4

Context: This question is meant as an addendum to my question Does the sequence $1, 2, 3, 4, 5, 6$ appear in the number of groups of order $n$ up to isomorphism? which I asked a while ago.

That question asked whether or not we could find an integer $n$ such that

$$\mathrm{gnu}(n+i) = i \quad i=1,...,6$$

where $\mathrm{gnu}(n)$ denotes the number of groups of order $n$ up to isomorphism.

It turns out you can, and I found $n = 29436120$ works.

I brute-forced a search for $n$ to find that, but what I didn't mention is I substantially reduced the possible $n$ I had to search by noting that $n$ has to be a multiple of $24$.

In particular, if

$$\mathrm{gnu}(n+i) = i \quad i=1,...,4$$

holds, then $24 \mid n$.

When I first submitted A373650 to the OEIS, which is the sequence containing the indexes of all the $1,2,3,4$s, someone named Hugo Pfoertner observed that all the terms I submitted are divisible by $24$. I'd only had that each term must be even at that point, so this sped up the search substantially. While I set up the search using this fact, I looked for a proof that this is true.

It is true, but I've never written up the proof in full anywhere. I thought would be a good idea to put it here, for completeness of the other question. Everything used in the proof is fairly elementary, but I think the steps in the proof are quite nice. Namely, we observe $2 \mid n$, then $4 \mid n$, then $8 \mid n$, then $3 \mid n$ successively, with each one of those observations hinging on the previous.

In fact, it's also possible to show:

  • $48 \not \mid n$ for any $n$
  • we can get further conditions on $n$ modulo various prime numbers, such as $5,7,11,13$

but that's a story for another time.

Robin
  • 6,383

2 Answers2

5

First, we'll need a few lemmas:

Lemma 1: If $a$, $b$ are coprime, then

$$\mathrm{gnu}(ab) \geq \mathrm{gnu}(a) \mathrm{gnu}(b).$$

Sketch of proof: Let $G_1, \dots G_n$ be nonisomorphic groups of order $a$, $H_1, \dots, H_m$, be nonisomoprhic groups of order $b$. Then we can show $(G_i \times H_j)_{i=1,...,n}^{j=1,...,m}$ are all nonisomorphic due to the coprimality of $a$ and $b$.

In fact, it is conjectured that the above holds for all $a, b$.

Lemma 2: For $p$ prime, $q$ coprime to $p$, $m > 1$, $\mathrm{gnu}(p^mq) \geq \mathrm{gnu}(p^{m-1}q)$. In particular, $\mathrm{gnu}(2^m) \geq 5$ for all $m \geq 3$.

Sketch of proof: Can be proved inductively by showing that the groups of the previous prime power order, direct product the cyclic group of order $p$, are all non-isomorphic. Second statement follows immediately after observing $\mathrm{gnu}(8)=5$.

Lemma 3: For $p$ an odd prime, $\mathrm{gnu}(4p) \geq 4$.

Sketch of proof: One can show $C_{4p}$, $C_{2} \times C_2 \times C_p$, $D_{4p}$, $\mathrm{Dic}_{4p}$ are all non-isomorphic, where $C_m$ is the cyclic group of order $m$, $D_m$ is the dihedral group of order $m$, $\mathrm{Dic}_m$ is the dicyclic group of order $m$.

Lemma 4: For $p$ an odd prime, $t \geq 1$, $\mathrm{gnu}(4p^t) \geq 4$.

Proof: If $t = 1$, this is just lemma 3. If $t \geq 1$,

$$ \begin{align} \mathrm{gnu}(4p^t) &\geq \mathrm{gnu}(4) \mathrm{gnu}(p^t) && \text{lemma 1}\\ &\geq \mathrm{gnu}(4) \mathrm{gnu}(p^2) && \text{lemma 2}\\ &\geq 2 \cdot 2 \\ &\geq 4. \end{align} $$


Now, we will successively show that $2 \mid n$, then $4 \mid n$, then $8 \mid n$, then $3 \mid n$.


$2 \mid n$:

In fact, this shows that if

$$\mathrm{gnu}(n+i) = i \quad i=1,...,2$$

holds, then $2 \mid n$.

Since $\mathrm{gnu}(n+1) = 1$, we have that either $n+1=2$ or $n+1$ is odd, as there exists a dihedral group of order $2m$ for all $m>2$, and $\mathrm{gnu}(4) = 2$.

But if $n=1$, $\mathrm{gnu}(n+2) = \mathrm{gnu}(3) = 1$.

So $n$ is even.


$4 \mid n$:

In fact, this shows that if

$$\mathrm{gnu}(n+i) = i \quad i=1,...,3$$

holds, then $4 \mid n$.

We already have $2 \mid n$, so suppose for contradiction $n \equiv 2 \mod 4$. Then $n + 2 \equiv 0 \mod 4$.

Write $n = 2^rs$ where $s$ is odd. We have $r \geq 2$. If $r \geq 3$

$$\begin{align} \mathrm{gnu}(n+2) &= \mathrm{gnu}(2^r s) \\ &\geq \mathrm{gnu}(2^r)\mathrm{gnu}(s) && \text{lemma 1} \\ &\geq \mathrm{gnu}(2^r) \\ &\geq 5 && \text{lemma 2} \\ \end{align}$$

which is a contradiction. $(\ast)$

So $r = 2$. As $m$ is odd, either $m=1$ or there is some odd prime $p \mid m$. Let $p^t$ be the largest power of $p$ dividing $m$, with $t \geq 1$. Then $n+2 = 4p^t u$ with $u$ coprime to $4, p^t$ and hence

$$\begin{align} \mathrm{gnu}(n+2) &= \mathrm{gnu}(4p^t u) \\ &\geq \mathrm{gnu}(4p^t) \mathrm{gnu}(u) && \text{lemma 1} \\ &\geq \mathrm{gnu}(4p^t) \\ &\geq 4 && \text{lemma 4} \end{align}$$

which is also a contradiction.

So, $4 \mid n$.


$8 \mid n$:

We have $4 \mid n$. Suppose $n = 4m$, for $m$ odd. Then $8 \mid n+4$. It follows similarly to how we previously argued at $(\ast)$ that $\mathrm{gnu}(n+4) \geq 5$, a contradiction. Hence $8 \mid n$.


$3 \mid n$:

We have $n = 8m$. Suppose $3 \not \mid m$ for contradiction.

Then $n+4 = 4(2m+1)$. $2m+1 \not \equiv 1 \mod 3$, as $m \not \equiv 0 \mod 3$.

Hence we can rewrite $n+4 = 4r$, $r$ odd, $r \not \equiv 1 \mod 3$.

If $r \equiv 0 \mod 3$, $12 \mid n+4$. Similarly to how we have previously argued using lemma 1 and lemma 2, we can show $\mathrm{gnu}(n+4) \geq \mathrm{gnu}(12) = 5$, a contradiction.

If $r \equiv 2 \mod 3$, then $n+4 = 4(3s+2) = 12s+8$ for some $s$.

But $n = 12s+4$, and $8 \mid n$, so we require $s = 2t+1$ is odd.

Hence $n = 12(2t+1)+4 = 24t+16$.

So $n+2 = 24t+18 = 6(4t+3)$.

If $3 \mid t$, then $18 \mid n+2$. Again we can argue with lemma 1 and lemma 2 that $\mathrm{gnu}(n+2) \geq \mathrm{gnu}(18) = 5$, a contradiction.

Otherwise, we have $n+2 = 6u$, for some $u$ with $2, 3 \not \mid u$. So there is a prime $p$ such that $6p \mid n+2$. Yet again, we argue with lemma 1 and 2 that $\mathrm{gnu}(n+2) \geq \mathrm{gnu}(6p)$.

But then, $C_2 \times C_3 \times C_p$, $C_3 \times D_{2p}$, $C_p \times D_6$, $D_{6p}$ can all be shown to be non-isomorphic groups of order $6p$. Thus, $\mathrm{gnu}(n+2) \geq 4$, another contradiction.

Hence, we have $3 \mid n$.


As $3 \mid n$ and $8 \mid n$, we have $24 \mid n$ as required.

Robin
  • 6,383
2

Here are some extensions, unfortunately I just used inelegant case bashing:

$16 \not\mid n$:

If $n$ is a multiple of $16$, we have $n+4 = 16m+4 = 4(4m+1)$. But per Miller 1932 "Orders for which there exist exactly four or five groups", $\mathrm{gnu}(n)=4$ in exactly six cases (I paraphrased these because the paper is genuinely horrible):

  • I: $n$ is squarefree and there are exactly two relations, $p_i\mid p_k-1$ and $p_j\mid p_k-1$
  • II: $n$ is squarefree and there are exactly two relations, $p_i\mid p_j-1$ and $p_x\mid p_y-1$
  • III: $n=2pq$ where $p$ and $q$ are not related (2 is always related to any odd prime)
  • IV: $n$ is squarefree aside from two square factors $p_1$ and $p_2$, ie $n=p_1^2p_2^2m$ with $m$ squarefree, there are no relations, and $p_1+1$ and $p_2+1$ are not divisible by any divisor of $n$
  • V: $n$ is squarefree aside from one square factor $p_1$, and there is exactly one relation $p_1\mid p_i-1$, and $p_1^2$ does not divide $p_i-1$, and $p_1 + 1$ is not divisible by any divisor of $n$
  • VI: $n$ is squarefree aside from one square factor $p_1$, and we have exactly one relation $p_i\mid p_j-1$, and $p_1 + 1$ is not divisible by any divisor of $n$

(in any case, $p_1$, $p_i$, $p_j$, etc are distinct)

Since $n+4=4(4m+1)$, we must be in a non-squarefree case. In case IV, we must have no relations, but $2$ is related to any odd prime, so we cannot be in case IV. In case V, there is one relation iff $4m+1$ is prime. But then $4m+1-1$ is divisible by $4$, which violates the next condition. Finally, case VI is impossible because if we have two primes $p_i$ and $p_j$ which are related but not $2$, that can't be the only relation because $2$ is also related to both of these primes.

So $16\not\mid n$.

$n$ is $0$, $2$, or $4\mod 5$:

Based on the mod $48$ restriction we have so far, $n=24\mod 48$, so we will just case bash the other residues mod $5$ until they go away. Thus, consider $n=24,\,168,\,\mathrm{or}\,216\mod 240$:

If $n=168\mod 240$, then $n+2=10(24m+17)$. But obviously, $24m+17$ has at least one prime factor besides $2$ and $5$, so let $24m+17 = pq$. The $5$ part is not totally obvious, but if $24m+17$ is a power of $5$, it's at least $5^2$, so we would have $n+2$ divisible by $5^3$, pushing $\mathrm{gnu}(n+2)$ to at least $5$.

Then $\mathrm{gnu}(10pq)\ge\mathrm{gnu}(10p)$.

By Holder's formula (see eg the Mahmoud paper linked in the next case), we have $\mathrm{gnu}(10p)\ge3$. So in any case, $\mathrm{gnu}(n+2)>2$ when $n=168\mod 240$.

If $n=216\mod 240$, then $n+4=20(12m+11)$, so (unless $12m+11$ is a power of $5$ at least $5^2$, in which case $n+4$ is divisible by $5^3$ and $\mathrm{gnu}(n+4)$ is at least $5$) $n+4$ is a multiple of $20p$ for some prime $p$ besides $2$ and $5$. But without even trying, we can invent more than $4$ groups of order $20p$: $C_4\times C_5\times C_p$, $C_2\times C_2\times C_5\times C_p$, $C_2\times D_{10}\times C_p$, $C_2\times C_5\times D_{2p}$, and $C_2\times D_{10p}$.

More extensions for $\mathrm{gnu}(n+5)=5$ and $\mathrm{gnu}(n+6)=6$:

$n$ is $0$ or $2\mod 5$:

We already showed that $n$ can't be $1$ or $3$ mod $5$ when we have a sequence of just $4$ consecutive group numbers starting at $1$, so we only have to cover the $4$ case. But $n=4\mod 5$ corresponds to $n=24\mod 240$, where $n+6=30(8m+1)$. Thus $\mathrm{gnu}(n+6)\ge\mathrm{gnu}(30p)$ (aside from if $8m+1$ is a power of $5$, but that also does not work), and $\mathrm{gnu}(30p)\ge8$ by Holder's formula, so this is impossible.

$9\mid n$:

Now I will use Mahmoud 2024, Orders for which there exist exactly six or seven groups.

This paper explains the conditions for "cyclic-free" $n$, which means in the following table, when we say "$n=2pq$" for example, we mean "$n=2pqm$ where $m$ is squarefree and has no relations (ie, if $a$ is a prime dividing the mentioned part, and $b$ is a prime dividing $m$, then neither $a\mid b-1$ nor $b\mid a-1$) (this includes half relations, that is, we should not have $b\mid a+1$ if $a^2$ divides the mentioned part, and we should not have $b\mid a^2+a+1$ if $a^3$ divides the mentioned part)".

Also within this table, $s$, $p$, $q$, and $r$ are distinct odd primes, and there are NO relations (as described in my comment above) other than those mentioned.

$\mathrm{gnu}(n)=6$ iff one of the following:

  • I: $n=pqrs$ where $q\mid r-1$, $s\mid r-1$, and $p\mid q-1$
  • II: $n=2pq$ where $p\mid q-1$
  • III: $n=3p^2$ where $p$ is not $3$ and $3\mid p-1$
  • IV: $n=p^3q$ where $q\mid p^2+p+1$
  • V.a: $n=p^2qr$ where $p\mid\mid q-1$ (that is, $p$ divides but $p^2$ does not divide) and $q\mid r-1$
  • V.b: $n=p^2qr$ where $q=3$, $3\mid p+1$, and $3\mid r-1$ (the paper omits this last condition due to a typo)
  • V.c: $n=p^2qr$ where $p^2\mid r-1$ and $q\mid p+1$
  • VI: $n=p^2q^2$ where $p\mid\mid q+1$
  • VII: $n=n_1n_2$ where $n_1$ and $n_2$ are arithmetically independent (have no relations as described above) and each has two independent choices:
    • a: $n_1=q_1r_1$ where $q_1\mid r_1-1$; OR $n_1=q_1^2$
    • b: $n_2=p_2q_2r_2$ where $q_2\mid r_2-1$ and $p_1\mid q_1-1$; OR $n_2=p_2^2q_2$ and $q_2\mid p_2+1$

(I refer to the choices left, left as "VII.1"; left, right as "VII.2"; right, left as "VII.3"; and right, right as "VII.4")

Ok, so now we are ready to eliminate $9\not\mid n$: since we know $n$ mod $48$, consider $n=144m+24$ or $n=144m+120$. Then $n+6$ is $6(24m+5)$ or $18(8m+7)$.

But thanks to Mahmoud's conditions, if $n+6$ is even and $\mathrm{gnu}(n+6)=6$, then ONLY case II is possible. But even more than that, in this particular case, $n+6$ CANNOT have any other prime factors (since $2$ would be related to them), so we have $6(24m+5)$ or $18(8m+7)$. In the latter case, we don't have a squarefree number so it can't be $2pq$. In the former case, we have $p=3$ and $q=24m+5$ is prime, but now $3\mid 24m+5-1$ is impossible, so $n$ cannot be $3$ or $6$ mod $9$, it MUST be divisible by $9$.

$n=6(p-1)$, where $p$ is a prime of the form $24m+13$:

We basically noticed this when analyzing $9\mid n$: since $n=72\mod 144$, $n+6=6(24m+13)$, but this can only match case II and only if $24m+13$ is prime.

Additional Comments:

I've been exploring this following your linked question about sequences where $\mathrm{gnu}(n+1)=1$ thru $\mathrm{gnu}(n+k)=k$. I did not know this modulo restriction, so I just wrote a multithreaded segmented sieve to find all $n$ with $3$ groups up to $10^{12}$, which took 160 minutes to run on my computer, and then wrote filters for $\mathrm{gnu}(n)=1$ thru $7$ following this paper for 1-3 and the above papers for 4-5 and 6-7.

I found 1809 sequences with 1-6, but none of these extend to 1-7.

My guess is that 1-7 sequences do exist, but looking at the asymptotic proportions of k-group numbers:

  • 1: 1/3.273
  • 2: 1/6.916
  • 3: 1/297.988
  • 4: 1/9.035
  • 5: 1/24.565
  • 6: 1/45.490
  • 7: 1/250.591

3 and 7 are both significantly rarer than other small numbers.

Finally, looking at the gcd of $n$ for all the 1-6 sequences I found

  • $n=8\mod 16$ is the highest mod $2^k$ restriction that seems to arise, and we have proved that
  • $n=0\mod 9$ is the highest mod $3^k$ restriction
  • for ALL greater primes $p$ from $5$ thru $137$, $p-3$ residues are permissible. This is a really weird pattern that probably keeps going further if you have enough data.

I'll try to extend the sieve to $10^{14}$ or so. The fact we only have 2 possible residues mod 720 in theory offers a 360x speedup so 100x'ing the upper bound SHOULD be possible, but it needs significant work.

Edit 1:

$n+2=2(72m+37)$ where $72m+37$ is prime, and $n+4=4(36m+19)$ where $35m+19$ is prime:

If we consider $n=144m+72$ and then inspect $n+2$ and $n+4$ to see which $\mathrm{gnu}(n+2)=2$ and $\mathrm{gnu}(n+4)=4$ cases are actually possible, we see that the ONLY cases are $2p$ and $4p$ respectively, like how the only possible case for $\mathrm{gnu}(n+6)=6$ was $n+6=6p$. We don't even really need the full conditions to prove this, once we have that $n+2$ etc are divisible by certain powers of $2$, since even numbers have their group number pushed up exponentially the more odd prime factors they have.

This explains why there are $3$ bad residues mod every prime $p>3$ that $n$ cannot have:

After showing that $n=144m+72$ where $24m+13$, $72m+37$, and $36m+19$ are all prime, this is completely trivial. If $p$ divides any of these three numbers, then one of them is a multiple of $p$, which is only allowed if it EQUALS $p$. But we can multiply these out to have the same coeff, so we need $72m+39$, $72m+37$, and $72m+38$ to all not be multiples of $p$ (or one of them could equal $p$). $72$ is invertible mod $p$, so this shows there are $3$ bad residues mod $p$ as desired.

hacatu
  • 859