12

$\underline{\text{Introduction}:-}$

For the last few days I've been wondering about the question below. I don't think that my approach is an elegant approach but this is the best I can do. (I am a high school student. But I have managed to gather a lot of information about advanced maths from the internet such as calculus, linear algebra, Set theory, number theory, coordinate geometry, modular arithmetic etc. So if the answer needs advanced maths then I don't have any problem, but not too advanced maths.)

Note: in this question I will include $0$ as a natural number, I know that this is rather controversial in the mathematical community but I will request not to debate about it the comment section. ($\mathbb{N}=\{0,1,2,\cdots\}$)


$\underline{\text{My Question}:-}$

If we put the natural numbers on the Ulam Spiral this wayenter image description here (Sorry I couldn't find a picture where the spiral starts at $0$, so I just put the wikipedia picture here. Just replace $n$ with $n-1$)

Then what is the formula for the cartesian coordinate on the integer lattice where each natural number lies?


$\underline{\text{My Attempt}:-}$

Define a bijective function $U:\mathbb{N}\rightarrow\mathbb{Z×Z}$

Where $U(n)$ gives the cartesian coordinates of $n$ on the ulam spiral.

For example:

$U(0)=(0,0)$

$U(1)=(0,1)$

$U(2)=(1,1)$

$U(3)=(1,0)$

$U(4)=(1,-1)$

$\vdots$

Let, $U(n)=(x_n,y_n)$

It's obvious that how $x_n$ and $x_{n+1}$ or $y_n$ and $y_{n+1}$ would compare to each other would depend on which part of the ulam spiral $n$ is in.

So, I divided the spiral into segments. I will denote them as $L_1$ and $L_2$ such that $|L_1|=L$ and $|L_2|=L+1$ for $L\in\mathbb{N}$ and define $0_2=\{0\}$. For example-

$0_1=\{\}=\varnothing$ $(\text{We won't consider this ever again, since it is just the null set})$

$0_2=\{0\}$

$1_1=\{1\}$

$1_2=\{2,3\}$

$2_1=\{4,5\}$

$2_2=\{6,7,8\}$

$\vdots$

So- enter image description here

(Please forgive me. I cannot create pictures, so I just drew it. But I hope you understand that what $L_1$ and $L_2$ are)

Here I tried to show how each segment looks like on the spiral.

A bit of investigations gives us the result-

$\begin{align}&L_1=\{L^2,\cdots,L^2+L-1\}\\&L_2=\{(L+1)^2-(L+1),\cdots,(L+1)^2-1\}\end{align}$

For $L\in\mathbb{N}$

It's obvious that-

●If $n\in L_1$ where $L$ is odd, then $x_{n+1}=x_n$ and $y_{n+1}=y_n+1$

●If $n\in L_2$ where $L$ is odd, then $x_{n+1}=x_n-1$ and $y_{n+1}=y_n$

●If $n\in L_1$ where $L$ is even, then $x_{n+1}=x_n$ and $y_{n+1}=y_n-1$

●If $n\in L_2$ where $L$ is even, then $x_{n+1}=x_n+1$ and $y_{n+1}=y_n$

So, $$U(n+1)=\begin{cases}(x_n,y_n+1),&\text{if $n\in L_1$ and $L$ is odd}\\(x_n,y_n-1),&\text{if $n\in L_1$ and $L$ is even}\\(x_n-1,y_n),&\text{if $n\in L_2$ and $L$ is odd}\\(x_n+1,y_n),&\text{if $n\in L_2$ and $L$ is even}\end{cases}$$

If we say that, $(a,b)+(x,y)=(a+x,b+y)$

Then, $$U(n+1)=\begin{cases}(x_n,y_n)+(0,1),&\text{if $n\in L_1$ and $L$ is odd}\\(x_n,y_n)-(0,1),&\text{if $n\in L_1$ and $L$ is even}\\(x_n,y_n)-(1,0),&\text{if $n\in L_2$ and $L$ is odd}\\(x_n,y_n)+(1,0),&\text{if $n\in L_2$ and $L$ is even}\end{cases}$$

And since,

$\begin{align}&(0,1)=U(1)\\&(1,0)=U(3)\\&(x_n,y_n)=U(n)\end{align}$

We have, $$U(n+1)=\begin{cases}U(n)+U(1),&\text{if $n\in L_1$ and $L$ is odd}\\U(n)-U(1),&\text{if $n\in L_1$ and $L$ is even}\\U(n)-U(3),&\text{if $n\in L_2$ and $L$ is odd}\\U(n)+U(3),&\text{if $n\in L_2$ and $L$ is even}\end{cases}$$

This is the worst recurrence relation I have ever seen.


Update:

To get the formula for $L$ and whether $n\in L_1$ or $n\in L_2$, we need to solve for $L$ and $k$ in-

$\begin{cases}L^2+k=n\\1<L<n\\-L\leq k\leq L-1\end{cases}$

With a little different definition for $L_1$ and $L_2$

The definition I used here will require solving for $L$ and $k$ in

$\begin{cases}L^2+k=n\\0\leq L<n\\0\leq k\leq 2L\end{cases}$

(for more details, read this question of mine)

The solutions would be:

$$\begin{align}&L=\left\lfloor\sqrt{n}\right\rfloor\\&k=n-\left\lfloor\sqrt{n}\right\rfloor^2\end{align}$$

(For more details, read this answer to the question I linked above)

It's clear that,

If $0\leq k\leq L-1$ then $n\in L_1$

If $L\leq k\leq 2L$ then $n\in L_2$

So we can write our recurrence formula as-

$$U(n+1)=\begin{cases}U(n)+U(1)&\text{if $0\leq k\leq L-1$ and $L$ is odd}\\U(n)-U(1)&\text{if $0\leq k\leq L-1$ and $L$ is even}\\U(n)-U(3)&\text{if $L\leq k\leq 2L$ and $L$ is odd}\\U(n)+U(3)&\text{if $L\leq k\leq 2L$ and $L$ is even}\end{cases}$$

$$\begin{align}U(n+1)=&\begin{cases}U(n)+U(1),&\text{if $0\leq n-\lfloor\sqrt{n}\rfloor^2\leq\lfloor\sqrt{n}\rfloor-1$ and $\lfloor\sqrt{n}\rfloor$ is odd}\\U(n)-U(1),&\text{if $0\leq n-\lfloor\sqrt{n}\rfloor^2\leq\lfloor\sqrt{n}\rfloor-1$ and $\lfloor\sqrt{n}\rfloor$ is even}\\U(n)-U(3),&\text{if $\lfloor\sqrt{n}\rfloor\leq n-\lfloor\sqrt{n}\rfloor^2\leq 2\lfloor\sqrt{n}\rfloor$ and $\lfloor\sqrt{n}\rfloor$ is odd}\\U(n)+U(3),&\text{if $\lfloor\sqrt{n}\rfloor\leq n-\lfloor\sqrt{n}\rfloor^2\leq 2\lfloor\sqrt{n}\rfloor$ and $\lfloor\sqrt{n}\rfloor$ is even}\end{cases}\end{align}$$

It made it worse, but at least now we have the recurrence relation only in terms of $n$.


$\underline{\text{Some more questions}:-}$

Besides my original question in the $\text{'My Question'}$ part, I have a few more questions.

$1)$How can I get further from my recurrence relation to derive an explicit formula?

$2)$If deriving a recurrence is really not the way to approach this problem then I would request some answer to question with a different approach.

Eric Wofsey
  • 330,363
  • 1
    Just to be extra clear about the coordinate system in your first picture, you want $U(1)=(0,0)$, $U(2)=(1,0)$ and $U(3)=(1,1)$? – Arthur Jul 16 '21 at 08:28
  • @Arthur. I made a mistake. Let me clear that out. – Rounak Sarkar Jul 16 '21 at 08:34
  • Ok, starting the natural numbers at $0$ or at $1$ isn't a big deal. My main point is, where in $\Bbb Z\times\Bbb Z$ do the first three natural numbers end up? – Arthur Jul 16 '21 at 08:55
  • @Arthur. The first one namely $0$ ends up at $(0,0)$, $1$ ends up at $(0,1)$, $2$ ends up at $(1,1)$. I didn't realized that I have touched a very controversial topic regarding the natural numbers. – Rounak Sarkar Jul 16 '21 at 09:44
  • 1
  • @TheSimpliFire. I think the post doesn't really answer my question. Even though it has the formula but the spiral is rotated $90$ degrees to the left. And a clear explanation is not really given, atleast not clear enough for a high school student like me to understand. – Rounak Sarkar Jul 18 '21 at 15:54
  • @RanjitKumarSarkar I agree there is absolutely no explanation to it. I don't have the time to give an elaboration within the bounty timeframe this week so I'll leave it to somebody else. – ə̷̶̸͇̘̜́̍͗̂̄︣͟ Jul 18 '21 at 15:56
  • 1
    Start at $0$ and go right-up-left-down, then $n=(2k+1)^2$ maps to $y=-k,x=k+1$; the $k$th odd square is mapped to a corner of the $k$th winding. Therefore, start by computing $k=\frac12(\sqrt{n}-1)$. If it is not an integer, you are somewhere after the $\lfloor k\rfloor$ winding but before $\lceil k\rceil$ winding, on one of the four sides of the winding. First horizontal side has $2k+1$ numbers preceding winding corner, so if $d_1=(2\lceil k\rceil+1)^2 - n$ is not more than $2k+1$, you are at $y=-k,x=k+1-d_1$. Otherwise, you can similarly find $d_2,d_3,d_4$ for other three sides. – Vepir Jul 18 '21 at 17:23
  • @Vepir. Can you please write an answer elaborating your point? – Rounak Sarkar Jul 18 '21 at 17:32

3 Answers3

9

As the spiral is winding around, we have four $90°$ turns before completing a full rotation. Notice that each rotation starts and ends with an odd square.

This is because, starting at $r^2-1$ and adding the four sides $4(r+1)$ to complete a rotation, gives $r^2+4r+3=(r+2)^2-1$, reaching the next odd square.

Therefore, odd squares $n=(2k+1)^2$ are mapped to $(y,x)=(-k, k+1)$.

Hence, given $n$ we start by computing $k=\frac12(\sqrt{n}-1)$ to determine in which rotation the given number is located in. That is, it is located between two odd squares $(2\lfloor k\rfloor+1)^2$ and $(2\lceil k\rceil+1)^2$ that determine the start and end of the rotation.

enter image description here

Now we simply backtrack from the end to the start of the rotation.

Given $n$, end of rotation is at $(-K, K+1)$, where $K=\left\lceil\frac12(\sqrt{n}-1)\right\rceil$.

Given $n$, distance from the end of the rotation is $d=(2K+1)^2 - n$.

This gives $U(n)=(y,x)$ as:

$$ U(n)= \begin{cases} (-K&,+K+1-d&), & 0K+0\le d\le 2K+1\\ (-3K-1+d&,-K&), & 2K+1\lt d\le 4K+1\\ (+K&,-5K-1+d&), & 4K+1\lt d\le 6K+1\\ (+7K+1-d&,+K&), & 6K+1\lt d\lt 8K+1\\ \end{cases} $$

which is as simple as it gets.

Verifying the formula in python, gives the expected result as in the picture:

36  35  34  33  32  31  30      
37  16  15  14  13  12  29      
38  17  4   3   2   11  28      
39  18  5   0   1   10  27      
40  19  6   7   8   9   26      
41  20  21  22  23  24  25      
42  43  44  45  46  47  48  49 
Vepir
  • 12,516
  • It's quite amazing that how a simple thing like ulam spiral gives so complicated formulas – Star Alpha Jul 19 '21 at 06:58
  • 1
    @JhumaSarkar I appreciate your amazment :) But, I'd say that complexity is subjective here, as at least for me, the formula feels quite simple, if you look at the idea behind it. That is, to get any coordinates, we simply subtract from the coordinates of the odd squares, which follow an easy pattern. And of course, since each rotation has four turning points, we have four cases in the formula. (We could combine everything into one line, but then I'd argue that such expression is unnecessarily complex.) – Vepir Jul 19 '21 at 08:09
  • @Vepir. I won't select your answer just yet, in case someone posts a better answer – Rounak Sarkar Jul 19 '21 at 13:05
  • @RounakSarkar That's expected. Accepting an answer will discourage others from answering. – Vepir Jul 20 '21 at 13:22
  • @Vepir. Why did the recurrence relation approach didn't work? – Rounak Sarkar Jul 21 '21 at 08:15
  • @RounakSarkar If you are asking about the recurrence you found, I didn't look into solving it. But, it looks like it works, your $L,k$ remind me of my $K,d$, but going another way around. It's just that instead of solving it, there are other better (easier) ways to find a direct (non-recursive) formula, e.g. the way I described in my comment (in this answer). – Vepir Jul 21 '21 at 10:16
4

We have the power set $\mathcal{P}(\Bbb Z \times \Bbb Z)$ and can define a transformation $\Gamma$ on the $3\text{-coordinate}$ set

$\tag 1 \mathcal U =\mathcal{P}(\Bbb Z \times \Bbb Z) \times (\Bbb Z \times \Bbb Z) \times \{-1,1\}$ by

$\quad \Gamma(S,(x,y),w))=$

$$ \begin{cases} \bigr(S \cup \{(x,y+1)\}&,(x,y+1)&,w\bigr), & w = -1 \land (x-1,y) \in S\\ \bigr(S \cup \{(x-1,y)\}&,(x-1,y)&,w\bigr), & w = -1 \land (x-1,y) \notin S \land (x-1,y-1) \in S\\ \bigr(S \cup \{(x-1,y)\}&,(x-1,y)&,-w\bigr), & w = -1 \land (x-1,y) \notin S \land (x-1,y-1) \notin S\\ \bigr(S \cup \{(x,y-1)\}&,(x,y-1)&,w\bigr), & w = 1 \land (x+1,y) \in S \\ \bigr(S \cup \{(x+1,y)\}&,(x+1,y)&,w\bigr), & w = 1 \land (x+1,y) \notin S \land (x+1,y+1) \in S\\ \bigr(S \cup \{(x+1,y)\}&,(x+1,y)&,-w\bigr), & w = 1 \land (x+1,y) \notin S \land (x+1,y+1) \notin S\\ \end{cases} $$

We have the projection mapping $\pi_2: \mathcal U \to \Bbb Z \times \Bbb Z$ onto the $2^{nd}$ coordinate.

The solution $U:\mathbb{N}\rightarrow\mathbb{Z×Z}$ is defined by

$\tag 2 U(n) = \pi_2 \circ \Gamma^{n}\big(\{(0,0)\},(0,0),1\bigr)\quad \text{for } n\ge 0$

CopyPasteIt
  • 11,366
  • Is this the general solution? – Rounak Sarkar Jul 20 '21 at 04:31
  • @RounakSarkar I would call it an elementary set theoretical solution. – CopyPasteIt Jul 20 '21 at 20:55
  • @Copypastelt. How did you found the definition of $\Gamma(S,(x,y),w)$? – Rounak Sarkar Jul 21 '21 at 04:32
  • @RounakSarkar I started out looking to define the function using set theory and some form or another of the recursion theorem. As I worked to put the pieces together I realized that a 'state machine' $\Gamma$ would work ($w = \pm1$ are the two states), with the orbit of $\big({(0,0)},(0,0),1\bigr)$ under $\Gamma$ describing the mapping for $U$. – CopyPasteIt Jul 21 '21 at 16:30
  • @RounakSarkar Note that once you see this recursion solution you can start working on a closed formulaic method. You might proceed by looking for the smallest square that hasn't been completed. Squares are odd or even, there are two states, and each state can be running horizontal or vertical; solving this your explicit solution will have an '$8$ logic breakdown' (Vespir's answer has '$4$ logic breakdown'). – CopyPasteIt Jul 22 '21 at 13:15
  • @RounakSarkar you might be able to use your work to provide an '8 logic breakdown' answer. – CopyPasteIt Jul 22 '21 at 13:18
  • @CopyPastelt. If you include this derivation within your answer, I think your answer will get more attention. – Rounak Sarkar Jul 23 '21 at 05:59
1

Using interval notation we can partition the natural numbers by writing

$\tag 1 \Bbb N = \displaystyle \bigsqcup_{k \in \Bbb N} \bigr[k^2,(k+1)^2\bigr)$

Let $\phi_k$ denote the restriction of the $\text{counter-clockwise Ulam 0-spiral}$ function $U$ to $[k^2,(k+1)^2\bigr)$.

If $k$ is even then $$ \quad \phi_k(n)= \begin{cases} \bigr(-\frac{k}{2}, \frac{k}{2} - (n-k^2) \bigr), & k^2 \le n \le k^2 +k\\ \bigr(-\frac{k}{2}+(n-k^2-k), \frac{k}{2} -k \bigr), & k^2 + k + 1 \le n \lt (k+1)^2\\\\ \end{cases} $$

If $k$ is odd then $$ \quad \phi_k(n)= \begin{cases} \bigr(\frac{k+1}{2}, \frac{1-k}{2} + (n-k^2) \bigr), & k^2 \le n \le k^2 +k\\ \bigr(\frac{k+1}{2}-(n-k^2-k), \frac{1-k}{2} + k \bigr), & k^2 + k + 1 \le n \lt (k+1)^2\\\\ \end{cases} $$

After some deliberation we can write

$\quad U(n) =$ $$ \small{ \begin{cases} & \text{when } \lfloor \sqrt n \rfloor \text{ is even:}\\ \bigr(-\frac{\lfloor \sqrt n \rfloor}{2}, \frac{\lfloor \sqrt n \rfloor}{2} - (n-\lfloor \sqrt n \rfloor^2) \bigr), & \lfloor \sqrt n \rfloor^2 \le n \le \lfloor \sqrt n \rfloor^2 +\lfloor \sqrt n \rfloor\\ \bigr(-\frac{\lfloor \sqrt n \rfloor}{2}+(n-\lfloor \sqrt n \rfloor^2-\lfloor \sqrt n \rfloor), \frac{\lfloor \sqrt n \rfloor}{2} -\lfloor \sqrt n \rfloor \bigr), & \lfloor \sqrt n \rfloor^2 + \lfloor \sqrt n \rfloor + 1 \le n \lt (\lfloor \sqrt n \rfloor+1)^2\\\\ & \text{when } \lfloor \sqrt n \rfloor \text{ is odd:}\\ \bigr(\frac{\lfloor \sqrt n \rfloor+1}{2}, \frac{1-\lfloor \sqrt n \rfloor}{2} + (n-\lfloor \sqrt n \rfloor^2) \bigr), & \lfloor \sqrt n \rfloor^2 \le n \le \lfloor \sqrt n \rfloor^2 +\lfloor \sqrt n \rfloor\\ \bigr(\frac{\lfloor \sqrt n \rfloor+1}{2}-(n-\lfloor \sqrt n \rfloor^2-\lfloor \sqrt n \rfloor), \frac{1-\lfloor \sqrt n \rfloor}{2} + \lfloor \sqrt n \rfloor \bigr), & \lfloor \sqrt n \rfloor^2 + \lfloor \sqrt n \rfloor + 1 \le n \lt (\lfloor \sqrt n \rfloor+1)^2\\\\\end{cases} } $$

CopyPasteIt
  • 11,366