arXiv reader

On primitive prime divisors of the orders of Suzuki-Ree groups (corrected version)

Maria Grechkoseeva

arXiv
2504.04729v1 · PDF · 7 April 2025
Categories
math.GR
Licence
CC BY 4.0 The site shows the text.

Abstract

There is a well-known factorization of the number 2^{2m}+1, with m odd, related to the orders of tori of simple Suzuki groups: 2^{2m}+1 is a product of a=2^m+2^{(m+1)/2}+1 and b=2^m-2^{(m+1)/2}+1. By the Bang-Zsigmondy theorem, there is a primitive prime divisor of 2^{4m}-1, that is, a prime r that divides 2^{4m}-1 and does not divide 2^i-1 for any i<4m. It is easy to see that r divides 2^{2m}+1, and so it divides one of the numbers a and b. The main objective of this paper is to show that for every m>5, each of a and b is divisible by some primitive prime divisor of 2^{4m}-1. Also we prove similar results for primitive prime divisors related to the simple Ree groups. As an application, we find the independence and 2-independence numbers of the prime graphs of almost simple Suzuki-Ree groups.

Items 10

Theorem

Theorem 1 (Bang–Zsigmondy)

9LB9

Let q and m be integers larger than 1. Then either there is a prime that divides q^m-1 and does not divided q^i-1 for all i<m, or one of the following holds:

  1. (a) m=2, q=2^s-1, where s\geqslant 2;

  2. (b) m=6, q=2.

Theorem

Theorem 2

FI0Y

Let v=2,3,2 and n=4,6,12, respectively, let \Psi_n(x) be the polynomial defined in (?), and suppose that m is an odd integer larger than 1. Then for every \varepsilon\in\{+,-\}, either there is a prime that divides \Psi_{n}(\varepsilon\sqrt{v^{m}}) and does not divide v^{i}-1 for any i<nm, or v=2, n=4, and m=3, 5.

Theorem

Theorem 3

Y5HV

Let L be a finite simple Suzuki or Ree group and L<G\leq\operatorname{Aut} L. Then one of the following holds:

  1. (a) t(G)=t(L) and t(2,G)=t(2,L);

  2. (b) L={}^2B_2(32), t(G)=t(L)-1 and t(2,G)=t(2,L)-1;

  3. (c) L={}^2F_4(2^m), \pi(G)\not\subseteq\pi(L), t(G)=t(L)+1, and t(2,G)=t(2,L).

  4. (d) L={}^2G_2(3^m), \pi(G)=\pi(L), 3 divides |G:L|, t(G)=t(L)-1 and t(2,G)=t(2,L).

Used by Remark 3
Lemma

Lemma 1

UDWV

Let (m,n)=1 and let s be an odd positive integer.

  1. (a) If n=4,6, then (\Phi_{nm}(v^s), \Psi_n(\varepsilon\sqrt{v^{sm}}))=|\Phi_m(\sqrt{v^s}/\xi)\Phi_m(\sqrt{v^s}/\overline \xi)|, where \xi is some suitable primitive 2nth root of unity.

  2. (b) If n=12, then (\Phi_{nm}(v^s),\Psi_{n}(\varepsilon \sqrt{v^{sm}}))=|\Phi_{3m}(\sqrt{v^s}/\xi)\Phi_{3m}(\sqrt{v^s}/\overline\xi)|, where \xi is some suitable primitive 8th root of unity.

Proof of Lemma 1

(a) Suppose that n=4,6. Observe that \varphi(2n)=4 and that the primitive 2nth roots of unity can be written as \pm\xi, \pm\overline \xi, where \xi is some fixed root. Since 2n and m are coprime, if \mu runs over the set of primitive mth roots of unity, then \pm\xi\mu, \pm\overline\xi\mu are all of the primitive 2nmth roots of unity. So we have

\Phi_{2nm}(x)=\Phi_m(x/\xi)\Phi_m(x/\overline\xi)\Phi_m(-x/\xi)\Phi_m(-x/\overline\xi).

Furthermore, since n is even, it follows that \Phi_{nm}(x^2)=\Phi_{2nm}(x). Thus

\Phi_{nm}(x^2)=f_m(x)f_m(-x),

where

f_m(x)=\Phi_m(x/\xi)\Phi_m(x/\overline\xi).

It is clear that the coefficients of the polynomial f_m(x) are real. Also it is easy to see that the coefficients of the terms x^{2i} are integral linear combinations of numbers of the form \xi^{2j}+\overline\xi^{2j}, while the coefficients of x^{2i+1} are integral linear combinations of numbers of the form \xi^{2j+1}+\overline\xi^{2j+1}. Since

\xi^{2j}+\overline\xi^{2j}\in\{0,\pm 2\}\text{ and }\xi^{2j+1}+\overline\xi^{2j+1}=\pm\sqrt{2}\text{ if }n=4,
\xi^{2j}+\overline\xi^{2j}\in\{\pm 1,\pm 2\}\text{ and }\xi^{2j+1}+\overline\xi^{2j+1}\in\{0,\pm \sqrt{3}\}\text{ if }n=6,

we deduce that

f_m(x)=x^{2\varphi(m)}+a_1\sqrt{v}x^{2\varphi(m)-1}+a_2x^{2\varphi(m)-2}+\dots+1,

where a_1,a_2,\dots are integers. Thus f_m(\sqrt{v^s}) is an integer for every odd s.

Note that

\underset{d\mid m}{\prod}f_d(x)=(x^m-\xi^m)(x^m-\overline \xi^m)=x^{2m}-(\xi^m+\overline \xi^m)x^m+1=\Psi_n(\epsilon x^m),

where \epsilon\in\{+,-\} is defined by the equality \xi^m+\overline\xi^m=-\epsilon\sqrt{v}. Replacing, if necessary, the root \xi, we may assume that \epsilon=\varepsilon. By the above, every number f_d(\sqrt{v^s}), where d divides m, is an integer, and so f_m(\sqrt{v^s}) divides \Psi_n(\varepsilon\sqrt{v^{ms}}). Similarly, f_m(-\sqrt{v^s}) divides \Psi_n(-\varepsilon\sqrt{v^{ms}}). The numbers \Psi_n(\sqrt{v^{ms}}) and \Psi_n(-\sqrt{v^{ms}}) are coprime, and hence (\Phi_{nm}(v^s), \Psi_n(\varepsilon\sqrt{v^{ms}}))=|f_m(\sqrt{v^s})|, as required.

(b) Applying (Proof) with n=4, we derive that \Phi_{12m}(x^2)=f_{3m}(x)f_{3m}(-x). Also since 3 does not divide m, we have

\prod_{d\mid m}\Phi_{3d}(x)=\prod_{d\mid 3m}\Phi_d(x)/\prod_{d\mid m}\Phi_d(x)=\frac{x^{3m}-1}{x^m-1}=\Phi_3(x^m).

It follows that

\prod_{d\mid m}f_{3d}(x)=\Phi_3((x/\xi)^m)\Phi_3((x/\overline \xi)^m))=(x^{2m}-\xi^mx^m+\xi^{2m})(x^{2m}-\overline\xi^mx^m+\overline\xi^{2m})=
=x^{4m}+\epsilon x^{3m}+x^{2m}+\epsilon x^m+1=\Psi_{12}(\epsilon x^m),

where \epsilon is defined in the same way as in (a). The rest of the proof is similar to the proof of (a).

Lemma

Lemma 2

LVBB

Let n=4,6, (m,n)=1 and f_m(x)= \Phi_m(x/\xi)\Phi_m(x/\overline\xi), where \xi is a primitive 2nth root of unity. If either x\geqslant \sqrt{2} and m>5, or x\geqslant \sqrt{3} and m>3, then |f_m(x)|>r, where r is the largest prime divisor of m.

Proof of Lemma 2

Let m=r^il, where (r,l)=1. We have \Phi_m(x)=\Phi_l(x^{r^{i}})/\Phi_l(x^{r^{i-1}}) and so

|f_m(x)|=\frac{|\Phi_l(({x}/\xi)^{r^{i}})\Phi_l(({x}/\overline\xi)^{r^{i}})|}{|\Phi_l(({x}/\xi)^{r^{i-1}})\Phi_l(({x}/\overline\xi)^{r^{i-1}})|}.

The nominator of the last expression is a product of \varphi(l) factors of the form |y^{2r}-(\lambda+\overline\lambda)y^r+1|, where y=x^{r^{i-1}} and |\lambda|=1. Since y\geqslant \sqrt{2}, each of these factors is not less than y^{2r}-2y^r+1=(y^r-1)^2. Similarly, every factor in the denominator is not larger than (y+1)^{2}. Hence

|f_m(x)|\geqslant \left(\frac{(y^r-1)^2}{(y+1)^2}\right)^{\varphi(l)}.

It ie easy to check that

\frac{y^r-1}{y+1}>\frac{y^{r-2}(y^2-1)}{y+1}=y^{r-2}(y-1)\geqslant y^{r-2} \text{ if } y\geqslant 2,
\frac{y^r-1}{y+1}\geqslant \frac{y^{r-3}(y^3-1)}{y+1}>y^{r-3} \text{ if } y\geqslant \sqrt{3};
\frac{y^r-1}{y+1}\geqslant \frac{y^{r-5}(y^5-1)}{y+1}>y^{r-5}y^{3/2}=y^{r-\frac{7}{2}} \text{ if } r\geqslant 5.

Thus for all r\geqslant 7, we have |f_m({x})|>y^{2(r-\frac{7}{2})}>y^{2(r-4)}\geqslant 2^{r-4}>r.

If r=3, then i>1, and so y\geqslant \sqrt{2^3}>2, and |f_m(x)|>y^2\geqslant 2^3>3.

Let r=5. If x\geqslant \sqrt{3} or i>1, then y\geqslant \sqrt{3} and hence |f_m({x})|>y^4\geqslant3^{2}>5. If x<\sqrt{3} and i=1, then l\geqslant 3, and therefore |f_m(x)|>(y^{3})^{\varphi(l)}\geqslant \sqrt{2}^6>5.

Remark

Remark 1

46XV

Let k\geqslant 2 be a square-free positive integer and suppose that v is a divisor of k such that k/v is odd. Let \tau=- if v\equiv 1\pmod 4 and \tau=+ otherwise. By [62Sch, Theorem 1], there are polynomials P_{k,v}(x), Q_{k,v}(x)\in \mathbb{Z}[x] such that \Phi_{2k}(\tau x)=P_{k,v}^2(x)-vxQ^2_{k,v}(x). Hence for every odd positive integer m, we have the following factorization of \Phi_{2k}(\tau v^m) as a product of two integers:

\Phi_{2k}(\tau v^m)=(P_{k,v}(v^m)+v^{(m+1)/2} Q_{k,v}(v^m))(P_{k,v}(v^m)-v^{(m+1)/2} Q_{k,v}(v^m)).

The factorizations of \Phi_4(2^m), \Phi_6(3^m) and \Phi_{12}(2^m) related to the Suzuki–Ree groups are in fact the factorization (1) with k=2,3, 6 and v=2,3,2 respectively. Since \pi(\Phi_{2k}(\tau v^m)) contains all primitive divisors of v^{2mk}-1 (if \tau=+) or of v^{mk}-1 (if \tau=-), one can ask under what conditions on k, v and m, each of the factors in (1) is divisible by the corresponding primitive divisor, and thus obtain some generalization of Theorem 2. However, the study of this question is beyond the scope of the present work. Also it is worth mentioning that [62Sch, Theorem 2] provides some sufficient conditions for v^{2mk}-1 (or v^{mk}-1) to have at least two primitive divisors.

Remark

Remark 2

CRY4

Primitive prime divisors of the orders of Suzuki–Ree groups are also studied in [22Katz, Section 3], and Theorem 2 of the present paper is Theorems 3.4, 3.7 and Corollary 3.5 in [22Katz]. Observe that the author discovered the work [22Katz] after submitting the present paper to “Algebra and Logic” and that the proofs of these two works, despite some common motifs, are quite different.

Remark

Remark 3

GHX8

Theorem 3 of [23Gre.t] was used in the proof of Theorem 1 of [24GrePan.t] to show that the spectrum of an almost simple group G with socle ^2G_2(3^m) cannot be equal to the spectrum of a classical simple group H [24GrePan.t, p. 1082]. A corrected argument is as follows. If \omega(G)=\omega(H), then 4\not\in\omega(H) and so H=L_2(u) for some u by Walter’s classification of simple finite groups with abelian Sylow 2-subgroups [69Wal]. Since t(L_2(u))=3 and t(G)\geqslant 4 by the corrected version of Theorem 3, we derive a contradiction.

Source and licence

The text on this page is from arXiv:2504.04729v1 by Maria Grechkoseeva, under the licence CC BY 4.0. The corpus made these changes: it read the items out of the LaTeX source, it expanded the macros of the paper, it put each proof after its statement, and it converted the text to HTML. The formulas are the TeX of the paper, and KaTeX renders them. A formula or a reference can look different from the PDF. The page is published under CC BY 4.0.

The record of each item is in analysis/objects/2504/2504.04729.jsonl.