Latest revision as of 00:05, 23 January 2025

• Computing
• Entertainment
• Humanities
• Language
• Mathematics
• Science
• Miscellaneous
• Archives

• WP:RD/MA

Main page: Help searching Misplaced Pages

How can I get my question answered?

• Select the section of the desk that best fits the general topic of your question (see the navigation column to the right).
• Post your question to only one section, providing a short header that gives the topic of your question.
• Type '~~~~' (that is, four tilde characters) at the end – this signs and dates your contribution so we know who wrote what and when.
• Don't post personal contact information – it will be removed. Any answers will be provided here.
• Please be as specific as possible, and include all relevant context – the usefulness of answers may depend on the context.
• Note:
  • We don't answer (and may remove) questions that require medical diagnosis or legal advice.
  • We don't answer requests for opinions, predictions or debate.
  • We don't do your homework for you, though we'll help you past the stuck point.
  • We don't conduct original research or provide a free source of ideas, but we'll help you find information you need.

How do I answer a question?

Main page: Misplaced Pages:Reference desk/Guidelines

• The best answers address the question directly, and back up facts with wikilinks and links to sources. Do not edit others' comments and do not give any medical or legal advice.

• Teahouse
• Help desk
• Village pump
• Help manual

January 8

Even Fermat pseudoprimes

There are infinitely many odd numbers which are Fermat pseudoprimes to all bases coprime to them (the Carmichael numbers), but it seems that all even numbers > 946 are Fermat pseudoprimes to at most 1/8 for the bases coprime to them, is this proven? (Like that all odd numbers are strong pseudoprimes to at most 1/4 for the bases coprime to them) 220.132.216.52 (talk) 12:40, 8 January 2025 (UTC)

For what it's worth, any counterexample >28 must have the form ${\displaystyle 2pq}$ where p and q are distinct odd primes such that ${\displaystyle 2pq-1}$ is divisible by both ${\displaystyle (p-1)/2}$ and ${\displaystyle (q-1)/2}$. Tito Omburo (talk) 18:55, 8 January 2025 (UTC)

If we denote ${\displaystyle a:=\gcd((p-1)/2,(q-1)/2)}$ such that ${\displaystyle P:=(p-1)/2a}$ and ${\displaystyle Q:=(q-1)/2a}$ are coprime, then ${\displaystyle (p-1)/2=aP}$ and ${\displaystyle (q-1)/2=aQ}$ divides ${\displaystyle 2pq-1=2(2aP+1)(2aQ+1)-1=8a^{2}PQ+4aP+4aQ+1}$. Naturally, this implies ${\displaystyle aP}$ divides ${\displaystyle 4aQ+1}$ and ${\displaystyle aQ}$ divides ${\displaystyle 4aP+1}$, which further implies that ${\displaystyle a=1}$ and thus ${\displaystyle (p-1)/2}$ and ${\displaystyle (q-1)/2}$ are coprime. So we must find (odd) coprime ${\displaystyle P,Q}$ such that ${\displaystyle P|4Q+1}$ and ${\displaystyle Q|4P+1}$, and from there ${\displaystyle 2P+1}$ and ${\displaystyle 2Q+1}$ must be prime. GalacticShoe (talk) 16:13, 16 January 2025 (UTC)

Suppose we have odd coprime ${\displaystyle P,Q}$ such that ${\displaystyle P|4Q+1}$ and ${\displaystyle Q|4P+1}$. Assume WLOG that ${\displaystyle P<Q}$. This means that ${\displaystyle 4P+1}$ can only be ${\displaystyle Q}$ or ${\displaystyle 3Q}$. If ${\displaystyle 4P+1=Q}$, then ${\displaystyle P|4Q+1\Rightarrow P|4(4P+1)+1\Rightarrow P|16P+5\Rightarrow P=1,5,Q=5,21}$. This yields ${\displaystyle p:=2P+1=3,11,q:=2Q+1=11,43}$ which both work, giving values ${\displaystyle 2pq=66,946}$. If ${\displaystyle 4P+1=3Q}$, then ${\displaystyle P|4(4P+1)/3+1\Rightarrow P|(16P+7)/3\Rightarrow P|16P+7\Rightarrow P=1,7}$. For either value though ${\displaystyle 4P+1}$ is not divisible by ${\displaystyle 3}$, so this doesn't yield any values. We conclude that the only counterexamples greater than ${\displaystyle 28}$ are indeed ${\displaystyle 66,946}$. GalacticShoe (talk) 01:32, 17 January 2025 (UTC)

It is notable that 28, 66, 946 are triangular numbers, and their indices (7, 11, 43) are Heegner numbers, is this a coincidence? 220.132.216.52 (talk) 10:59, 17 January 2025 (UTC)

The other triangular numbers whose indices are Heegner numbers are 190 (19), 2278 (67), 13366 (163), but 190 is Fermat pseudoprime only to 1/8 for the bases coprime to it, 2278 is Fermat pseudoprime only to 1/32 for the bases coprime to it, 13366 is Fermat pseudoprime only to 1/16 for the bases coprime to it (8, 32, 16 are powers of 2). 220.132.216.52 (talk) 13:09, 17 January 2025 (UTC)

I seem to know the reason:

(Heegner numbers corresponding to the prime-generating polynomial n^2+n+p, i.e. the number p = (the Heegner number + 1)/4

66 -> 11th triangular number -> 11 and (11+1)/4 = 3 -> 11-1 and 3-1 totally have 2 prime factors of 2 -> 1/(2^2) = 1/4 of the bases coprime to it

190 -> 19th triangular number -> 19 and (19+1)/4 = 5 -> 19-1 and 5-1 totally have 3 prime factors of 2 -> 1/(2^3) = 1/8 of the bases coprime to it

946 -> 43rd triangular number -> 43 and (43+1)/4 = 11 -> 43-1 and 11-1 totally have 2 prime factors of 2 -> 1/(2^2) = 1/4 of the bases coprime to it

2278 -> 67th triangular number -> 67 and (67+1)/4 = 17 -> 67-1 and 17-1 totally have 5 prime factors of 2 -> 1/(2^5) = 1/32 of the bases coprime to it

13366 -> 163rd triangular number -> 163 and (163+1)/4 = 41 -> 163-1 and 41-1 totally have 4 prime factors of 2 -> 1/(2^4) = 1/16 of the bases coprime to it 220.132.216.52 (talk) 13:16, 17 January 2025 (UTC)

@Lambiam: @GalacticShoe: @RDBury: @Bubba73: 220.132.216.52 (talk) 14:45, 16 January 2025 (UTC)

Rectifying latitude

Is it true that using the rectifying latitude with the best sphere radius (possibly not the same radius for all 3) minimizes the worst-case error for distance (%), distance (km) & max km the 2 paths get from each other? (maximum separation between the great circle & the geodesic for the surface of the WGS84 ellipsoid)? Or is another latitude better like the geocentric latitude? (the geocentric latitude can get ~0.2° from the (by far) most kind of used latitude which is more separation than any kind of latitude (besides the Mercator one that's 0° to ∞°)) What's the best radius for each of these 3 metrics & how well do the worst point pairs for these 3 metrics approximate the ellipsoidal trigonometry answer? (the one where the geodesic latA lon A alt0 to latB lonB alt0 is considered perfect accuracy even though most places aren't on the 2D surface) Sagittarian Milky Way (talk) 22:56, 8 January 2025 (UTC)

January 10

Knot equivalence

Our knot theory article gives two definitions, asserted to be equivalent, for when one knot is equivalent to another. Both of these involve auto-homeomorphisms of ${\displaystyle \mathbf {R} ^{3}}$.

To me this just kind of feels -- heavy. To move one trefoil knot to another, verifying their equivalence, I apparently have to account for all the rest of space. Is this actually necessary?

Concretely, suppose I define knot equivalence as follows. If ${\displaystyle K_{0}}$ and ${\displaystyle K_{1}}$ are two knots, defined as continuous maps from into ${\displaystyle \mathbf {R} ^{3}}$ that are injective except that ${\displaystyle K_{i}(0)=K_{i}(1)}$ (${\displaystyle i=0,1}$), then say ${\displaystyle K_{0}}$ and ${\displaystyle K_{1}}$ are equivalent if there is a continuous map ${\displaystyle f}$ from ${\displaystyle \times }$ into ${\displaystyle \mathbf {R} ^{3}}$ such that:

• the map ${\displaystyle x\mapsto f(x,0)}$ equals ${\displaystyle K_{0}}$
• the map ${\displaystyle x\mapsto f(x,1)}$ equals ${\displaystyle K_{1}}$
• for ${\displaystyle 0<t<1}$, the map ${\displaystyle x\mapsto f(x,t)}$ is injective except that ${\displaystyle f(0,t)=f(1,t)}$

Is this notion of equivalence the same as the one in the article? --Trovatore (talk) 07:01, 10 January 2025 (UTC)

Since no ${\displaystyle K_{t}}$ intersects itself, my intuition tells me there ought to be a lower bound on how close it can come to self-intersection. More precisely, the boundary of the Minkowski sum ${\displaystyle K_{t}+B_{\delta }}$ of ${\displaystyle K_{t}}$ and a constant ball ${\displaystyle B_{\delta }}$ with a sufficiently small (${\displaystyle t}$-independent) radius ${\displaystyle \delta }$ should be a torus. Proving this formally may not be easy, but a proof will establish the required ambient isotopy.

If your conjecture is indeed correct – I'm hedging my bet because counterexamples in topology can be quite counterintuitive – it would be amazing if your markedly simpler characterization is not found in the literature.  --Lambiam 10:20, 10 January 2025 (UTC)

I think knot (lol). I believe any nontrivial not is "injectively homotopic" to the unknot. Start with your favorite knot and stretch one arc while compressing the rest so that in the limit the compressed part tends to a point. This can be done injectively. Tito Omburo (talk) 12:03, 10 January 2025 (UTC)

Yeah, that sounds right. Thanks. --Trovatore (talk) 18:22, 10 January 2025 (UTC)

With the R embedding definition you can definitely say that if the complements are not homeomorphic then the knots are inequivalent. That's a useful property since you may be able to show that two knots are different without having to go into the details of knot theory, in fact you can show the trefoil knot is not the unknot with a little algebraic topology. In practice, two knots are equivalent if their diagrams can be transformed into each other by a sequence of Reidemeister moves, so the problem is really to find an intuitive definition that's equivalent. As Lambiam pointed out, topology can be counterintuitive, and perhaps "injectively homotopic" = "homotopic" is an example of that. --RDBury (talk) 00:40, 11 January 2025 (UTC)

Think yourself lucky. Maybe. When I had a look at knot theory a while ago the definition was in terms of a finite set of straight lines and Reidemeister moves, and things like infinite knots were an extension. And then there were the extensions like knotting a sphere in 4d. NadVolum (talk) 14:50, 10 January 2025 (UTC)

January 15

Least common multiple

What do you call a subset of the factors of x where all 3 are true?

1. LCM(subset)=x
2. no member >√x
3. can't be done with less members

(obviously only some x would have even 1 subset passing all 3)

Is there also a name for a version where no member ≥√x or a version with tiebreaks (maybe smallest largest member then smallest 2nd largest member and so on?) or a version with both extra strictures? Sagittarian Milky Way (talk) 22:56, 15 January 2025 (UTC)

It seems unlikely that anyone has before come up with this specific set of conditions, let alone coined a term for it.  --Lambiam 22:24, 16 January 2025 (UTC)

I was looking at factor lists of numbers with many factors & thinking most of these are superfluous to define a least common multiple & wondering how many you could remove without the LCM becoming less than the number (obviously there are other applications where you need all factors) but if you don't set a max size it's too easy you could always get it down to 2 with lcm(1,x) sometimes also others like lcm(prime, bigger prime) or lcm(2,x/2 if odd) so if you must use factors ≤√x it'll at least be over 2 members so more interesting. Sagittarian Milky Way (talk) 16:32, 17 January 2025 (UTC)

Consider a set ${\displaystyle S}$ of numbers whose lcm equals a given number ${\displaystyle n}$. Let ${\displaystyle a}$ and ${\displaystyle b}$ be two different elements of that set that are not coprime. This means that they have nontrivial factors ${\displaystyle p^{\alpha }}$ and ${\displaystyle p^{\beta }}$ in which ${\displaystyle p}$ is a prime number and ${\displaystyle \alpha }$ and ${\displaystyle \beta }$ are at least ${\displaystyle 1.}$ Assume wlog that ${\displaystyle \alpha \leq \beta .}$ Then the lcm of set ${\displaystyle S'}$ oobtained by replacing ${\displaystyle a}$ by ${\displaystyle a'=a/p^{\alpha }}$ is the same as that of ${\displaystyle S.}$ This means that we can keep things simpler by only considering sets with pairwise coprime elements. Then the lcm function can be replaced by the product operator ${\displaystyle \textstyle \prod ,}$ where ${\displaystyle \textstyle \prod \{a_{1},a_{2},...a_{k}\}=a_{1}a_{2}\cdots a_{k}.}$

Given the factorization ${\displaystyle n=p_{1}^{m_{1}}p_{2}^{m_{2}}...p_{k}^{m_{k}},}$ one possible choice for ${\displaystyle S}$ is the set ${\displaystyle F=\{p_{1}^{m_{1}},p_{2}^{m_{2}},...,p_{k}^{m_{k}}\}.}$ Given a partition ${\displaystyle {\mathcal {P}}}$ of ${\displaystyle F,}$ define ${\displaystyle \textstyle \prod ^{\ast }\!({\mathcal {P}})}$ by

${\displaystyle \textstyle \prod ^{\ast }\!({\mathcal {P}})=\{\textstyle \prod X\,|\,X\in {\mathcal {P}}\}.}$

Then ${\displaystyle \textstyle \prod (\textstyle \prod ^{\ast }\!({\mathcal {P}}))=n,}$ so for each partition ${\displaystyle {\mathcal {P}},}$ the set ${\displaystyle \textstyle \prod ^{\ast }\!({\mathcal {P}})}$ is also a candidate for ${\displaystyle S.}$ Conversely, each set ${\displaystyle S}$ of mutually coprime numbers such that ${\displaystyle \textstyle \prod S=n}$ can be written as ${\displaystyle \textstyle \prod ^{\ast }\!({\mathcal {P}})}$ for some partition ${\displaystyle {\mathcal {P}}}$ of ${\displaystyle F.}$ The original ${\displaystyle F}$ itself equals ${\displaystyle \textstyle \prod ^{\ast }\!(\{\{p_{1}^{m_{1}}\},\{p_{2}^{m_{2}}\},...,\{p_{k}^{m_{k}}\}\}).}$

Instead of partitioning ${\displaystyle F}$ and then applying ${\displaystyle \textstyle \prod ^{\ast }\!(\cdot )}$ to the partition, we can obtain the same candidates for ${\displaystyle S}$ by starting with ${\displaystyle S_{0}=F}$ and obtaining ${\displaystyle S_{n{+}1}}$ from ${\displaystyle S_{n}}$ — provided that ${\displaystyle S_{n}}$ is not a singleton set — by choosing two elements from ${\displaystyle S_{n}}$ and replacing these two by a single element, namely their product. If we wish to keep the values low, a reasonable greedy heuristic is to pick each time the smallest two elements.

Applying this to Plato's favourite number, we get:

${\displaystyle 5040=2^{4}3^{2}5^{1}7^{1},}$ so ${\displaystyle S_{0}=\{2^{4},3^{2},5^{1},7^{1}\}=\{16,9,5,7\}.}$

${\displaystyle S_{0}=\{5,7,9,16\}}$; the smallest two elements are ${\displaystyle 5}$ and ${\displaystyle 7,}$ so ${\displaystyle S_{1}=\{5{\times }7,9,16\}=\{35,9,16\}.}$

${\displaystyle S_{1}=\{9,16,35\}}$; the smallest two elements are ${\displaystyle 9}$ and ${\displaystyle 16,}$ so ${\displaystyle S_{2}=\{9{\times }16,35\}=\{144,35\}.}$

${\displaystyle S_{2}=\{35,144\}}$; the smallest two elements are ${\displaystyle 35}$ and ${\displaystyle 144,}$ so ${\displaystyle S_{3}=\{35{\times }144\}=\{5040\}.}$

If the largest element must not exceed the square root of ${\displaystyle n,}$ the set ${\displaystyle S}$ has to contain at least three elements, so with ${\displaystyle k}$ being the number of distinct prime factors, there is no point in going farther than ${\displaystyle S_{k-3}.}$  --Lambiam 23:11, 17 January 2025 (UTC)

January 17

Probability questions

First Question: You walk into a room filled with random people. You want to find another person in that room who has the same birthday as you. For example, June 15. How many people would need to be in the room? How do you go about solving this question?

Second Question: Same as above. However, you want to find another person in that room who has the same birth date as you. For example, June 15, 1985. How many people would need to be in the room? How do you go about solving this question?

Thanks, 32.209.69.24 (talk) 08:08, 17 January 2025 (UTC)

These are both different from the well-known birthday problem.

For the first, let's ignore the possibility of people born on February 29 in a leap year, so there are only 365 possible birthdays. Let us also assume that all 365 birthdays are equally likely, so for any fixed day D of the year, such as January 17, the probability ${\displaystyle p}$ that a randomly selected person's birthday falls on that very same day is equal to ${\displaystyle {\tfrac {1}{365}}.}$ The probability that this person's birthday falls on a different day is then equal to the complement ${\displaystyle 1{-}p={\tfrac {364}{365}}.}$

It is easier now to consider the complementary question: What is the probability ${\displaystyle q_{N}}$ that none among ${\displaystyle N}$ randomly selected persons has a given birthday D. The answer to the original question is then given by its complement, ${\displaystyle 1-q_{N}.}$

If ${\displaystyle N=0,}$ there is no one whose birthday could be D, so ${\displaystyle q_{0}=1.}$ If ${\displaystyle N=1,}$ with just one other (randomly selected) person present, ${\displaystyle q_{1}}$ is just the probability that this person's birthday is D, so ${\displaystyle q_{1}=1{-}p.}$ Now suppose we already know ${\displaystyle q_{n}}$ for some value of ${\displaystyle n.}$ Then we can determine ${\displaystyle q_{n{+}1}}$ by considering that the joint probability of two independent events co-occurring is equal to the product of their individual probabilities. Therefore ${\displaystyle q_{n{+}1}=q_{n}\times q_{1}.}$ We can conclude that in general

${\displaystyle q_{n}=q_{1}^{n}=(1{-}p)^{n}=({\tfrac {364}{365}})^{n}.}$

The probability of the same birthday as yours among a random selection of ${\displaystyle N}$ people is therefore ${\displaystyle 1-({\tfrac {364}{365}})^{N}.}$

Now note that as ${\displaystyle n}$ gets larger and larger, the value of ${\displaystyle q_{n}}$ gets smaller and smaller, but it never reaches zero exactly. Even if ${\displaystyle N=365,}$ we find that ${\displaystyle 1-({\tfrac {364}{365}})^{365}=0.632625...~.}$ To get to 99%, ${\displaystyle N}$ should be at least ${\displaystyle 1679}$; ${\displaystyle 1-({\tfrac {364}{365}})^{1678}=0.98998...}$ falls still short, but ${\displaystyle 1-({\tfrac {364}{365}})^{1679}=0.99001...}$ reaches the target.

The approach assumes that the possible birthdays are uniformly distributed over the population, which is not the case in reality. However, to account for this, you only need to know the real value of ${\displaystyle p}$ for day D and not for any other day.

To find a somewhat realistic answer to the second question is harder. In reality, the people in a room will not be a random sample from the total population. People below the age of 3 and over the age of 97 will be underrepresented, so if your own birthdate is January 17, 1925, the likelihood of today finding someone present to jointly celebrate your 100th birthday with is much smaller than that of finding a co-celebrant for your 35th birthday if your birthdate is January 17, 1990. The notion of "random selection" is not clearly applicable. You need to know at least the distribution of birthyears among the population from which the people in the room are selected, accounting both for the actual population pyramid and for age-based selection bias. When you have determined ${\displaystyle p_{\text{Y}},}$ the probability that a person randomly selection from those present in the room has the same birthyear Y as you, instead of ${\displaystyle p=1/365}$ you can use ${\displaystyle p=p_{\text{Y}}/365}$ and proceed as above.  --Lambiam 11:44, 17 January 2025 (UTC)

Wow. Very thorough, detailed, and comprehensive. You certainly have a gift for math. Thanks! Let me read this over and process it all. I'll need a day or two. Thanks so much. 32.209.69.24 (talk) 08:39, 18 January 2025 (UTC)

January 18

Is it possible to tweak Kunerth's algorithm so that it returns a different possible solution ?

The Kunerth's algorithm is a non generic modular square root algorithm that compute modular square roots without factoring the modulus…

Let’s say I’ve a valid input for which the algorithm can return a solution, is it possible to tweak it so that it returns a different possible solution ? So far I only found how to modify it to return the modular inverse… — Preceding unsigned comment added by 2A01:E0A:401:A7C0:9D9:50BB:6262:E787 (talk) 06:39, 18 January 2025 (UTC)

This question has also been asked (and answered) at User talk:Endo999 § A question about Kunerth’s algorithm….

Being unfamiliar with Kunerth's algorithm I tried to understand it from our article, but the presentation is so confusing (also in the Example section) that I kept getting lost.  --Lambiam 09:21, 18 January 2025 (UTC)

Personally, I used this implementation to understand it. As the article do contains errors, my question is merely how to modify the implementation in order to get a different valid ouput for the same input 2A01:E0A:401:A7C0:9D9:50BB:6262:E787 (talk) 14:15, 18 January 2025 (UTC)

Regarding the answer I received on the other page, it seems to me the author doesn’t fully understand the algorithm. 2A01:E0A:401:A7C0:416E:83F1:C73F:88BD (talk) 12:46, 20 January 2025 (UTC)

January 19

Looking for a function like y=1/x with adjustable corner curve

I'm trying to figure out how to make a function that passes through (1,1) like ${\displaystyle y=1/x}$ (I'm interested only in x>0), but lets me adjust the sharpness of that "corner" at (1,1), thereby increasing or decreasing the tail thickness of the curve while still passing through (1,1), with the tails remaining between 0<y<1 and 0<x<1, and the slope at (1,1) remaining -1.

My application for this is to create metaballs that can maintain a reasonably small radius while having a blobby connection to other metaballs at long distances. I've been experimenting with Desmos to come up with such a function but haven't hit on anything yet. ~Anachronist (talk) 09:11, 19 January 2025 (UTC)

The usual equation for a hyperbola symmetric about the y axis passing through (a, 0) would be ⁠x/a⁠-⁠y/b⁠=1 so if you want to rotate it by π/4 I guess it would be ⁠(x cos ⁠π/4⁠ + y sin ⁠π/4⁠)/2⁠-⁠(ycos ⁠π/4⁠ - x sin ⁠π/4⁠)/b⁠=1 or something like that? Alpha3031 (t • c) 14:28, 19 January 2025 (UTC)

Thank you. Yes I tried something like that but while I could fix the knee of the curve in one place, I couldn't keep the asymptotes 90° apart at the same time and adjust the sharpness of the knee. ~Anachronist (talk) 17:32, 19 January 2025 (UTC)

Oh, right, if you want right hyperbolas only then you'd need to hold the eccentricity constant as well at √2 which means they're all the same shape and you're just rescaling it by a factor. The easiest equation for that is probably ⁠xy/A⁠ = 1 (or equivalently, y = A/x) which has a vertex at (√A, √A). Translating (√A, √A) to (1, 1) would make the equation y = ⁠A/x+√A-1⁠-√A+1, A ∈ (0, 1]. Alpha3031 (t • c) 01:48, 20 January 2025 (UTC)

Yes, that holds the knee at (1,1) but it also moves the asymptotes so that the positive side of the function no longer approaches the axis, but instead approaches a constant ${\displaystyle 1-{\sqrt {a}}}$. I did finally come up with something (see below) but when applying it to metaballs, it didn't have the effect I had hoped for. ~Anachronist (talk) 23:21, 20 January 2025 (UTC)

I would start from 1/x but use a varying exponent. 1/x is x^-1, i.e. with an exponent of minus one. Larger negative exponents will I think do as you describe. X ^ -2 or 1/x^2 will give a twice as steep/sharp slope at (1, 1) and go to zero more rapidly. You can try other values, including non-integer values, for the exponent. --2A04:4A43:909F:F990:5C10:A535:8952:E94D (talk) 15:53, 19 January 2025 (UTC)

Well, that's the first thing I tried, but I need the slope at (1,1) to be -1 always. Forgot to mention that; I'll correct it above. Varying the exponent doesn't give me that, it moves the knee of the curve off (1,1). ~Anachronist (talk) 17:27, 19 January 2025 (UTC)

If your threshold is set to ${\displaystyle 1,}$ the precise function values for ${\displaystyle x<1}$ are immaterial, as long as they are at least as large as the threshold. So only the shape of the tail for ${\displaystyle x>1}$ is relevant. To get long-distance connections, this tail should be fat.

The notion of "knee" is not very useful IMO. As ${\displaystyle \alpha \downarrow 0,}$ the fatness of the tail of ${\displaystyle f(x)=x^{{-}\alpha }}$ increases. The slope of the graph of ${\displaystyle f(x)}$ at ${\displaystyle x=1}$ equals ${\displaystyle f'(1)=-\alpha ,}$ so when ${\displaystyle \alpha }$ approaches ${\displaystyle 0}$ the graph becomes increasingly horizontal in that neighbourhood, ultimately just below ${\displaystyle y=1.}$ A side effect of fat tails is that two blobs, approaching each other, will start sprouting "feelers" towards each other well before these turn into a connection, and more so with a shallow slope. I suppose this is undesirable. It is possible to keep a steeper slope like ${\displaystyle {-}1}$ while having a fat tail, but then a more intricate function definition will be needed.

Our article mentions the criterion function ${\displaystyle \textstyle \sum _{i}{\mbox{metaball}}_{i}(x,y,z).}$ Consider two blobs A and B that are not far apart, but too far to have a connection. Now, if a third blob moves toward the area in the middle between A and B, its contribution to the criterion function may cause it to locally exceed the threshold. This is especially likely when you have fat tails. If this is unwanted, a better criterion may be

${\displaystyle \max _{i\neq j}~({\mbox{metaball}}_{i}(x,y,z)+{\mbox{metaball}}_{j}(x,y,z))\geq {\mbox{threshold}}.}$

Can you reveal at which distance (between their centres) two blobs should become connected? Then we can (perhaps) give a better response. Even better, a series of sketches of blobs coming increasingly closer and becoming distended before mating. Also, can you control the criterion function or only the threshold and individual metaball functions?

 --Lambiam 17:51, 19 January 2025 (UTC)

Because I'm writing the code myself (this is an OpenSCAD project), I have control over everything, so I can make any criterion function with any number of inputs. I already completed the marching cube isosurface algorithm and it's working well. Here's an example of metaballs I made, and an example of a manifold thick gyroid surface, which I used to test my isosurface module.

The notion of the "knee" is necessary to establish a threshold higher than 1 that still maintains a reasonable ball radius without clipping it to some minimum value, while at the same time having a fat tail at x>1 that allows two balls to combine at long distances without the ball radius growing too much. The function ${\displaystyle f(x)=x^{-a}}$ causes the diameter of the ball to shrink too far when isolated from other balls, and grow too fast when in proximity to other balls. That's why I'm looking for a function that would be, at the extreme, nearly vertical at x<1 and nearly horizontal at x>1 (both legs connected by a small-radius knee), with each leg approaching the axis at the same rate.

Picture two balls, say with an approximate radius of 10 (give or take), 70 units apart, and connected by a long tendril, like two wads of bubblegum stretched far apart after being stuck together. That's kind of what I'm going for. ~Anachronist (talk) 23:38, 19 January 2025 (UTC)

The x-scale and y-scale have no a priori established relationship. All ascending exponential curves ${\displaystyle (y=a\exp(\lambda x)}$ with positive ${\displaystyle a}$ and ${\displaystyle \lambda )}$ are similar. You can define the "knee" of a curve as the point where the angle of the slope equals ${\displaystyle 45^{\circ },}$ but where this is depends on the ratio of the x- and y-scale, which is why I doubt the usefulness of the concept without an established relationship between these scales. You can kind of create a relationship by equating the threshold value (measured on the y-scale) with the radius of a typical solitary ball (measured on the x-scale). Then a slope of ${\displaystyle {-}1}$ means that a 1% increase of the ${\displaystyle {\text{metaball}}}$ function means a 1% increase in the radius of the ball.

How wide do you envisage the diameter of the tube at its thinnest? What should we see when these balls are 50 or 100 units apart? And what would you use as the threshold?  --Lambiam 01:51, 20 January 2025 (UTC)

I don't have any preconceived notions of the width of the tube at its thinnest, and I expect the balls to separate and join like regular metaballs, just at bigger distances. The answers depend on my experimentation now that I have finally found a function that does what I want:

${\displaystyle f(x)=\left(a\left(\max \left(x,d\right)-d\right)\right)^{-{\frac {1}{a}}},{\text{ where }}d=1-{\frac {1}{a}}}$

(on Desmos here). The vertical asymptote is always at ${\displaystyle x=d}$ and stays in the range ${\displaystyle 0<x<1}$, the horizontal asymptote is always ${\displaystyle y=0}$, the function always passes through (1,1) and the slope at (1,1) is always -1. Increasing ${\displaystyle a}$ sharpens the corner and thickens the tail. I didn't expect the vertical asymptote would need to move but it should work well for constraining the minimum radius of the metaball. ~Anachronist (talk) 02:45, 20 January 2025 (UTC)

...and, in practice, it turns out not much different than the usual metaball functions. Oh well. ~Anachronist (talk) 03:04, 20 January 2025 (UTC)

When I wrote "at its thinnest", I meant for the case of two balls with a radius of 10 units having their centres 70 units apart. Is it more like 5 units or more like half a unit?  --Lambiam 10:50, 20 January 2025 (UTC)

It doesn't matter to me as long as the connection is thinner than the ball on each end, and the connection exists. ~Anachronist (talk) 23:14, 20 January 2025 (UTC)

January 23

• Misplaced Pages resources for researchers
• Misplaced Pages help forums
• Misplaced Pages reference desk