83
$\begingroup$

I've been comparing the sequence of the Least Common Multiple of the first $n$ integers, $L_n = \text{lcm}(1, 2, \dots, n)$, with the sequence of Highly Abundant Numbers (HA).

The two sequences in question are:

A positive integer $N$ is Highly Abundant (HA) if the sum of its divisors, $\sigma(N)$, is strictly greater than the sum of the divisors of any smaller positive integer $m$: $$N \text{ is HA} \iff \sigma(N) > \sigma(m) \quad \text{for all } m \in \mathbb{Z}^+, m < N$$ My observation leads to the following conjecture:

Conjecture: Every number in the $\text{lcm}(1, 2, \dots, n)$ sequence is a Highly Abundant Number or in terms of sets: $$\{L_n\}_{n \ge 1} \subset \text{HA}$$

Is there a known theorem or a published proof that confirms that the sequence of Least Common Multiples of the first $n$ integers is a strict subset of the Highly Abundant Numbers?

$\endgroup$
21
  • 16
    $\begingroup$ I predict that this proposition in fact fails for all sufficiently large $n$. If one adds $k$ primes slightly larger than $n^{1/2}$ to $\mathrm{lcm}(1,2,\dots,n)$, and then takes away slightly more than $k/2$ primes slightly less than $n$, in such a way that the product of the primes added is very slightly smaller than the product of the primes taken away, one should end up with a quantity a little smaller than $\mathrm{lcm}(1,2,\dots,n)$ with a larger sum of divisors. It may be possible to locate a numerical example of this type by computer. $\endgroup$ Commented Oct 2, 2025 at 6:35
  • 8
    $\begingroup$ Sorry, I wasn't able to specify all the parameters within the comment. The $p_j$ will range between $n^{1/2}$ and $(1+\varepsilon) n^{1/2}$ for some not too small $\varepsilon$, and the $q_m$ between $(1-\varepsilon) n$ and $n$, introducing factors of $(1\pm O(\varepsilon))^n$ that can counteract the gap you mention. It's easier to see things multiplicatively rather than additively: the ratio between the products over $j$ is like $1 + (1+O(\varepsilon)) k/n$ and the ratio between the products over $m$ is like $1 + (1+O(\varepsilon)) \ell/n$, revealing the favorable gap. $\endgroup$ Commented Oct 2, 2025 at 14:46
  • 9
    $\begingroup$ In fact one could probably make this rigorous with some Fourier analysis and known bounds on the zeta function. In any case we have an explicit criterion for a counterexample: find distinct primes $n^{1/2} < p_j, q_m < n$ such that $0 < \sum_m \log q_m - \sum_j \log p_j < \sum_j \log(1+\frac{1}{p_j^2+p_j}) - \sum_m \log(1+\frac{1}{q_m})$. I think a computer search using primes $p_j$ near $n^{1/2}$ and $q_m$ near $n$ perhaps combined with some shortest lattice vector type methods can produce a very small positive value of $\sum_m \log q_m - \sum_j \log p_j$ that would work. $\endgroup$ Commented Oct 2, 2025 at 15:08
  • 20
    $\begingroup$ Everything about this story, including how it is unfolding, is wonderful. I would like to congratulate everyone involved, and in particular @283212123 for the bravery of the initial post (as a new user!) which has brought joy to so many people. I really hope this story is published (ideally with the backstory) so it can be shared by a wider audience for posterity. $\endgroup$ Commented Oct 5, 2025 at 19:12
  • 11
    $\begingroup$ I actually got around to reading the Alaoglu-Erdos paper at users.renyi.hu/~p_erdos/1944-03.pdf . On pages 466-467 they write "it would be easy to see that neither n! nor the multiple of the first n integers can be highly abundant infinitely often", but offer no proof. (I found this claim through an AI deep research tool, though I initially believed it to be a hallucination.) Strangely, though, I am almost glad that we didn't find this reference earlier, given the outcome! $\endgroup$ Commented Oct 6, 2025 at 15:16

10 Answers 10

39
$\begingroup$

Here is a new method designed to handle very large $n$ (combining ideas from several previous constructions) in a manner that is more quantitatively efficient than the existing method.

Theorem: Suppose that $$ \sqrt{n} < p_1 < p_2 < p_3 < q_1 < q_2 < q_3 < n$$ are primes obeying the condition $$ \prod_{i=1}^3 \left(1+\frac{1}{q_i}\right) \leq \left(\prod_{i=1}^3 \left(1+\frac{1}{p_i(p_i+1)}\right)\right) \left(1+\frac{3}{8n}\right) \left(1 - \frac{4p_1p_2p_3}{q_1q_2q_3}\right) \tag{1}\label{eq1}$$ (so in particular $4p_1p_2p_3 < q_1q_2q_3$). Then $L_n$ is not highly abundant.

Proof: Write $L_n = q_1 q_2 q_3 L'$, where $L'$ is not divisible by $q_1,q_2,q_3$. Then we have $$ \frac{\sigma(L_n)}{L_n} = \frac{\sigma(L')}{L'} \prod_{i=1}^3 \left(1+\frac{1}{q_i}\right).\tag{2}\label{eq2} $$ We also write $$ q_1 q_2 q_3 = 4p_1p_2p_3 m + r$$ for some natural number $m$ and remainder $0 < r < 4p_1p_2p_3$. The competitor $M$ to $L_n$ is then defined as $$ M := 4p_1p_2p_3 m L'.$$ Clearly $M < L_n$, and in fact $$ 1 < \frac{L_n}{M} = \left(1 - \frac{r}{q_1q_2q_3}\right)^{-1} < \left(1 - \frac{4p_1p_2p_3}{q_1q_2q_3}\right)^{-1}.$$ To show that $\sigma(M) \geq \sigma(L_n)$, it thus suffices to show that $$ \frac{\sigma(M)}{M} \left(1 - \frac{4p_1p_2p_3}{q_1q_2q_3}\right) \geq \frac{\sigma(L_n)}{L_n}.$$ By \eqref{eq1}, \eqref{eq2}, it suffices to show that $$ \frac{\sigma(M)}{M} \geq \frac{\sigma(L')}{L'} \left(\prod_{i=1}^3 \left(1+\frac{1}{p_i(p_i+1)}\right)\right) \left(1+\frac{3}{8n}\right).$$ But each prime $p_i$ already divided $L'$ once, so inserting an additional factor of $p_i$ increases $\frac{\sigma(L')}{L'}$ by a factor of $$ \frac{(1+p_i+p_i^2)/p_i^2}{(1+p_i)/p_i} = 1 + \frac{1}{p_i(p_i+1)}.$$ Next, if $2^k$ is the largest power of $2$ less than or equal to $n$, then $L'$ is divisible by $2^k$, and $M$ divisible by at least $2^{k+2}$. Hence the effect of the prime $2$ on $\frac{\sigma(L')}{L'}$ is at least $$ \frac{(1+2+\dots+2^{k+2})/2^{k+2}}{(1+2+\dots+2^k)/2^k} = 1 + \frac{3}{2^{k+3}-4}$$ $$ \geq 1 +\frac{3}{8n}.$$ The contribution of the remaining factors $m$ only serve to increase $\sigma(L')/L'$ further. The claim follows.

Remark By a similar method, the pair of parameters $(4, 3/8 = 0.375)$ can be replaced by $(2, 1/4 = 0.25)$, $(6, 17/36 = 0.472\dots)$, $(30,0.632\dots)$, etc.. One could also use $(1,0)$, which no longer is an asymptotic win; but if one takes five "p" primes $p_1,\dots,p_5$ and four "q" primes $q_1,\dots,q_4$ instead of three of each, one can restore an asymptotic win without having to perform any analysis at small primes.

For $n$ large, the prime number theorem lets us choose $p_1,p_2,p_3$ close to $\sqrt{n}$ and $q_1,q_2,q_3$ close to $n$. So the LHS behaves like $1+3/n$ and the RHS behaves like $1+3/n + 3/8n$, so this should win for large $n$. Indeed, from Proposition 5.4 of this paper of Dusart, for every $x \geq 89693$ there is a prime in the range $x < p \leq x (1 + \frac{1}{\log^3 x})$. So, for $n \geq 89693^2 \approx 8.04 \times 10^9$, we can choose the primes so that $$ p_i \leq \sqrt{n} \left(1 + \frac{1}{\log^3 \sqrt{n}}\right)^i$$ and $$ q_{4-i} \geq n \left(1 + \frac{1}{\log^3 \sqrt{n}}\right)^{-i}$$ so $$ \prod_{i=1}^3 \left(1 + \frac{1}{q_i}\right) \leq \prod_{i=1}^3 \left(1 + \frac{\left(1 + \frac{1}{\log^3 \sqrt{n}}\right)^i}{n} \right)$$ and (bounding $p_i+1 \leq (1+1/\sqrt{n}) p_i$) $$ \prod_{i=1}^3 \left(1 + \frac{1}{p_i (p_i+1)}\right) \geq \prod_{i=1}^3 \left(1 + \frac{1}{\left(1 + \frac{1}{\log^3 \sqrt{n}}\right)^{2i} (n + \sqrt{n})} \right)$$ and $$ 1 - \frac{4p_1p_2p_3}{q_1q_2q_3} \geq 1 - \frac{4 \left(1+\frac{1}{\log^3 \sqrt{n}} \right)^{12}}{n^{3/2}}$$ so this theorem should handle all $n \geq 8.04 \times 10^9$ for which $$ \prod_{i=1}^3 \left(1 + \frac{\left(1 + \frac{1}{\log^3 \sqrt{n}}\right)^i}{n} \right) \leq \prod_{i=1}^3 \left(1 + \frac{1}{\left(1 + \frac{1}{\log^3 \sqrt{n}}\right)^{2i} (n + \sqrt{n})} \right)$$ $$ \times \left(1 - \frac{4 \left(1+\frac{1}{\log^3 \sqrt{n}} \right)^{12}}{n^{3/2}}\right) \left(1+\frac{3}{8n}\right).$$ For $n \geq 89693^2$, we upper bound $$\frac{1}{\log^3 \sqrt{n}} \leq \frac{1}{\log^3 \sqrt{n}} (1+\frac{1}{\sqrt{n}}) \leq 0.000675$$ and $$ \frac{1}{n^{3/2}} \leq \frac{1}{89693} \frac{1}{n}$$ so on setting $\varepsilon = 1/n$, we reduce to verifying $$ \prod_{i=1}^3 \left(1 + 1.000675^i \varepsilon \right) \leq \prod_{i=1}^3 \left(1 + \frac{\varepsilon}{1.000675^{2i}} \right)$$ $$ \times \left(1 - \frac{4 \times 1.000675^{12}}{89693} \varepsilon\right) \left(1+\frac{3}{8}\varepsilon\right)$$ for $0 \leq \varepsilon \leq 1/89693^2$. But the RHS is at least $$ \left(1 + \sum_{i=1}^3 \frac{\varepsilon}{1.000675^{2i}} + \frac{3}{8}\varepsilon\right)\left(1- \frac{4 \times 1.000675^{12}}{89693} \varepsilon\right) \geq 1 + 3.37 \varepsilon-0.01 \varepsilon^2$$ and the LHS multiplies to at most $$ 1 + 3.01 \varepsilon + 3.01 \varepsilon^2 + 1.01 \varepsilon^3$$ and the inequality easily follows. Thus we can cover the range $n \geq 8.04 \times 10^9$.

EDIT: In case it is helpful, I am recording my thought process leading to the above construction. We had previously used constructions in which we removed many large primes (slightly less than $n$) and added many medium primes (slightly larger than $\sqrt{n}$), chosen so that the products closely matched. Eduardo, Saul and GH from MO also utilized a construction in which one large prime was removed, and many small primes (e.g., factors of 2310), together with additional surplus factors requiring no further analysis, were added. It seemed natural to try to combine the two approaches, but I found that if only one large prime was removed, then the approach of Eduardo, Saul and GH from MO was essentially optimal, and further forced an annoying congruence condition on the large prime which had significantly weakened the asymptotic results. I then tried removing two large primes, which did allow me to squeeze in a medium prime, but the asymptotic performance was about the same, except that the congruence condition was weakened. This eventually led me to the idea to try removing three large primes, and experimented with several combinations of small and medium primes, before finally realizing that the congruence condition could be dispensed with altogether by the ancient method of natural number division with remainder (once the product of the large primes was more than $n$ times as large as the product of the medium primes), leading to the above argument after some trial and error experimenting with different numbers of medium primes.

UPDATE, Jan 20 2026: The above argument has now been formalized in Lean, conditionally on the prime number theorem in short interval result of Dusart cited above, as part of the Integrated Explicit Analytic Number Theory Network. It is a future plan of this network to also be able to formalize results such as Dusart's as well.

$\endgroup$
22
  • 3
    $\begingroup$ A quick and dirty script suggests that taking the 3 smallest primes greater than $\sqrt n$ and the 3 largest primes less than $n$ handles every $n$ above $200,000$, which would complete the project if true. I am going to add some sanity checks to my code and switch it to use exact arithmetic, then publish the code + the updated bounds. Edit: sorry, this is slightly misstated. Rather I am again creating a "ladder", and I am able to make a new step of the ladder by choosing $n$ appropriately and the primes as described. $\endgroup$ Commented Oct 5, 2025 at 21:25
  • 5
    $\begingroup$ Here is a "ladder" covering from 200000 up to $10^{10}$. Probably someone else should confirm that these choices of primes meet the necessary conditions. gist.github.com/mjtb49/efa27d51ec3f193d51356c60b402a784 $\endgroup$ Commented Oct 5, 2025 at 21:52
  • 9
    $\begingroup$ Excellent! It would be good to check, but it seems tentatively that we can declare victory in the "beat the AI" challenge! $\endgroup$ Commented Oct 5, 2025 at 21:54
  • 3
    $\begingroup$ Why do all the fun developments happen when I'm away from my desk? Happened on EqTheories too.. anyway, I can confirm that @MatthewBolan's primes meet the conditions of the theorem, and the windows have no gaps. $\endgroup$ Commented Oct 5, 2025 at 22:14
  • 3
    $\begingroup$ Actually, I can do a little better than my previous comment might suggest. I think the 5 prime version covers everything greater than $20101$ except for entries in the interval $[39601,40759]$. $\endgroup$ Commented Oct 6, 2025 at 4:01
78
$\begingroup$

OK, after an hour-long conversation with an AI I was able to produce a numerical counterexample that I could verify separately. (Admittedly, the verification code was also provided by the AI (see the very end of the conversation), but it is only 29 lines long, and simple enough that I could read it line by line.)

Let $n = 200000$, and let $p_1,\dots,p_{25}$ and $q_1,\dots,q_{13}$ be the primes $$ 479, 487, 491, 499, 503, 509, 521, 523, 541, 547, 557, 563, 569, 571, 577, 587, 593, 599, 601, 607, 613, 617, 619, 631, 641$$ and $$ 187963, 187973, 187987, 188011, 188017, 188021, 188029, 188299, 199931, 199933, 199961, 199967, 199999$$ respectively. Thus the primes $p_j$ are slightly larger than $\sqrt{n} \approx 447$ and the primes $q_m$ are slightly less than $n$. (The conversation with the AI will describe how these primes and parameters such as $25$, $13$, and $200000$ were located.)

Direct calculation shows that $$ \sum_{m=1}^{13} \log q_m - \sum_{j=1}^{25} \log p_j \approx 1.8484\dots \times 10^{-6} \tag{1}\label{501125_1}$$ is positive, but smaller than $$ \sum_{j=1}^{25} \log(1 +\frac{1}{p_j^2+p_j}) - \sum_{m=1}^{13} \log(1+\frac{1}{q_m}) \approx 1.33489\dots \times 10^{-5}. \tag{2}\label{501125_2}$$ If we define the competitor $$ M := \frac{p_1 \dots p_{25}}{q_1 \dots q_{13}} L_n$$ to $L_n$, then by the positivity of \eqref{501125_1}, $M$ is (slightly) less than $L_n$. On the other hand, since the primes $p_1,\dots,p_{25},q_1,\dots,q_{13}$, being between $\sqrt{n}$ and $n$, all divide $L_n$ exactly once, we have $$ \frac{\sigma(M)}{\sigma(L_n)} = \frac{\prod_{j=1}^{25} (1+p_j+p_j^2)/(1+p_j)}{\prod_{m=1}^{13} (1+q_m)}$$ $$ = \exp\biggl( \eqref{501125_2} - \eqref{501125_1} \biggr). $$ Since \eqref{501125_2} is larger than \eqref{501125_1}, we thus see that $\sigma(M) > \sigma(L_n)$ and hence $L_n$ is not highly abundant.

EDIT: there is significant room in the analysis, and I could imagine that a more intelligent search (without relying as much on AI assistance) could reduce the value of $n$ significantly; perhaps those who are inclined towards such computational numerical challenges would like to see how well they can "beat the AI".

$\endgroup$
15
  • 18
    $\begingroup$ It is amusing to compare this conversation with Gowers's "imagined dialogue between a mathematician and a computer in two or three decades' time", written in 1999, on pp. 6-8 at dpmms.cam.ac.uk/~wtg10/gafavisions.ps $\endgroup$ Commented Oct 3, 2025 at 12:11
  • 10
    $\begingroup$ This ought to make it into eventual counterexamples $\endgroup$ Commented Oct 3, 2025 at 15:40
  • 15
    $\begingroup$ Terence Tao in partnership with a (paid?) AI... I am afraid that all remaining math problems will be solved (and I will be unemployed). $\endgroup$ Commented Oct 3, 2025 at 22:20
  • 5
    $\begingroup$ I had taken some screenshots of Gowers' imagined conversation in an 2023 Mathstodon post: mathstodon.xyz/@tao/110084623890144657 $\endgroup$ Commented Oct 4, 2025 at 16:55
  • 10
    $\begingroup$ This is only a very minor comment, but if I give GPT-5 on the API access to the Python sandbox tool but not to web search (in order to avoid polluting results with this discussion), set reasoning to high, and prompt it to find a counterexample to the conjecture given by OP, plus give it your three comments on the original posting, it correctly finds L_{97} as a counterexample all on its own, but hallucinates that this is the smallest counterexample (I'd have to check the chain-of-thought to see how it gets there). Output here: pastebin.com/mu5GbLcx $\endgroup$ Commented Oct 5, 2025 at 6:37
42
$\begingroup$

Third response. By improving on Saúl RM's nice argument, I will show below that if $n>\exp(\exp(7812))$, then $L_n$ is not highly abundant. Under GRH, this range can be extended to $n>\exp(15667)$. With little tweaks the constants $7812$ and $15667$ can be reduced by a factor of about $6$ (see the Remark after the Theorem).

Subsequently, Terry Tao found a better method, and as a result we know that if $n$ is a prime power, then $L_n$ is highly abundant if and only if $$n\in\{1, 2, 3, 4, 5, 7, 8, 9, 11, 13, 16, 17, 19, 23, 25, 27, 29, 31, 32, 37, 41, 43, 47, 49, 53, 59, 61, 64, 67, 81, 83, 89, 125, 127, 128, 131, 137, 139, 169\}.$$

Theorem. Assume that the prime number $p$ is such that $p-1\in[(400/401)n,n-1]$, and $p-1$ is divisible by the first thousand primes. Then $L_n$ is not highly abundant.

Proof. We shall prove that $$\sigma(L_n(p-1)/p)>\sigma(L_n).$$ Note that $p\in(\sqrt{n},n]$, hence $p\parallel L_n$. Let $p-1=\prod_j p_j^{s_j}$ be the prime factorization of $p-1$, and let $r_j$ be the exponent of $p_j$ in $L_n$. Then, $$\frac{\sigma(L_n(p-1)/p)}{\sigma(L_n)}=\frac{1}{p+1}\prod_j\frac{\sigma(p_j^{r_j+s_j})}{\sigma(p_j^{r_j})},$$ hence we need that $$\prod_j\frac{\sigma(p_j^{r_j+s_j})}{\sigma(p_j^{r_j})}>p+1.$$ Since $$\frac{\sigma(p_j^{r_j+s_j})}{\sigma(p_j^{r_j})}=p_j^{s_j}+\frac{p_j^{s_j}-1}{p_j^{r_j+1}-1}>p_j^{s_j}+\frac{p_j^{s_j}-1}{np_j},$$ it suffices that $$\prod_j\left(p_j^{s_j}+\frac{p_j^{s_j}-1}{np_j}\right)\geq p+1.$$ Dividing both sides by $p-1=\prod_j p^{s_j}$, this becomes $$\prod_j\left(1+\frac{1-p_j^{-s_j}}{np_j}\right)\geq 1+\frac{2}{p-1}.$$ For this it suffices that $$\sum_j\frac{1-p_j^{-s_j}}{np_j}\geq\frac{2}{p-1},$$ that is, $$\sum_j\frac{1-p_j^{-s_j}}{p_j}\geq\frac{2n}{p-1}.$$ The $p_j$'s on the left-hand side include the first thousand primes, hence the sum exceeds $401/200$ according to a numerical calculation, while the right-hand side does not exceed $401/200$. The proof is complete.

Remark. The crux of the above proof is that the sum in the last display exceeds $2$. For larger exponents $s_j$, one needs fewer primes $p_j$ for this to happen. For example, it would suffice to assume that $p-1$ is divisible by the square of the first $150$ primes, or the cube of the first $90$ primes, or the fourth power of the first $70$ primes, or the eight power of the first $60$ primes.

Corollary. If $n>\exp(\exp(7812))$, then $L_n$ is not highly abundant. If we assume the Riemann hypothesis for Dirichlet $L$-functions, then the above range can be extended to $n>\exp(15667)$.

Proof. Assume that $n>\exp(\exp(7812))$, and put $m:=n/401-1$. Let $q$ denote the product of the first thousand primes. Numerical calculation shows that $$q<\exp(7812.3)\qquad\text{and}\qquad\varphi(q)<\exp(7809.6).$$ By the Theorem above, it suffices to show that $$\theta(n-m;q,1)\neq\theta(n;q,1).$$ Assume that the two sides are equal. Then, by Theorem 1.2 of Bennett-Martin-O'Bryant-Rechnitzer (which appeared as Illinois J. Math. 62 (2018), 427–532), we conclude that $$\frac{m}{\varphi(q)}<\frac{n}{80\log m}.$$ In particular, $\log m<6\varphi(q)<\exp(7811.4)$, whence $n<\exp(\exp(7812))$, which is a contradiction. If we assume GRH and relax the assumption on $n$ to $n>\exp(15667)$, we use Corollary 1.1 of Lee (which appeared as Q. J. Math. 74 (2023), 1505-1533) to obtain the following strengthening of the previous display: $$\frac{m}{\varphi(q)}<\frac{1}{4\pi}(\log n+31008)\sqrt{n}\log n.$$ In particular, $$\sqrt{n}<\exp(7813)(\log n+31008)\log n,$$ whence $n<\exp(15667)$, which is again a contradiction.

Second response. This is an improvement of Max Alekseyev's nice counterexample. Consider $n=71$ with $$L := L_{71} = 2^6 \cdot 3^3 \cdot 5^2 \cdot 67 \cdot 71 \cdot L',$$ where the co-factor $L'$ is coprime to listed primes. Then set $$M := 2^8 \cdot 3^4 \cdot 5^3 \cdot 79 \cdot L'.$$

We have $$L = 205502400\cdot L' > 204768000\cdot L' = M,$$ while $$\sigma(L) = 771022080\cdot \sigma(L') < 771650880\cdot \sigma(L') = \sigma(M).$$ That is, $L_{71}$ is not highly abundant.

First response. I took on Terry's challenge "beat the AI". With my own little program (which is simpler than the AI's), starting from Terry's parameters, I got down to $n=13757$: $$\{p_1,\dotsc,p_{15}\}=\{127, 131, 137, 139, 149, 151, 157, 163, 167, 173, 179, 181, 191, 193, 197\},$$ $$\{q_1,\dotsc,q_8\}=\{13553, 13679, 13691, 13709, 13721, 13729, 13751, 13757\},$$ $$\sum_{m=1}^8 \log q_m - \sum_{j=1}^{15} \log p_j = 3.4179\dotsc \times 10^{-6},$$ $$\sum_{j=1}^{15} \log(1 +\frac{1}{p_j^2+p_j}) - \sum_{m=1}^8 \log(1+\frac{1}{q_m}) = 1.5531\dotsc \times 10^{-5}.$$

Added. The following text file contains variations of the data above. It shows that $L_n$ is not highly abundant for any $19919 \leq n < 201601$.

$\endgroup$
37
  • 2
    $\begingroup$ @DanielWeber My code is not professionally written, so I would be ashamed to share it, but I can tell you how it works. It starts with a given consecutive 15-tuple of primes $\mathcal{P}$, and it picks a random 8-tuple of primes $\mathcal{Q}$ from a given interval such that $\prod\mathcal{Q}>\prod\mathcal{P}$. It checks if $\mathcal{Q}$ can be updated by exchanging one if its elements, to decrease $\prod\mathcal{Q}$ but still maintaining $\prod\mathcal{Q}>\prod\mathcal{P}$. This last step is repeated several hundred times. For the final $\mathcal{Q}$, Terry's $(1)$ and $(2)$ are displayed. $\endgroup$ Commented Oct 3, 2025 at 9:16
  • 2
    $\begingroup$ @DanielWeber So my algorithm is not deterministic in the sense that it starts from a random tuple $\mathcal{Q}$, and it tries to modifiy it into a locally minimizing tuple $\mathcal{Q}$. By running the algorithm several times, and tweaking the parameters, better and better pairs of tuples $(\mathcal{P},\mathcal{Q})$ arose, which I kept recording manually in a text file. Hope this helps! $\endgroup$ Commented Oct 3, 2025 at 9:30
  • 3
    $\begingroup$ Interesting that the newest example ($67 \times 71 = 4757$ versus $2^2 \times 3 \times 5 \times 79 = 4740$) uses a prime larger than $n$ for the first time. So it's all down to $L_{67}$ now... $\endgroup$ Commented Oct 4, 2025 at 1:11
  • 4
    $\begingroup$ I'm beginning to suspect that $L_{67}$ is in fact highly abundant. Define the "score" of $M$ to be $\sigma(M)/M^{1.0045}$. If there is to be an $M$ for which $M < L_{67}$ and $\sigma(M) \geq \sigma(L_{67})$, then $M$ has to have a higher score than $L_{67}$. But of all the ways to add and remove primes from $L_{67}$, only the addition of $2$, $3$, or $3^2$, or the deletions of $59$, $61$, or $67$, increase the score; all other moves decrease the score, often by a lot (i.e. $L_{67}$ is nearly collossally abundant). This should reduce things to a feasible case check. $\endgroup$ Commented Oct 4, 2025 at 3:33
  • 5
    $\begingroup$ Using the score function of @TerryTao I find that no primes greater than $109$ can occur in the prime factorization of $M$, and get the following inclusive ranges for the exponents {2: (5, 12), 3: (3, 7), 5: (2, 4), 7: (1, 3), 11: (1, 2), 13: (1, 2), 17: (1, 2), 19: (1, 2), 23: (1, 1), 29: (1, 1), 31: (1, 1), 37: (0, 1), 41: (0, 1), 43: (0, 1), 47: (0, 1), 53: (0, 1), 59: (0, 1), 61: (0, 1), 67: (0, 1), 71: (0, 1), 73: (0, 1), 79: (0, 1), 83: (0, 1), 89: (0, 1), 97: (0, 1), 101: (0, 1), 103: (0, 1), 107: (0, 1), 109: (0, 1)}. Checking these, I find that $L_{67}$ is highly abundant. $\endgroup$ Commented Oct 4, 2025 at 7:03
34
$\begingroup$

This is a community wiki answer to collate the known status of $L_n$ for various $n$, restricting to the prime power case as these are the only $n$ for which $L_n$ is different from any previous $L_{n'}$. Please update with new results as they come in.

Results are also reported in OEIS A389482.

$n$ $L_n$ highly abundant? References Code / Data
1-64 Yes OEIS (Confirmed: DSM, Matthew Bolan) A002093, A003418
67 Yes Matthew Bolan (Confirmed: DSM) Python
71, 73, 79 No GH from MO, Max Alekseyev (Confirmed: Matthew Bolan, B. Mehta + Lean) Lean, or see below
81, 83, 89 Yes Max Alekseyev (Confirmed: Matthew Bolan) Python, SAGE
97, 101, 103, 107, 109, 113, 121 No Max Alekseyev, GH from MO (Confirmed: Matthew Bolan, Polytropos + AI, B. Mehta + Lean) Lean, or see below
125, 127, 128, 131, 137, 139 Yes Max Alekseyev (Confirmed: DSM) SAGE
149, 151, 157, 163, 167 No Max Alekseyev (Confirmed: DSM, B. Mehta + Lean) Lean, or see below
169 Yes Max Alekseyev SAGE
173-10,000,000,000 No Max Alekseyev, DSM, GH from MO, Matthew Bolan (173-229, 256: Max Alekseyev) (367, 373: aorq) (2,311: Eduardo Rodríguez Golvano) (4483: Der Schutz) (13,757-16,128: GH from MO) (200,000-229,440: Terry Tao + AI) (Confirmed: Matthew Bolan, DSM, B. Mehta + Lean) data, Lean
beyond 10,000,000,000 No Terry Tao, Dusart (beyond $\exp(15667)$ under GRH: GH from MO, Saúl RM, Lee) (beyond $\exp(\exp(7812))$: GH from MO, Saúl RM, Bennett et al.) theoretical analysis, Lean

For selected $n$, one can demonstrate that $L_n$ is not highly abundant by dividing $L_n$ by some product of primes, then multiplying $L_n$ by a slightly smaller product of primes, according to the following table.

$n$ Divide $L_n$ by Multiply $L_n$ by
71, 73 $67 \times 71 = 4757$ $2^2 \times 3 \times 5 \times 79 = 4740$
79 $67 \times 79 = 5293$ $2^5 \times 3 \times 5 \times 11 = 5280$
97, 101, 103, 107, 109, 113 $89 \times 97 = 8633$ $2^2 \times 3 \times 5 \times 11 \times 13 = 8580$
121 $101 \times 113=11413$ $2 \times 5 \times 7 \times 163=11410$
149, 151 $137 \times 149=20413$ $2 \times 5 \times 13 \times 157=20410$
157, 163, 167 $149 \times 151=22499$ $2 \times 5 \times 13 \times 173=22490$
173, 179, 181, 191, 193, 197, 199, 211, 223, 227, 229, 233, 239, 241 $157\times 173 = 27161$ $2^2\times 3\times 7\times 17\times 19 = 27132$
243, 251 $197\times 227=44719$ $2\times 5\times 17\times 263 = 44710$
256 $2 \times 227 \times 239 \times 241=26149946$ $3^2\times 5\times 7\times 17 \times 19 \times 257 = 26148465$
367, 373 $337 \times 347 \times 359=41981101$ $2^2 \times 3 \times 23 \times 383 \times 397 = 41966076$

For all $n$ in the above table, the failure of $L_n$ to be highly abundant was also verified in Lean.

For general $n$, we still divide by some primes and multiply by others, but according to the following procedures:

  • For all $n \leq 10^{10}$ for which $L_n$ is not highly abundant one can use this ladder of 3477 different ways to add and remove primes, each valid for a separate range of $n$; this ladder was formed by merging together several previous partial results.
  • For $n > 10^{10}$, one uses a theoretical analysis based on Dusart's effective version of the prime number theorem to verify the construction (which involves removing three large primes, then adding three medium primes, a factor of 4, and some additional terms).

Connections to the literature. The falsity of the original conjecture was already claimed in the 1944 paper of Alaoglu and Erdős, where on pages 466-467 they write "it would be easy to see that neither $n!$ nor the least common multiple of the first $n$ integers can be highly abundant infinitely often". While no proof is explicitly given, one can deduce this claim from the following (unnamed) Corollary on page 463, that describes rather precisely (most of) the prime factorization of a highly abundant number. It states the following:

Proposition. Fix $\varepsilon > 0$. There is a constant $c$ satisfying the following. Fix $N$ highly abundant. For $q$ a prime factor of $N$, let $k_q$ be the exponent of $q$ in $N$. Then less than $c \log \log N / \log \log \log N$ primes $q$ fail to satisfy the inequality: $$ (1 - \varepsilon) \log N \log \log N / (q \log q) < q^{k_q} < (1 + \varepsilon) \log N \log \log N / \log q. $$

Thus for instance, were $L_n = \exp((1+o(1))n)$ to be highly abundant, this proposition would force $L_n$ to be divisible by $p$ exactly once for almost all $(1+\varepsilon) (2n)^{1/2} \leq p \leq (1-\varepsilon) n$, exactly twice for almost all $(1+\varepsilon) (3n)^{1/3} \leq p \leq (1-\varepsilon) (2n)^{1/2}$, and so forth, where $\varepsilon>0$ is fixed, and "almost all" means that there are only $O_\varepsilon(\log n/\log\log n)$ exceptions. But this is not what actually happens: $L_n$ is in fact is divisible by $p$ exactly once for $n^{1/2} < p \leq n$, exactly twice for $n^{1/3} < p \leq n^{1/2}$, and so forth. This explains why $L_n$ is not highly abundant; a similar analysis also works for $n!$.

It is also instructive to compare $L_n$ with the collossally abundant number $C_n$ associated to the parameter $\varepsilon = \frac{1}{n \log n}$, which is a quintessential example of a highly abundant number. Standard calculations show that $C_n$ is divisible by $p$ exactly once for $(1+o(1)) (2n)^{1/2} \leq p \leq (1+o(1)) n$, exactly twice for $(1+o(1)) (3n)^{1/3} \leq p \leq (1+o(1)) (2n)^{1/2}$, and so forth, which is consistent with the proposition.

$\endgroup$
66
  • 4
    $\begingroup$ @DSM: 256 is a No with M/L = 2^-1 * 3^2 * 5 * 7 * 17 * 19 * 227^-1 * 239^-1 * 241^-1 * 257. $\endgroup$ Commented Oct 5, 2025 at 10:03
  • 4
    $\begingroup$ 367 and 373 are no with M/L = 2^2 * 3 * 23 * 337^-1 * 347^-1 * 359^-1 * 383 * 397. $\endgroup$ Commented Oct 5, 2025 at 11:02
  • 4
    $\begingroup$ I think we're almost ready to declare victory, but given that we've made some errors in the past, it would be good to have each of the main intervals needed to cover the entire range confirmed by someone other than the initial contributor for that interval. (For instance, even though there is no reason to doubt the OEIS data, it may be good to double-check that range, which presumably should be easy with all the tools we now have.) $\endgroup$ Commented Oct 5, 2025 at 22:28
  • 4
    $\begingroup$ For lean formalization purposes it turned out to be desirable to minimize the size of the ladder (well, really the number of distinct large primes used) as much as possible. I managed to reduce the ladder to just 102 rungs between $173$ and $10^{10}$, which I have added as a second file to my gist. $\endgroup$ Commented Oct 8, 2025 at 20:18
  • 4
    $\begingroup$ With Matthew's shorter ladder, I've verified 173 - $10^{10}$ in Lean, with code available here. $\endgroup$ Commented Oct 10, 2025 at 7:22
27
$\begingroup$

UPDATE (2025-10-06). Here is a SageMath implementation of my algorithm that for given integers $\sigma_L$ and $B$, finds the smallest integer $m\leq B$ with $\sigma(m)\geq \sigma_L$. Running it on $(\sigma_L,B)=(\sigma(L_n),L_n)$ either returns $L_n$ and thus proves that $L_n$ is HA, or returns some $M<L_n$ proving that $L_n$ is not HA.

At its core the algorithm relies on the inequality $$\frac{\sigma(m)}m \leq \prod_{\text{prime }p|m} \frac{p}{p-1}$$ and the following theorem.

Theorem. Let $p_1=2 < p_2=3 < \dots$ denote the prime numbers. Suppose that $p_k$ is the smallest prime factor of $m\leq B$ with $\sigma(m)\geq \sigma_L$. Then for $\ell := \omega(m)$, we have $$\prod_{i=k}^{k+\ell-1} p_i \leq B$$ and $$\prod_{i=k}^{k+\ell-1} \frac{p_i}{p_i-1} \geq \frac{\sigma_L}{B}.$$

The proof follows from noticing that $\frac{p}{p-1}$ is a decreasing function of $p$. My algorithm uses this theorem to effectively prune primes that may form a suitable $m$. It also uses parallelization.

To prove non-HA of $L_n$, we don't really need the smallest $m$, and the code can be stopped as soon as some $m<L_n$ is discovered. Such early termination can be requested by setting parameter just_any=True. As an example, the code checks HA-ness of $L_{67}$ and $L_{97}$.


UPDATE (2025-10-04). I've computationally verified that for an integer $n\in [67,113]$, $L_n$ is highly abundant only when $n\leq 70$ or $81\leq n\leq 96$. It follows that $n=71$ is the smallest index, for which $L_n$ is not highly abundant.

Just out of curiosity, I've also verified that $M$ constructed for $L_{71}$ by @GHfromMO is highly abundant, while $M$ for $L_{113}$ given below is not.


Changing primes that come with exponents greater than 1 enables much smaller counterexamples.

Consider $n=113$ with $$L_{113} = 2^6 \cdot 5^2 \cdot 7^2 \cdot 11 \cdot 13 \cdot 89\cdot 113 \cdot L',$$ where the co-factor $L'$ is coprime to listed primes. Then set $$M := 2^7 \cdot 5^3 \cdot 7^3 \cdot 11^2 \cdot 13^2 \cdot L'.$$

We have $$L = 112751038400\cdot L' > 112224112000\cdot L' = M,$$ while $$\sigma(L) = 386809305120\cdot \sigma(L') < 387282168000\cdot \sigma(L') = \sigma(M).$$ That is, $L_{113}$ is not highly abundant.

$\endgroup$
13
  • 4
    $\begingroup$ Nice! According to the OEIS lists provided in the comments, $L_n$ is highly abundant for $n \leq 66$, so it is now not inconceivable that with a bit more cleverness and computation (maybe involving integer programming?) one could actually find the minimal counterexample. $\endgroup$ Commented Oct 3, 2025 at 21:07
  • 3
    $\begingroup$ In fact, since $L_n$ only varies at prime powers, there are only 13 remaining candidates for the minimal counterexample $n$: $67,71,73,79,81,83,89,97,101,103,107,109,113$. Anyone want to place some bets? $\endgroup$ Commented Oct 3, 2025 at 21:31
  • 2
    $\begingroup$ @TerryTao If I did not make a mistake, $L_{97}$ is not highly abundant. See the new section in my response. $\endgroup$ Commented Oct 3, 2025 at 21:37
  • 5
    $\begingroup$ It's remarkable that the OEIS data was almost maximally misleading as to the truth of the OP's question. $\endgroup$ Commented Oct 4, 2025 at 0:54
  • 3
    $\begingroup$ @TerryTao: I've added a draft sequence oeis.org/draft/A389482 (and its sister A389483) Please update as suitable. $\endgroup$ Commented Oct 5, 2025 at 17:10
21
$\begingroup$

Following Terry's method and GH from MO, I started from the upper bound of GH from MO and found this pretty quick: $n=4483$ $$ \{p_1,\dots,p_{13}\}=\{67,71,73,79,83,89,97,101,103,107,109,113,127\} $$

$$ \{q_1,\dots,q_7\}=\{4349,4441,4447,4451,4457,4481,4483\} $$

$$ \sum_{m=1}^{7}\log q_m-\sum_{j=1}^{13}\log p_j = 1.66862917368\times 10^{-5}, $$

$$ \sum_{j=1}^{13}\log\!\left(1+\frac{1}{p_j^2+p_j}\right) -\sum_{m=1}^{7}\log\!\left(1+\frac{1}{q_m}\right) = 5.62404029424\times 10^{-5}. $$

$\endgroup$
6
  • 2
    $\begingroup$ Nice! For $\{q_1,\dotsc,q_7\}$, one can also take $\{4357, 4421, 4447, 4457, 4463, 4481, 4483\}$. $\endgroup$ Commented Oct 3, 2025 at 15:42
  • 1
    $\begingroup$ Here is a near-counterexample, in hopes that it might inspire something similar that works: Let $p=19$, $q=19^2-2$, $r=19^3-2$, $n=19^6+16$. Then $p,q,r,n$ are all prime. Let $M=pqrL_n/n<L_n$. In this case $\sigma(M)>0.994\, \sigma(L_n)$, and $\sigma(M)>\sigma(L_n)$ would provide a counterexample. $\endgroup$ Commented Oct 3, 2025 at 16:09
  • 1
    $\begingroup$ {59,61,67,71,73,79,83,89,101,131,179} and {3413,3457,3461,3463,3491,3499} $\endgroup$ Commented Oct 3, 2025 at 20:31
  • 1
    $\begingroup$ @TerryTao: Oops, I took the "integer square root" (the floor of the square root). I was trying $n=3500$. My verification code checked the log inequalities but not the sqrt one! :( Thank you very much for pointing this out, even though it's nowhere near the record results now. $\endgroup$ Commented Oct 5, 2025 at 1:35
  • 1
    $\begingroup$ @aorq it is possible your code could still be used to cover some of the missing regions (see the community wiki page). If you can find a counterexample where the first prime $p_1$ is significantly greater than the square root of the last prime $q_m$ then this covers a region $n \in [q_m, p_1^2-1]$; chaining together a few of these may be able to cover some of the remaining intervals. $\endgroup$ Commented Oct 5, 2025 at 1:40
21
$\begingroup$

Update (14-Oct-2025): The range 173 to $10^{40}$ is also verified.

Update: The range 173 to 10,000,000,000 is now verified in Lean, using @Matthew Bolan's shorter ladder mentioned in another comment.

Update: I have now verified, in Lean, all the disproofs for small $n$ (below $373$) listed in the community wiki table (at time of writing). The code is available here (in the highly_abundant_small.lean file), https://gist.github.com/b-mehta/75d7f118e5bb6440c131c4c6ceaf7478#file-highly_abundant_small-lean. They can also be interacted with in the online Lean sandbox here.

Original post:

Here is a quick and direct Lean verification of the failure of this conjecture, using the idea given in the second response of GH from MO's answer.

import Mathlib

open Nat ArithmeticFunction

-- definition of highly abundant number
def IsHighlyAbundant (N : ℕ) : Prop :=
  ∀ m > 0, m < N → σ 1 m < σ 1 N

-- definition of lcm of the range {1..n}
def lcmRange (n : ℕ) : ℕ := (Finset.Icc 1 n).lcm _root_.id

-- a few small examples of lcmRange
example : lcmRange 1 = 1 := rfl
example : lcmRange 2 = 2 := rfl
example : lcmRange 6 = 60 := rfl

-- two general helper lemmas
lemma sigma_one_apply_prime {p : ℕ} (hp : p.Prime) : σ 1 p = p + 1 := by
  have : σ 1 p = σ 1 (p ^ 1) := by simp
  rw [this, sigma_one_apply_prime_pow hp]
  simp

lemma sigma_one_apply_prime_pow' {p k : ℕ} (hp : p.Prime) :
    σ 1 (p ^ k) = (p ^ (k + 1) - 1) / (p - 1) := by
  rw [sigma_one_apply_prime_pow hp, Nat.geomSum_eq]
  exact hp.two_le

namespace verify_l71

-- the choice of L to verify that the conjecture fails at 71
def L : ℕ := lcmRange 71

-- prove L is what we say it is
example : L = 5624043567677125526551547131200 := by decide +kernel

-- we follow the proof given by "GH from MO"
def L' : ℕ := 7 ^ 2 * 11 * 13 * 17 * 19 * 23 * 29 * 31 * 37 * 41 * 43 * 47 * 53 * 59 * 61

def M : ℕ := 2 ^ 8 * 3 ^ 4 * 5 ^ 3 * 79 * L'

lemma L_eq : L = 205502400 * L' := by decide +kernel
lemma coprime_L'L : Nat.gcd (2 ^ 6 * 3 ^ 3 * 5 ^ 2 * 67 * 71) L' = 1 := by decide +kernel
lemma L_eq' : L = 2 ^ 6 * 3 ^ 3 * 5 ^ 2 * 67 * 71 * L' := by decide +kernel

lemma M_eq : M = 204768000 * L' := rfl
lemma coprime_L'M : Nat.gcd (2 ^ 8 * 3 ^ 4 * 5 ^ 3 * 79) L' = 1 := by decide +kernel
lemma M_eq' : M = 2 ^ 8 * 3 ^ 4 * 5 ^ 3 * 79 * L' := by decide +kernel

lemma M_lt_L : M < L := by decide +kernel

lemma sigma_L_aux : σ 1 (2 ^ 6 * 3 ^ 3 * 5 ^ 2 * 67 * 71) = 771022080 := by
  repeat rw [isMultiplicative_sigma.map_mul_of_coprime rfl]
  repeat rw [sigma_one_apply_prime_pow' (by norm_num)]
  repeat rw [sigma_one_apply_prime (by norm_num)]
  rfl

lemma sigma_M_aux : σ 1 (2 ^ 8 * 3 ^ 4 * 5 ^ 3 * 79) = 771650880 := by
  repeat rw [isMultiplicative_sigma.map_mul_of_coprime rfl]
  repeat rw [sigma_one_apply_prime_pow (by norm_num)]
  repeat rw [sigma_one_apply_prime (by norm_num)]
  rfl

-- the main result: L is not highly abundant
theorem not_HA_L71 : ¬ IsHighlyAbundant L := by
  intro h
  specialize h M (by decide +kernel) M_lt_L
  rw [L_eq', M_eq', isMultiplicative_sigma.map_mul_of_coprime coprime_L'M,
    isMultiplicative_sigma.map_mul_of_coprime coprime_L'L] at h
  simp only [sigma_M_aux, sigma_L_aux] at h
  simpa using Nat.lt_of_mul_lt_mul_right h

-- there exists an n (namely 71) such that lcmRange n is not highly abundant
theorem conjecture_fails : ∃ n, ¬ IsHighlyAbundant (lcmRange n) := ⟨71, not_HA_L71⟩

end verify_l71
$\endgroup$
2
  • $\begingroup$ Just to be clear, Lean is using GMP for calculations, right? Is it relying on any other external libraries for this calculation? $\endgroup$ Commented Oct 10, 2025 at 21:40
  • $\begingroup$ For numbers below 2^64, GMP isn't used so below 10^10 there's no external libraries whatsoever. From 2^64 upwards, GMP is used for the basic operations on natural numbers. No other external libraries are used. $\endgroup$ Commented Oct 11, 2025 at 0:03
17
$\begingroup$

I will show $L_{67}$ is highly abundant, and so $L_{71}$ is the smallest counterexample. My program is also able to recover the example proving $L_{71}$ is not highly abundant.

Following Terry's suggestion in the comments on GH from MO's answer, fix some $\alpha > 0$ and define the "score" $s(m) = \frac{\sigma(m)}{m^{\alpha}}.$ If there were some $M$ such that $M < L_{67}$ and $\sigma(M) \ge \sigma(L_{67})$, we must have $s(M) > s(L_{67})$, so it suffices to search those $M$ with $s(M) > s(L_{67})$. Terry suggests we take $\alpha = 1.0045$, which I will write this answer assuming, but I later found that $\alpha = 1.004$ made my search slightly faster.

For each prime $p_i$ there is some exponent $k_i$ such that $s(p_i^{k_i})$ is maximal, which I found with a computer search. I was able to search a finite space here as $\ln(s(p^k))$ is bounded above above by $(1-0.0045k) \ln p - \ln{(p-1)},$ which is decreasing in both $p$ and $k$. Combining these, we find that the $M$ with the largest score is $$\prod_{i \in \mathbb N} p_i^{k_i} = L_{67} \cdot 2 \cdot 3 \cdot 59^{-1} \cdot 61^{-1} \cdot 67^{-1}.$$ The fact it is so close to $L_{67}$ should be Terry's reason for the choice of $1.0045$.

The same process allows us to determine, for a given prime power $p^k$, the highest possible score of any $M$ with $p^k\mid M$ but $p^{k+1} \nmid M$ - they are just the maximal $M$ but with the power of $p$ replaced by $p^k$. This leads to the determination that no prime $p > 109$ can divide $M$, and that the other prime exponents must fall into the ranges {2: (5, 12), 3: (3, 7), 5: (2, 4), 7: (1, 3), 11: (1, 2), 13: (1, 2), 17: (1, 2), 19: (1, 2), 23: (1, 1), 29: (1, 1), 31: (1, 1), 37: (0, 1), 41: (0, 1), 43: (0, 1), 47: (0, 1), 53: (0, 1), 59: (0, 1), 61: (0, 1), 67: (0, 1), 71: (0, 1), 73: (0, 1), 79: (0, 1), 83: (0, 1), 89: (0, 1), 97: (0, 1), 101: (0, 1), 103: (0, 1), 107: (0, 1), 109: (0, 1)} in order to have a chance of beating $s(L_{67})$.

I checked these by iterating through the possible prime power factors recursively from largest to smallest, exiting the recursion early whenever the partial product exceeded $L_{67}$. Probably one can speed up this step by incorporating the score into the search and exiting early when you have lost too much score compared to $L_{67}$, or by doing integer linear programming. For the final run I used the score function $s(m) = \frac{\sigma(m)^{2000}}{m^{2009}}$ to determine all the bounds, which allowed me to do everything in exact arithmetic.

My code is here as a github gist. To find the counterexample for $L_{71}$, one simply changes N to 71 and SCORE_NUMERATOR and SCORE_DENOMINATOR to $2007$ and $2000$ respectively.

$\endgroup$
5
  • 2
    $\begingroup$ Thanks! It seems that this informal project has led to some useful ways to either prove or disprove the claim that a given number is highly abundant. Regarding the provenance of the exponent 1.0045, I had asked an AI at chatgpt.com/share/68e1444f-902c-800e-ab87-35a6d7a6e203 for code to plot the vectors $(\log p^{k+h}, \log \sigma(p^{k+h})) - (\log p^{k}, \log \sigma(p^{k}))$ where $p^k$ is the largest power of $p$ dividing $L_{67}$ and $h$ was a shift. It did so but the plot was almost a diagonal line, reflecting the fact that $\sigma(p^j) \approx p^j$. (cont) $\endgroup$ Commented Oct 4, 2025 at 20:02
  • 2
    $\begingroup$ I then manually adjusted the code (outside of the chat) by linear transformations to accentuate the interesting features of the plot, finding through trial and error that replacing $\log \sigma(p^j)$ by $\log \sigma(p^j) - 1.0045 \log p^j$ pushed almost all of the vectors to the bottom two quadrants, leaving only six points in the upper quadrants. Undoing the logarithm gave me the score-based interpretation of this plot. $\endgroup$ Commented Oct 4, 2025 at 20:04
  • 3
    $\begingroup$ In case anyone is interested, the transformed plot can be found at terrytao.wordpress.com/wp-content/uploads/2025/10/shift-fig.png $\endgroup$ Commented Oct 4, 2025 at 20:06
  • 2
    $\begingroup$ One other remark: one could use multiple scores (e.g., the 1.0045 and 1.004 one) and then take the intersection of all the exponent intervals to narrow the search space before checking all the remaining candidates. $\endgroup$ Commented Oct 4, 2025 at 20:13
  • 2
    $\begingroup$ Incidentally: the maximal $M = 1970992304700453905270400$ is the 27th collossally abundant number. en.wikipedia.org/wiki/Colossally_abundant_number $\endgroup$ Commented Oct 4, 2025 at 20:23
17
$\begingroup$

Here is a way to, assuming the conjecture below, prove that for big enough $N$, lcm$(1,\dots,N)$ is not highly abundant. This was discussed in the comments to the question, but is too long for a comment. The argument is based on a simple proof that $L_{2311}=L_{2\cdot3\cdot5\cdot7\cdot11+1}$ is not highly abundant by Eduardo Rodríguez Golvano.

Edit: The conjecture below was completely unnecessary for the argument to work (we don't need $p-1$ to be squarefree), as shown in GH from MO's answer (the third response, in particular).

Let $p_i$ be the $i^{\text{th}}$ prime. Let $r\in\mathbb{N}$ be big enough that $\sum_{i=1}^r\frac{1}{p_i}>8$.

The following would be needed to complete the argument (see the comment below by Terry Tao for ideas about how to prove it):

Conjecture: Let $k>0$. For big enough $N$ there exists a prime $p\in(2N/3,N)$ such that $p\equiv1$ mod $k$ and $p-1$ is squarefree.

Now suppose the conjecture is true and let $k=2\cdot3\cdot5\cdots p_r$. For big enough $N$ consider $p\in(2N/3,N)$ prime such that $p\equiv1$ mod $k$ and $p-1$ is squarefree. It would be enough to prove that $\sigma(L_N)<\sigma\left(L_N\cdot\frac{p-1}{p}\right)$. We can compute. Let $I=\{i\in\mathbb{N};p_i|p-1\}$ and $k_i$ the biggest exponent such that $p_i^{k_i}\leq N$. Then,

$$\frac{\sigma(L_N)}{\sigma\left(L_N\cdot\frac{p-1}{p}\right)}=(p+1)\prod_{i\in I}\frac{p_i^{k_i+1}-1}{p_i^{k_i+2}-1} =\frac{p+1}{p-1}\prod_i\frac{p_i^{k_i+1}-1}{p_i^{k_i+1}-\frac{1}{p_i}}$$ $$<\left(1+\frac{4}{N}\right)\cdot\prod_{i\in I}\left(1-\frac{1}{2p_i^{k_i+1}}\right)<\left(1+\frac{4}{N}\right)\cdot\prod_{i\in I}\left(1-\frac{1}{2p_iN}\right).$$ So taking logs, $$\ln\left(\frac{\sigma(L_N)}{\sigma\left(L_N\cdot\frac{p-1}{p}\right)}\right)<\frac{4}{N}-\sum_{i\in I}\frac{1}{2p_iN}\leq\frac{4}{N}-\sum_{i=1}^r\frac{1}{2p_iN}<0.$$

$\endgroup$
7
  • 4
    $\begingroup$ Nice! I think you mean "not highly abundant" in the first sentence. The desired claim should follow from a combination of the prime number theorem in arithmetic progressions and the Brun-Titchmarsh inequality en.wikipedia.org/wiki/Brun%E2%80%93Titchmarsh_theorem to subtract off multiples of $p^2$, perhaps after passing to a finer congruence class to dodge multiples of very small squares such as $2^2$. By the way, this opens up the possibility of a computer assisted proof that $L_n$ is not highly abundant for all $n \geq 71$. $\endgroup$ Commented Oct 4, 2025 at 19:17
  • 1
    $\begingroup$ @TerryTao: In fact, $L_{148}$ ($=L_{139}$ if we restrict $n$ to ptime powers) is highly abundant. I can suggest it as a candidate for the largest one. (I'm searching for a larger one now with no success.) $\endgroup$ Commented Oct 4, 2025 at 22:13
  • 1
    $\begingroup$ Ops. $L_{169}$ is also highly abundant. $\endgroup$ Commented Oct 4, 2025 at 22:18
  • 2
    $\begingroup$ @TerryTao We do not need Brun-Titchmarsh. In fact by elaborating on Saúl RM's nice argument, I prove in my "Third response" that $L_n$ is not highly abundant if there exists a prime $p$ such that $p-1$ is divisible by $2310$ and lies in the interval $[(120/121)n,n-1]$. This will quickly give an effective upper bound for the largest $n$ such that $L_n$ is highly abundant. $\endgroup$ Commented Oct 4, 2025 at 23:15
  • 1
    $\begingroup$ After thinking more about it, I am not sure how to use Brun-Titchmarsh to prove the claim (probably just my lack of experience with this kind of argument), so for now I'll return the answer to its previous state. Hopefully it won't matter thanks to @GHfromMO's new arguments. If anyone wants to edit the answer proving the claim anyways, that would be great though, it seems like an interesting claim on its own $\endgroup$ Commented Oct 5, 2025 at 0:11
8
$\begingroup$

This is an answer to the following question from the comments:

Is every factorial a highly abundant number?

My claim is, for $n>15$, there exists $k$ such that $\sigma(\frac{2^k−1}{2^k}n!)>\sigma(n!)$.

It can be easily checked for these intervals (see below):

$k=9$ for $16 \leq n \leq 72$

$k=60$ for $73 \leq n \leq 1320$

At this point, with such high numbers, we can pick any $k$ around $n/2$ with the only requirement $2^k−1$ is not a Mersenne prime (EDIT: In the proof below, the exact requirement is $k$ is divisible by $3$, that guarantees $2^k-1$ is divisible by $7$).

To check the previous intervals we can do the following. For $n=16$ and $k=9$ we have $2^9−1=511=7 \times 73$, and $16! = 2^{15}\cdot7^2\cdot m$, where $m$ is coprime to $2$, $7$ and $73$. Then:

$\sigma(\frac{511}{512}\cdot16!)=\frac{2^7−1}{2^{16}−1}\cdot\frac{7^4−1}{7^3−1}\cdot74 \cdot \sigma(16!)>\frac{2^7−1}{2^{16}−1}\cdot518 \cdot \sigma(16!)>\sigma(16!)$

And this is not going to change until at least $n=73$, so the claim holds for the whole interval. The other interval can be checked with similar calculations.

EDIT: The proof for big $n$ could be like this. Suppose $s$ is the exponent of $7$ in $n!$, so $s<\frac{n}{6}$, and $r$ is the exponent of $2$ in $n!$. For $n>1023$ we have $r>\frac{99n}{100}$. We choose $k$ divisible by $3$ (that implies $2^k-1$ is divisible by $7$) such that $r-4 \leq 2k \leq r+1$, so:

$2^k>2^{\frac{99n-400}{200}}$

Then we have:

$\frac{\sigma(\frac{2^k−1}{2^k}n!)}{\sigma(n!)} > \frac{2^{r+1-k}-1}{2^{r+1}-1} \cdot \frac{7^{s+2}-1}{7^{s+1}-1} \cdot \frac{2^k-1}{7} = \frac{2^{r+1-k}-1}{2^{r+1}-1} \cdot (1+\frac{6}{7^{s+2}-7}) \cdot (2^k-1) = \frac{2^{r+1-k}-1}{2^{r+1}-1} \cdot (2^k-1+\frac{6}{7} \cdot \frac{2^k-1}{7^{s+1}-1}) > \frac{2^{r+1-k}-1}{2^{r+1}-1} \cdot (2^k-1+\frac{6}{7} \cdot \frac{2^{\frac{99n-400}{200}}-1}{7^{\frac{n+6}{6}}-1})$

The last term is increasing with $n$ and, for $n>1023$, is much bigger than $2$:

$\frac{\sigma(\frac{2^k−1}{2^k}n!)}{\sigma(n!)} > \frac{2^{r+1-k}-1}{2^{r+1}-1} \cdot (2^k+1) = \frac{2^{r+1}-2^k+2^{r+1-k}-1}{2^{r+1}-1} = 1 + \frac{2^{r+1-k}-2^k}{2^{r+1}-1} \geq 1$

This works for $n>1023$ so the proof is complete.

$\endgroup$
11
  • 2
    $\begingroup$ Eduardo decided to create an account (this is the same person that told me that $L_{2310}$ is not highly abundant, which then led to the proof that $L_{n}$ is not highly abundant for big $n$) $\endgroup$ Commented Oct 9, 2025 at 11:35
  • 2
    $\begingroup$ As Terry Tao pointed out, this was also mentioned in the Alaoglu-Erdos paper. $\endgroup$ Commented Oct 9, 2025 at 11:36
  • 3
    $\begingroup$ This together with the OEIS data means $n!$ is highly abundant if and only if $n \le 8$. $\endgroup$ Commented Oct 9, 2025 at 15:29
  • 3
    $\begingroup$ Nice! In particular one is able to handle the asymptotic regime without needing any form of the prime number theorem (as opposed to the Alaoglu-Erdos analysis, which needs variants of PNT at many places in their paper). $\endgroup$ Commented Oct 9, 2025 at 18:55
  • 1
    $\begingroup$ @EduardoRodríguezGolvano Your proof looks nice, but can you provide the details for $n>1320$? How to choose $k$ precisely, and why does it really work? Thanks in advance. $\endgroup$ Commented Oct 10, 2025 at 0:19

You must log in to answer this question.

Start asking to get answers

Find the answer to your question by asking.

Ask question

Explore related questions

See similar questions with these tags.