite valid sources in a way following the policy, if we put non-ambiguous description. For example, an ambiguous statement without full quantifications of variables or full conditions should be avoided. Also, when we describe an algorithm, we need to clearly describe it so that whether it works or not mathematically makes sense. For example, referring to an ambiguous method name without specifying how it actually works can cause the trouble in the past days.
Concerning the warning, we need to be responsible for what we have spread, because many people actually refered to this wiki as "sources" of mathematical facts. Therefore, we are keeping warnings on incorrect statements in this wiki (e.g. TREE_sequence#Wrong_arguments_of_the_growth_rate, Buchholz_hydra#Growth_rate_of_BH(n)_function, Rathjen's psi function#Common misconceptions, and so on.) That is why we should not irresponsibly delete them as if this wiki has no responsibility for the spreading. I think that this discussion log might be what you want to read. As the discussion concludes, reducing the emphasis by changing the bold font to the usual font is ok, but removing past mistakes is not ok. Thank you for sincerely asking here before change!
p-adic 09:41, 20 December 2023 (UTC)

We proven this result here: https://arxiv.org/abs/2208.02622 and we went even further here: https://arxiv.org/abs/2210.07956 Basically, at least in radix-10, if you want an easy (far too redundant) condition for being sure that the last k digits of the integer tetration a^^b are frozen (i.e., stable digits), it is sufficient to take b>=a+k since, for every a>1, b>=a+1 is a sufficient (but not necessary) condition for having a constant congruence speed of the base. Now, we know that the constant congruence speed of tetration is a strictly positive integer for any a>1, so that at least k digits will be frozen after k steps by starting from b=a+1 (including a^^(a+1) itself), and this number of frozen digits for any unit increment of b is fixed and it does not depend on a anymore! Thus, you can safely state that, for any positive integers k, a, and b, a^^(a+k)==a^^(a+k+c)(mod 10^k) holds for every c=0,1,2,3,... SPIqr (talk)

Thank you. Since they seem to be preprints, it is good to cite them as non-peer-reviewed sources. If there are published versions, then it is better to cite them.
By the way, concerning the stability, although Alam states that the article's methods are not applicable to base 10 when x is not coprime to 10. However, if I remember correctly, isn't the article already dealing with such a case at the paragraph starting with "For the case where b is not necessarily squarefree or the length d of digits is not necessarily equal to 1"? I am sorry if I am missing a point since it was written years ago. (Or are there any known issues on the argument? At least, Fish also invented a similar computation method for non-coprime case here.)
p-adic 22:59, 20 December 2023 (UTC)

My bad. The part "Even if x is not necessarily coprime with b if b is squarefree..." actually addresses that (as 10 is squarefree), so really the only condition for the method to work in base 10 is that x not be divisible by 10 (And that d be at least 2, of course). Allam(2^^n mod 10^6 for n >= 8) (talk) 00:21, 21 December 2023 (UTC)

Thank you for the clarification. If I correctly understand, we do not need to care about the corner case where x is fortunately divisible by 10, because the last d digits of x^y is always 0...0 when y>=d in that case, do we?
p-adic 07:58, 21 December 2023 (UTC)

@p-adic: Here are the peer-review papers (IMO, the corresponding arXiv versions are simply better than the journal versions since they are in LaTeX), so feel free to cite the whole trilogy of my published papers on this topic, including the last two that I mentioned above: 1) "On the constant congruence speed of tetration" (2020): https://nntdm.net/volume-26-2020/number-3/245-260/ ; 2) "The congruence speed formula" (2021): https://nntdm.net/volume-27-2021/number-4/43-61/ ; 3) "Number of stable digits of any integer tetration" (2022): https://nntdm.net/volume-28-2022/number-3/441-457/ Now, the provided proof for the rightmost stable digits assumes radix-10, but I suspect that the same proof idea (by using the same ring homomorphism and g-adic numbers, solving the related fundamental equation y^c(g)=y, and so on...) can be easily extended to any other radix-g system, and this belief is supported by Germain's paper entitled "On the Equation a^x ≡ x (mod b)", but it is still an open problem (as far as I know). SPIqr (talk)

Thank you for telling me published versions! (I understand such preferences, but this academic wiki policy recommends peer-reviewed sources. Also, as both of a mathematician and a referee of many mathematical journals, I personally prefer to cite peer-reviewed journals in order to thank not only to the author but also all contributors, e.g. referees, editors, and so on.)
By the way, since you are (perhaps the best in this wiki) an expert on this topic, I appreciate if you yourself add the new results in a precise manner. If you have some troubles in editing or wiki formatting, of course we will help you. Would you mind editing the article?
p-adic 13:45, 21 December 2023 (UTC)

You are welcome. Unfortunately, at the moment I am ill (very bad flu), but I think that I could help you here if you have some concerns about the published results. Basically, we call V(a) the constant congruence speed of tetration and its value is given by Equation (16) of https://nntdm.net/volume-28-2022/number-3/441-457/, which give us the number of new(!) rightmost frozen digits that appears after every unit increment of the hyperexponent, when it is sufficiently large (this minimum value depends on the base). Thus, a (trivial) sufficient condition for having k stable digits at the end of the result of a^^b is to take b=a+ceiling(k/V(a)) (and since V(a)>=1 for any a>1, b=a+k is always enough), but we can achieve a very tight bound on the minimum value of the hyperexponent by taking b=2+tilde(v)(a), where tilde(v)(a) is stated in Definition 2.1 of https://nntdm.net/volume-28-2022/number-3/441-457/. In the above, we are assuming that a=/=0(mod 10), but if not, then a is not characterized by a constant congruence speed... it has a constant congruence acceleration and we get more and more new(!) trailing 0s at the end of the result for any further unit increment of the hyperexponent as it grows. I think that this would be enough for a Wiki; the papers tell a lot more and people can read them (at any time) by following the references. SPIqr (talk) So, can we add the following to the article? The last digits of \(^yx\) for sufficient values of y for specific values of x in base 10 are given below: 2: ...98,615,075,353,432,948,736 3: ...04,575,627,262,464,195,387 4: ...22,302,555,290,411,728,896 5: ...17,493,002,618,408,203,125 6: ...67,965,593,227,447,238,656 7: ...43,331,265,511,565,172,343 8: ...21,577,035,416,895,225,856 9: ...10,748,087,597,392,745,289 Allam(2^^n mod 10^6 for n >= 8) (talk) 22:01, 21 December 2023 (UTC)

@SPIqr
Thank you for the introduction. Oh, I hope that you will get well soon... Since I am currently working on two other papers for reviewing, I am not ready for editing the article deeply. However, I (or others, perhaps Allam) will edit the article later.
@Allam
We can, if we cite sources for the values.
p-adic 22:12, 21 December 2023 (UTC)

Added a short section trying to resume what I previously wrote in my last comment above. I hope that someone of you will edit it in the right way ASAP (thanks in advance). SPIqr (talk)

Thanks!
p-adic 22:33, 27 December 2023 (UTC)

As for the edit I made earlier this month, my intention was not to be unconstructive. I was actually trying to edit the explanation to be a bit more fitting for a mainspace article. Allam(2^^n mod 10^6 for n >= 8) (talk) 15:37, 30 December 2023 (UTC)


When I made the edit I did just over a month ago, my intention was not to be unconstructive. I was actually just trying to rewrite an explanation in a way that I felt might be more fitting for a mainspace article. As such, it was not vandalism, although I was wrong to remove the statement that the formula was wrong, for reasons I have already explained. Allam(2^^n mod 10^6 for n >= 8) (talk) 03:25, 7 January 2024 (UTC)

I think that your two comments above are suitable for your own talk page rather than this section. Therefore I recommend you to move them (together with removing this reply). In addition, I think that nobody is doubting you.
p-adic 06:21, 7 January 2024 (UTC)

Something interesting I found[]

Usually, the last digits of tetrations converge to a random sequence of digits. However, there are some instances where this is actually not the case. For instance, the last digits of \(27 \uparrow\uparrow n\) in base 5 converge to an alternating pattern: \(...13131313131313131313_5\), which is in fact the 5-adic value of -1/3. Additionally, a power tower of 256s in base 5 converges to ...11111111111111111111, which is the 5-adic -1/4. Now, what is special about 27 and 256? Both of these can be expressed as n^^2 for an integer n, and a power tower of \(n^n\) will become \(n^{n^{n^{...^{n^n + 1}...} + 1}}\). This cannot be computed using the "wrong" formula on the article, because \(\phi(5^n) = 4 \times 5^{n-1} \nmid 5^n\) and 5 is an odd prime. Additionally, are there any numbers that generate such patterns in base 10? Allam(2^^n mod 10^6 for n >= 8) (talk) 16:01, 12 January 2024 (UTC)

Advertisement