I have been investigating a procedure for rearranging the natural numbers called “THROWBACK”. This procedure was described in the article “Coding Fun: Rearranging All The Numbers”, published in Popular Computing #55, Oct. 1977, Vol. 5, No. 10.^{1} A partial copy of the article is available from https://oeis.org/A155167/a155167.pdf.

Here is the THROWBACK procedure: Start with the infinite sequence of natural numbers beginning with 3, that is 3, 4, 5, 6, 7, 8, … The first number in the sequence at any step of the procedure is called the “leader”. At each step, the leader is moved back in the sequence the number of places equal to its value. So, in the first step, 3 is moved back 3 places giving a new sequence 4, 5, 6, 3, 7, 8, … (See listing 1 for the results of the first 26 steps). This simple procedure turns out to produce a wealth of interesting number sequences, patterns, and questions.

For example, one can ask whether every number eventually becomes the leader. The Popular Computing article claims that the answer is yes and asks for the sequence of numbers giving the first step at which each number becomes the leader. (I will call this the “first indices sequence”). This sequence begins 1, 2, 3, 5, 7, 10, 14, 19, 26 and — as noted by N. J. A. Sloane — appears to match sequence A155167 in the Online Encyclopedia of Integer Sequences.^{2}

#### Listing 1. The first 26 steps of THROWBACK

```
Step Sequence after step N
---- ---------------------
0 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, ...
1 4, 5, 6, 3, 7, 8, 9, 10, 11, 12, 13, 14, 15, ...
2 5, 6, 3, 7, 4, 8, 9, 10, 11, 12, 13, 14, 15, ...
3 6, 3, 7, 4, 8, 5, 9, 10, 11, 12, 13, 14, 15, ...
4 3, 7, 4, 8, 5, 9, 6, 10, 11, 12, 13, 14, 15, ...
5 7, 4, 8, 3, 5, 9, 6, 10, 11, 12, 13, 14, 15, ...
6 4, 8, 3, 5, 9, 6, 10, 7, 11, 12, 13, 14, 15, ...
7 8, 3, 5, 9, 4, 6, 10, 7, 11, 12, 13, 14, 15, ...
8 3, 5, 9, 4, 6, 10, 7, 11, 8, 12, 13, 14, 15, ...
9 5, 9, 4, 3, 6, 10, 7, 11, 8, 12, 13, 14, 15, ...
10 9, 4, 3, 6, 10, 5, 7, 11, 8, 12, 13, 14, 15, ...
11 4, 3, 6, 10, 5, 7, 11, 8, 12, 9, 13, 14, 15, ...
12 3, 6, 10, 5, 4, 7, 11, 8, 12, 9, 13, 14, 15, ...
13 6, 10, 5, 3, 4, 7, 11, 8, 12, 9, 13, 14, 15, ...
14 10, 5, 3, 4, 7, 11, 6, 8, 12, 9, 13, 14, 15, ...
15 5, 3, 4, 7, 11, 6, 8, 12, 9, 13, 10, 14, 15, ...
16 3, 4, 7, 11, 6, 5, 8, 12, 9, 13, 10, 14, 15, ...
17 4, 7, 11, 3, 6, 5, 8, 12, 9, 13, 10, 14, 15, ...
18 7, 11, 3, 6, 4, 5, 8, 12, 9, 13, 10, 14, 15, ...
19 11, 3, 6, 4, 5, 8, 12, 7, 9, 13, 10, 14, 15, ...
20 3, 6, 4, 5, 8, 12, 7, 9, 13, 10, 14, 11, 15, ...
21 6, 4, 5, 3, 8, 12, 7, 9, 13, 10, 14, 11, 15, ...
22 4, 5, 3, 8, 12, 7, 6, 9, 13, 10, 14, 11, 15, ...
23 5, 3, 8, 12, 4, 7, 6, 9, 13, 10, 14, 11, 15, ...
24 3, 8, 12, 4, 7, 5, 6, 9, 13, 10, 14, 11, 15, ...
25 8, 12, 4, 3, 7, 5, 6, 9, 13, 10, 14, 11, 15, ...
26 12, 4, 3, 7, 5, 6, 9, 13, 8, 10, 14, 11, 15, ...
```

*Note: All of the Python code that I used in my investigations is available as a gist on GitHub: THROWBACK code. It is intended to be used mostly from the Python interactive prompt.*

## The Leader Sequence

The first sequence that I want to look at is the sequence of leaders at each step. Here are the first 200 values of the leader sequence:

```
3, 4, 5, 6, 3, 7, 4, 8, 3, 5, 9, 4, 3, 6, 10, 5, 3, 4, 7, 11, 3, 6, 4,
5, 3, 8, 12, 4, 3, 7, 5, 6, 3, 4, 9, 13, 3, 5, 4, 8, 3, 6, 7, 4, 3, 5,
10, 14, 3, 4, 6, 5, 3, 9, 4, 7, 3, 8, 5, 4, 3, 6, 11, 15, 3, 4, 5, 7, 3,
6, 4, 10, 3, 5, 8, 4, 3, 9, 6, 5, 3, 4, 7, 12, 3, 16, 4, 5, 3, 6, 8, 4,
3, 7, 5, 11, 3, 4, 6, 9, 3, 5, 4, 10, 3, 7, 6, 4, 3, 5, 8, 13, 3, 4, 17,
5, 3, 6, 4, 7, 3, 9, 5, 4, 3, 8, 6, 12, 3, 4, 5, 7, 3, 10, 4, 6, 3, 5,
11, 4, 3, 8, 7, 5, 3, 4, 6, 9, 3, 14, 4, 5, 3, 18, 6, 4, 3, 7, 5, 8, 3,
4, 10, 6, 3, 5, 4, 9, 3, 7, 13, 4, 3, 5, 6, 8, 3, 4, 11, 5, 3, 7, 4, 6,
3, 12, 5, 4, 3, 9, 8, 6, 3, 4, 5, 7, 3, 10, 4, 15, ...
```

The leader sequence does not appear directly in the OEIS but the Superseeker server at OEIS found connections to some other known sequences. If 2 is subtracted from every term of the leader sequence then the result appears to match sequence A087165 which is defined as “a(n) = 1 when n ≡ 1 (mod 4), otherwise a(n) = a(n – ceiling(n/4)) + 1”. The starting index for sequence A087165 is 1 whereas I am using 0 as the starting index for the leader sequence. So, the precise relationship is leader(n) – 2 = a(n+1). I have verified this relationship for the first 10,000,000 terms of each sequence.

By definition, every fourth term of sequence A087165 is 1. Correspondingly, every fourth term of the leader sequence is 3. This is fairly easy to prove since whenever 3 is the leader it is moved back to the fourth position and the three terms in front of it must all be larger than 3. Therefore, over the next three steps, those terms are moved back further in the sequence than the 3 and the 3 advances one position towards the front on each step so that on the fourth step it is again the leader. Another interesting property that the 3’s in the leader sequence should share with the 1’s in sequence A087165 is that if every 3 is removed then the result should be a sequence with every term 1 greater than the same term of the leader sequence.

```
3, 4, 5, 6, 3, 7, 4, 8, 3, 5, 9, 4, 3, 6, 10, 5, 3, 4, 7, 11, 3, 6, ...
4, 5, 6, 7, 4, 8, 5, 9, 4, 6, 10, 5, 4, 7, 11, 6, 4, 5, 8, 12, 4, 7, ...
```

The OEIS notes that the “indices of records” for sequence A087165, that is the locations in the sequence at which a new number greater than all previous terms appears, are the terms of sequence A087192. Since our “first indices sequence” for THROWBACK also gives the indices of new records in the leader sequence, it is not surprising that sequence A155167 appears to be related to sequence A087192. This (conjectured) relationship, which has not been recorded by OEIS contributors yet, is given by A155167(n) + 1 = A087192(n+1).

## The N-indices Sequences

Each of the other values in the leader sequence also appears to occur an infinite number of times and the intervals at which they occur show some fascinating patterns. I will define the “N-indices sequence” as the sequence of indices at which the value N occurs in the leader sequence. E.g. The 3-indices sequence is 0, 4, 8, 12, … Listing 2 gives the beginning of the N-indices sequences for N = 3 to 10.

#### Listing 2. The N-indices sequences for N = 3 to 10

```
N Indices at which N occurs
- -------------------------
3 0, 4, 8, 12, 16, 20, 24, 28, 32, 36, 40, 44, 48, 52, 56, ...
4 1, 6, 11, 17, 22, 27, 33, 38, 43, 49, 54, 59, 65, 70, 75, ...
5 2, 9, 15, 23, 30, 37, 45, 51, 58, 66, 73, 79, 87, 94, 101, ...
6 3, 13, 21, 31, 41, 50, 61, 69, 78, 89, 98, 106, 117, 126, 135, ...
7 5, 18, 29, 42, 55, 67, 82, 93, 105, 119, 131, 142, 157, 169, 181, ...
8 7, 25, 39, 57, 74, 90, 110, 125, 141, 159, 175, 190, 210, 226, 242, ...
9 10, 34, 53, 77, 99, 121, 147, 167, 189, 213, 234, 254, 281, 302, 323, ...
10 14, 46, 71, 103, 133, 162, 197, 223, 253, 285, 313, 339, 375, 403, 431, ...
```

Taking the differences between successive values in each of the N-indices sequences gives the intervals at which the value N occurs. For N > 3, these intervals are not constant as they are for N = 3, but the values of each sequence of intervals appear to be cyclic (i.e. periodic starting with the first term). (See listing 3). The amazing thing is that the period for the N-indices differences appears to be 3^{N-3}. I have confirmed these patterns for the first 1,000,000 terms of the leader sequence.

#### Listing 3. First differences of the N-indices sequences for N = 3 to 10

```
N Period Intervals at which N occurs
- ------ ---------------------------
3 1 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, 4, ...
4 3 5, 5, 6, 5, 5, 6, 5, 5, 6, 5, 5, 6, 5, 5, 6, 5, 5, 6, ...
5 9 7, 6, 8, 7, 7, 8, 6, 7, 8, 7, 6, 8, 7, 7, 8, 6, 7, 8, ...
6 27 10, 8, 10, 10, 9, 11, 8, 9, 11, 9, 8, 11, 9, 9, 11, 8, 9, 11, ...
7 81 13, 11, 13, 13, 12, 15, 11, 12, 14, 12, 11, 15, 12, 12, 14, 11, 12, 15, ...
8 243 18, 14, 18, 17, 16, 20, 15, 16, 18, 16, 15, 20, 16, 16, 19, 14, 16, 20, ...
9 729 24, 19, 24, 22, 22, 26, 20, 22, 24, 21, 20, 27, 21, 21, 26, 18, 22, 26, ...
10 2187 32, 25, 32, 30, 29, 35, 26, 30, 32, 28, 26, 36, 28, 28, 35, 24, 29, 35, ...
```

The values within a single period of the sequences with longer periods have some interesting patterns too. Take the simple example of N = 5. The repeating cycle is 7, 6, 8, 7, 7, 8, 6, 7, 8. Formatting the cycle in 3 columns reveals “subperiodic” patterns at the same indices modulo 3:

```
7 6 8
7 7 8
6 7 8
```

The 6-indices differences have period 27 and there are subperiodic patterns at indices modulo 9 and 3.

```
10 8 10 10 9 11 8 9 11
9 8 11 9 9 11 8 9 11
9 8 11 9 10 10 8 10 10
10 8 10
10 9 11
8 9 11
9 8 11
9 9 11
8 9 11
9 8 11
9 10 10
8 10 10
```

In the 9-column list above, the 2nd and 7th columns are all 8’s and these are the only columns that 8 occurs in. The 1st, 4th, 5th, and 8th columns contain only 9’s and 10’s and the 3rd, 6th, and 9th columns contain only 10’s and 11’s (also a modulo 3 pattern). In the 3-column list, the 1st column is the reverse of the 2nd.

Similar patterns for higher N-indices differences occur for subperiods of various powers of 3. The cycle of differences for N = 7 is arranged in various numbers of columns in listing 4. When I looked at the specific indices at which each value in this cycle occurs and took the differences of those indices (see listing 5), I finally noticed a global pattern that explains some of the other patterns: one cycle of the differences of each N-indices sequence forms a palindrome if the last value is omitted! I have confirmed this up to N = 13. You can also see that the indices differences within the cycle for some values (e.g. 11 and 12 for the 7-indices differences) are made up of smaller palindromes.

#### Listing 4. The 7-indices differences in 27, 9, and 3 columns

```
13 11 13 13 12 15 11 12 14 12 11 15 12 12 14 11 12 15 12 10 15 12 13 14 10 14 13
13 11 13 14 12 14 11 12 15 12 10 15 12 12 15 10 12 15 12 11 14 12 14 13 11 13 13
14 10 14 13 12 15 10 12 15 12 11 14 12 12 15 11 12 14 12 11 15 12 13 13 11 13 14
13 11 13 13 12 15 11 12 14
12 11 15 12 12 14 11 12 15
12 10 15 12 13 14 10 14 13
13 11 13 14 12 14 11 12 15
12 10 15 12 12 15 10 12 15
12 11 14 12 14 13 11 13 13
14 10 14 13 12 15 10 12 15
12 11 14 12 12 15 11 12 14
12 11 15 12 13 13 11 13 14
13 11 13
13 12 15
11 12 14
12 11 15
12 12 14
11 12 15
12 10 15
12 13 14
10 14 13
13 11 13
14 12 14
11 12 15
12 10 15
12 12 15
10 12 15
12 11 14
12 14 13
11 13 13
14 10 14
13 12 15
10 12 15
12 11 14
12 12 15
11 12 14
12 11 15
12 13 13
11 13 14
```

#### Listing 5. Indices of values within a single cycle of the 7-indices differences

```
Value Indices
----- -------
10 19, 24, 37, 42, 55, 60
11 1, 6, 10, 15, 28, 33, 46, 51, 64, 69, 73, 78
12 4, 7, 9, 12, 13, 16, 18, 21, 31, 34, 36, 39, 40, 43, 45, 48, 58, 61, 63, 66, 67, 70, 72, 75
13 0, 2, 3, 22, 26, 27, 29, 50, 52, 53, 57, 76, 77, 79
14 8, 14, 23, 25, 30, 32, 47, 49, 54, 56, 65, 71, 80
15 5, 11, 17, 20, 35, 38, 41, 44, 59, 62, 68, 74
Differences of the above indices:
Value Differences
----- -----------
10 5, 13, 5, 13, 5
11 5, 4, 5, 13, 5, 13, 5, 13, 5, 4, 5
12 3, 2, 3, 1, 3, 2, 3, 10, 3, 2, 3, 1, 3, 2, 3, 10, 3, 2, 3, 1, 3, 2, 3
13 2, 1, 19, 4, 1, 2, 21, 2, 1, 4, 19, 1, 2
14 6, 9, 2, 5, 2, 15, 2, 5, 2, 9, 6, 9
15 6, 6, 3, 15, 3, 3, 3, 15, 3, 6, 6
```

## Other Starting Sequences

One of the first questions I wondered about when reading the Popular Computing article was “why did they omit 1 and 2 from the starting list of numbers?” Is the THROWBACK procedure trivial or broken if 1 and/or 2 are included? When all natural numbers starting with 1 are used, the first indices sequence has a simple formula: a(n) = 2^{n-1} – 1. The leader sequence still has an interesting if simpler structure. Starting with 2, however, turns out to be just as rich as 3.

Here are the first 64 values in the leader sequence when the THROWBACK procedure is performed on all natural numbers starting with 1:

```
1, 2, 1, 3, 1, 2, 1, 4, 1, 2, 1, 3, 1, 2, 1, 5,
1, 2, 1, 3, 1, 2, 1, 4, 1, 2, 1, 3, 1, 2, 1, 6,
1, 2, 1, 3, 1, 2, 1, 4, 1, 2, 1, 3, 1, 2, 1, 5,
1, 2, 1, 3, 1, 2, 1, 4, 1, 2, 1, 3, 1, 2, 1, 7
```

This is a well-known sequence called the ruler function and can be defined as f(n) equals the exponent of the largest power of 2 that divides 2n. That is, f(n) is the largest integer such that 2^{f(n)} divides 2n. As stated before, each value N first appears in this leader sequence at the (2^{N-1}-1)-th step. After that, each value N appears at a constant interval of 2^{N}. The ruler function pops up in many different contexts and you can find out more from its entry on the OEIS as sequence A001511.

If the THROWBACK procedure is performed on all natural numbers starting with 2 then the first indices sequence begins

```
1, 2, 4, 7, 11, 17, 26, 40, 61, 92, 139, 209, 314, 472, 709, 1064, 1597,
2396, 3595, 5393, 8090, 12136, 18205, 27308, 40963, 61445, 92168,
138253, 207380, 311071, 466607, 699911, 1049867, 1574801, 2362202,
3543304, 5314957, 7972436
```

This appears to match sequence A006999 (excluding the leading 0) on the OEIS described there as “partitioning integers to avoid arithmetic progressions of length 3”. It is interesting to note that A006999 has a comment that “it appears that, aside from the first term, this is the (L)-sieve transform of … {2,5,8,11,…,3n+2….}”. Sequence A155167 — which matches the first indices sequence of THROWBACK(3, 4, 5, …) — is defined as the (L)-sieve transform of {3,7,11,15,…,4n-1,…}. So there appears to be a link between the THROWBACK procedure and the (L)-sieve transform of certain arithmetic progressions. (The (L)-sieve transform is defined at https://oeis.org/A152009).

What do the leader sequence and the N-indices sequences look like when we start with 2? If you guessed that the value 2 occurs every third term than you are catching on (see listing 6). The other N-indices sequences are similar to what we saw above but now their differences are nearly palindromic cycles with periods that are powers of 2. Based on further investigations of the THROWBACK procedure on the natural numbers starting with higher values, I conjecture that the periods of the N-indices sequences for THROWBACK(k, k+1, k+2, …) are k^{N-k}. And I have found potential matches for more of the first indices sequences on the OEIS (see listing 7).

#### Listing 6. The leader sequence for THROWBACK starting with 2

```
2, 3, 4, 2, 5, 3, 2, 6, 4, 2, 3, 7, 2, 5, 3, 2, 4, 8, 2, 3, 6, 2, 4, 3,
2, 5, 9, 2, 3, 4, 2, 7, 3, 2, 5, 4, 2, 3, 6, 2, 10, 3, 2, 4, 5, 2, 3, 8,
2, 4, 3, 2, 6, 5, 2, 3, 4, 2, 7, 3, 2, 11, 4, 2, 3, 5, 2, 6, 3, 2, 4, 9,
2, 3, 5, 2, 4, 3, 2, 7, 6, 2, 3, 4, 2, 5, 3, 2, 8, 4, 2, 3, 12, 2, 5, 3,
2, 4, 6, 2, ...
```

#### Listing 7. Sequences that match the first indices for starting numbers 1 to 9

```
THROWBACK(1, 2, 3, ...) See A000225.
THROWBACK(2, 3, 4, ...) See A006999.
THROWBACK(3, 4, 5, ...) See A155167.
THROWBACK(4, 5, 6, ...) See A279075.
THROWBACK(5, 6, 7, ...) See A279076.
THROWBACK(6, 7, 8, ...) See A279077.
THROWBACK(7, 8, 9, ...) See A279078.
THROWBACK(8, 9, 10, ...) See A279079.
THROWBACK(9, 10, 11, ...) See A279080.
```

All of these sequences also match the (L)-sieve transform of an arithmetic progression and an “Adar-based transformation” as mentioned in footnote no. 2 below.

I have only given a cursory look at what happens when the THROWBACK procedure is performed on starting sequences that are not just an ascending list of the natural numbers. The behavior seems similar for strictly increasing sequences such as all even numbers or the Fibonacci numbers (with only one leading 1). It is more difficult to simulate the procedure very far on sequences with large gaps. One interesting behavior that I noticed is that if you start with a reversed or randomized list of the first N natural numbers^{3}, then THROWBACK appears to slowly sort the initial values after which it behaves as described above!

## Next Steps: Proofs?

All of the patterns mentioned above are fascinating but most of them are just conjectures at this point. The proof that I sketched for the 3’s occurring every fourth term on the original THROWBACK leader sequence is the only proof that I have considered so far (although similar reasoning should apply for the lowest term in THROWBACK variants). What remains to be done is to find precise formulas or recurrences for the leader and N-indices sequences of THROWBACK with different starting sequences and to prove the various properties I’ve described as well as their identification with the OEIS sequences that they appear to match. Also, are there general relationships between the THROWBACK procedure, the (L)-sieve transform, and the “Adar-based transformation” to which I have alluded? While I hope to work on answering all of these questions, I don’t often get enough concentrated time right now to put my recreational math explorations on a more rigorous footing. Perhaps someone reading this will be able to fill in the gaps first?

#### Footnotes

^{1} There has been more than one publication with the name “Popular Computing” over the years. The one referenced here was published in the 1970’s by Audrey and Fred Gruenberger and appears to have featured monthly coding problems. I could not locate any digital archive of it and was only able to find a few excerpts online.

^{2} I came across sequence A155167 while investigating a new sequence transform based on the esoteric programming language Adar. The Adar-based transform provides a new way to generate numerous known sequences in the OEIS. Hopefully, I will find the time to say more about that in the future. Until then, if you interested you can wade through this Python code that gives numerous examples.

^{3} You need one or more extra values (“sentinels”) at the end of the sequence when starting with these kinds of input sequences.