betaFeedback

WorksMathematics

The multiplicity bound for near-group fusion categories of type Z/4Z+k\mathbb Z/4\mathbb Z+k

Abstract

Let C\mathcal{C} be a near-group fusion category of type Z/4Z+k\mathbb Z/4\mathbb Z+k. We show k∈{0,3,4}k\in\{0,3,4\}, which then gives a complete characterization of near-group fusion categories determined by Z/4Z\mathbb Z/4\mathbb Z.

Introduction

The fusion rules of a near-group fusion category are determined by a finite group GG and a nonnegative integer kk, with the additional simple object ρ\rho satisfying ρ⊗ρ=⨁g∈Gg⊕kρ\rho\otimes\rho= \bigoplus_{g\in G} g \oplus k\rho. We denote this fusion category by G+kG + k. For every finite abelian group GG, near-group fusion categories of type G+0G + 0 are exactly Tambara–Yamagami categories [12]. A natural question is therefore to determine which positive values of kk admit a near-group fusion category of type G+kG + k. Ostrik proved that, for G={e}G = \{e\} and G=Z/2ZG = \mathbb{Z}/2\mathbb{Z}, the admissible levels are precisely k∈{0,1}k \in\{0, 1\} and k∈{0,1,2}k \in\{0, 1, 2\}, respectively [7, 8]. For G=Z/3ZG = \mathbb{Z}/3\mathbb{Z}, Larson showed that the admissible levels are exactly k∈{0,2,3,6}k \in\{0, 2, 3, 6\} [5]. More recently, Schopieray classified the admissible levels for all nontrivial elementary abelian 2-groups [10 Theorem 1.1]: a near-group fusion category of type G+kG + k exists if and only if k=0k = 0, or G=Z/2ZG = \mathbb{Z}/2\mathbb{Z} and k∈{1,2}k \in\{1, 2\}, or G=Z/2Z×Z/2ZG = \mathbb{Z}/2\mathbb{Z} \times\mathbb{Z}/2\mathbb{Z} and k=4k = 4.

Let C\mathcal{C} be a near-group fusion category of type Z/4Z+k\mathbb{Z}/4\mathbb{Z} + k, O(C)={1,g,g2,g3,ρ}\mathcal{O}(\mathcal{C}) = \{1, g, g^{2}, g^{3}, \rho\} and fusion rules are defined as

g4=1,gjρ=ρgj=ρ,ρ2=∑j=03gj+kρ,k∈Z≥0.(1)g^{4} = 1,\qquad g^{j}\rho= \rho g^{j} = \rho,\qquad\rho^{2} = \sum_{j=0}^{3} g^{j} + k\rho,\qquad k \in\mathbb{Z}_{\geq0}. \tag*{(1)}

The main results of this paper are as follows:

Theorem 1.1 (Theorem 4.3). Let C\mathcal{C} be a fusion category with fusion rules (1). Then k∈{0,3,4}k \in\{0, 3, 4\}.

For k=0k = 0, these are the Tambara–Yamagami categories TY⁡(Z/4Z,χ,τ)\operatorname{TY}(\mathbb{Z}/4\mathbb{Z}, \chi, \tau), where χ\chi is a nondegenerate symmetric bicharacter on Z/4Z\mathbb{Z}/4\mathbb{Z} and τ=±12\tau= \pm\frac{1}{2} [12]. For k=3k = 3, a categorification is provided by Rep⁡(Z/5Z⋊Z/4Z)\operatorname{Rep}(\mathbb{Z}/5\mathbb{Z} \rtimes\mathbb{Z}/4\mathbb{Z}) and Z/4Z\mathbb{Z}/4\mathbb{Z} acts faithfully on Z/5Z\mathbb{Z}/5\mathbb{Z}. For k=4k = 4, Izumi’s construction gives unitary near-group categories of type Z/4Z+4\mathbb{Z}/4\mathbb{Z} + 4, two inequivalent unitary categorifications are recorded in [3 Table 2] and are complex conjugate to one another [4 Example 12.12]. Thus the possible values of kk in the pseudo-unitary setting are exactly 0, 3, and 4.

The method follows Larson’s use of central induction, twist traces, and sums of roots of unity [5]. We also use the dimension row formula recorded in [11 Lemma 2.1]. Its application here is an integrality obstruction for a fusion matrix. The numerical inequality in [11 Theorem 5.1] by itself does not exclude k=8k = 8; this distinction is explained in § 4.

Central induction and preliminary reductions

Standard identities

For objects U,VU,V in a fusion category C\mathcal{C}, let [U,V]:=dim⁡Hom⁡(U,V)[U,V] := \dim\operatorname{Hom}(U,V). We recall the following standard facts about a spherical fusion category and its Drinfeld center Z(C)\mathcal{Z}(\mathcal{C}); see [8 Section 2], [6].

Let F:Z(C)→CF:\mathcal{Z}(\mathcal{C}) \to\mathcal{C} be the forgetful functor, and let II be its right adjoint. By [2 Proposition 9.2.2], adjunction and the induction formula give

F(I(V))=⨁X∈O(C)X⊗V⊗X∗,[I(U),I(V)]=[U,F(I(V))].(2)F(I(V)) = \bigoplus_{X\in\mathcal{O}(\mathcal{C})} X \otimes V \otimes X^{*}, \qquad[I(U),I(V)] = [U,F(I(V))]. \tag*{(2)}

Let D:=dim⁡(C)D := \dim(\mathcal{C}). Notice that the scalar twists θX\theta_X of simple objects X∈O(Z(C))X \in\mathcal{O}(\mathcal{Z}(\mathcal{C})) are roots of unity, let νn(V)\nu_n(V) be the nn-th Frobenius-Schur indicator of object V∈CV \in\mathcal{C}, then [6 Theorem 4.1] states that

Tr⁡(θI(V))=D⋅νn(V)(n≥1),(3)\operatorname{Tr}(\theta_{I(V)}) = D \cdot\nu_n(V) \qquad(n \ge1), \tag*{(3)}

Let gg be an invertible object of order nn. Then the vector space Hom⁡(1,g⊗n)\operatorname{Hom}(1,g^{\otimes n}) is one-dimensional. Since νn(g)\nu_n(g) is the trace of an endomorphism of finite order on this space, it is a root of unity. Hence ∣νn(g)∣=1\lvert\nu_n(g)\rvert= 1. Since dim⁡I(g)=D\dim I(g) = D and dimensions of an objects are positive, Equation (3) forces θX=νn(g)\theta_X = \nu_n(g) for every simple summand XX of I(g)I(g).

Let C\mathcal{C} be a fusion category with Grothendieck ring K0(C)K_0(\mathcal{C}). Then the algebra K0(C)⊗ZCK_0(\mathcal{C}) \otimes_{\mathbb{Z}} \mathbb{C} is semisimple. Given an irreducible representation χ\chi of K0(C)⊗ZCK_0(\mathcal{C}) \otimes_{\mathbb{Z}} \mathbb{C}, there is a positive algebraic integer fχf_\chi, called the formal codegree of C\mathcal{C}, satisfying ∑χdim⁡χfχ=1\sum_\chi\frac{\dim\chi}{f_\chi}=1 by [8 proposition 2.10]. For example, FPdim⁡(C)\operatorname{FPdim}(\mathcal{C}) and dim⁡(C)\dim(\mathcal{C}) are formal codegrees of C\mathcal{C}. Moreover, [8 Theorem 2.13] gives the decomposition

I(1)=⨁χAχ,dim⁡Aχ=Dfχ,θAχ=1.(4)I(1) = \bigoplus_\chi A_\chi,\qquad\dim A_\chi= \frac{D}{f_\chi},\qquad\theta_{A_\chi}=1. \tag*{(4)}

Computation of the bi-adjoint functor and twists

Let C\mathcal{C} be a near-group fusion category of type Z/4Z+k\mathbb{Z}/4\mathbb{Z}+k, and let d=dim⁡(ρ)d=\dim(\rho). Then dd is the positive root of x2−kx−4x^2-kx-4, i.e. d=k+k2+162d=\frac{k+\sqrt{k^2+16}}{2}. Set D=dim⁡(C)=4+d2=8+kdD=\dim(\mathcal{C})=4+d^2=8+kd and its Galois conjugate D′=8+k(k−d)D'=8+k(k-d).

Since the fusion ring is commutative, its formal codegrees are the eigenvalues of multiplication by ∑X∈O(C)XX∗=4⋅1+ρ2\sum_{X\in\mathcal{O}(\mathcal{C})}XX^{*}=4\cdot1+\rho^2 on K0(C)⊗ZCK_0(\mathcal{C})\otimes_{\mathbb{Z}}\mathbb{C} [8 Remark 2.11]. With respect to the ordered basis (1,g,g2,g3,ρ)(1,g,g^2,g^3,\rho), this operator has matrix

R=(5111k1511k1151k1115kkkkk8+k2).R=\begin{pmatrix} 5 & 1 & 1 & 1 & k \\ 1 & 5 & 1 & 1 & k \\ 1 & 1 & 5 & 1 & k \\ 1 & 1 & 1 & 5 & k \\ k & k & k & k & 8+k^2 \end{pmatrix}.

Then the formal codegrees of C\mathcal{C} are precisely the roots of the characteristic polynomial of RR, which are D,D′,4,4,4D,D',4,4,4. In particular, Equation (4) means

dim⁡X≤D4,for all simple objects X⊂I(1).(5)\dim X \le\frac{D}{4},\qquad\text{for all simple objects }X\subset I(1). \tag*{(5)}

If dd is rational, it is an integer, then k=0,3k=0,3. Assume dd is irrational below, then it follows from [8 Theorem A.6(i)] that k=4tk=4t.

Proposition 2.1. Let CC be a near-group fusion category of type Z/4Z+k\mathbb{Z}/4\mathbb{Z}+k, then each I(gj)I(g^j) has exactly five distinct simple summands, for 0≤j≤30 \le j \le3.

Proof. For I(1)I(1), it is proved by Equation (4). For any h∈{g,g2,g3}h \in\{g,g^2,g^3\}, the induction formula gives

F(I(h))=4h⊕(⨁j=03gj)⊕kρ.(6)F(I(h)) = 4h \oplus\left(\bigoplus_{j=0}^{3}g^j\right) \oplus k\rho. \tag*{(6)}

Write I(h)=⨁ZmZZI(h)=\bigoplus_Z m_Z Z with mZ∈Z≥0m_Z \in\mathbb{Z}_{\ge0}. By definition, we have ∑ZmZ2=5\sum_Z m_Z^2=5. If I(h)I(h) were not multiplicity-free, the only possible decomposition would then be I(h)=X⊕2YI(h)=X\oplus2Y, X≇YX\not\cong Y, for some X,Y∈O(Z(C))X,Y\in\mathcal{O}(\mathcal{Z}(C)). Since [I(h),I(1)]=1[I(h),I(1)]=1 and I(1)I(1) is multiplicity-free, we have 1=[I(1),X]+2[I(1),Y]1=[I(1),X]+2[I(1),Y]. Consequently, X⊂I(1)X\subset I(1) and Y⊄I(1)Y\not\subset I(1). In particular, θX=1\theta_X=1.

By [6 Theorem 4.1], we know 0=ν1(h)dim⁡(C)=Tr⁡(θI(h))=dim⁡(X)+2θYdim⁡(Y)0=\nu_1(h)\dim(C)=\operatorname{Tr}(\theta_{I(h)})=\dim(X)+2\theta_Y\dim(Y). Hence, dim⁡(X)=∣dim⁡(X)∣=∣2dim⁡Y∣=2dim⁡(Y)\dim(X)=|\dim(X)|=|2\dim Y|=2\dim(Y) as CC is pseudo-unitary, consequently, θY=−1\theta_Y=-1. Since dim⁡I(h)=D\dim I(h)=D, it follows that D=dim⁡(X)+2dim⁡(Y)=2dim⁡(X)D=\dim(X)+2\dim(Y)=2\dim(X). Thus dim⁡(X)=D/2\dim(X)=D/2, contradicting Equation (5). This contradiction proves multiplicity-freeness for every hh.

Finally, Equation (6) shows that each induction has five distinct simple summands, any two different inductions have exactly one summand in common.

For all X∈CX\in C and Z∈Z(C)Z\in\mathcal{Z}(C), the adjointness condition gives [I(X),Z]=[X,F(Z)][I(X),Z]=[X,F(Z)].

Lemma 2.2. There are pairwise non-isomorphic simple objects 1,A,B01,B02,B03∈O(Z(C))1,A,B_{01},B_{02},B_{03}\in\mathcal{O}(\mathcal{Z}(C)) such that

I(1)≅1⊕A⊕B01⊕B02⊕B03.I(1)\cong1\oplus A\oplus B_{01}\oplus B_{02}\oplus B_{03}.

Their dimensions and forgetful images are

dim⁡A=1+td,F(A)≅1⊕tρ,dim⁡B0j=2+td,F(B0j)≅1⊕gj⊕tρ (1≤j≤3).\dim A=1+td,\qquad F(A)\cong1\oplus t\rho,\qquad\dim B_{0j}=2+td,\qquad F(B_{0j})\cong1\oplus g^j\oplus t\rho\ (1\le j\le3).

Proof. It follows from Equation (4) and Proposition 2.1 that the decomposition I(1)=1⊕A⊕C1⊕C2⊕C3I(1)=1\oplus A\oplus C_1\oplus C_2\oplus C_3 is multiplicity-free and

dim⁡A=DD′=d24=1+td,dim⁡Ci=D4=2+td.\dim A=\frac{D}{D'}=\frac{d^2}{4}=1+td,\qquad\dim C_i=\frac{D}{4}=2+td.

Notice that dd is irrational, so F(A)F(A) contains exactly one invertible summand and tt copies of ρ\rho, whereas each F(Ci)F(C_i) contains exactly two invertible summands and tt copies of ρ\rho. Meanwhile, Equation (2) implies F(I(1))≅4⋅1⊕(⨁j=03gj)⊕4tρF(I(1))\cong4\cdot1\oplus\left(\bigoplus_{j=0}^{3}g^j\right)\oplus4t\rho. Each of g,g2,g3g,g^2,g^3 occurs exactly once in this decomposition. Thus we may label the objects CiC_i as B01,B02,B03B_{01},B_{02},B_{03} so that F(B0j)≅1⊕gj⊕tρF(B_{0j})\cong1\oplus g^j\oplus t\rho (1≤j≤3)(1\le j\le3).

By Proposition 2.1, for any simple object X∈O(Z(C))X\in\mathcal{O}(\mathcal{Z}(C)), every invertible summand of F(X)F(X) has multiplicity at most one. Let k=4tk=4t with t≥1t\ge1 as in Section 2. Denote the simple objects of Z(C)\mathcal{Z}(C) by 1,A,Bij1,A,B_{ij} (0≤i<j≤3)(0\le i<j\le3), Pj,QjP_j,Q_j (1≤j≤3)(1\le j\le3), LαL_\alpha (α∈J)(\alpha\in\mathcal{J}), where J\mathcal{J} is a finite index set. Here BijB_{ij} denotes the unique common simple summand of I(gi)I(g^i) and I(gj)I(g^j); the six objects BijB_{ij} are shown to be pairwise non-isomorphic in the following proposition.

Proposition 2.3. There are integers p,q,r,s≥0p,q,r,s\ge0 and γα>0\gamma_\alpha>0, and η∈{i,−i}\eta\in\{i,-i\}, such that p+q=r+s=tp+q=r+s=t. The block above the twist row is the matrix WW of the forgetful functor FF, with entries WV,Y=[F(Y),V]W_{V,Y}=[F(Y),V]. Thus each column records the decomposition of F(Y)F(Y), by adjunction, each row records the decomposition of I(V)I(V). In particular, the first four rows determine the decompositions of I(1),I(g),I(g2),I(g3)I(1),I(g),I(g^2),I(g^3). The last

row records the twists and is not part of WW. The final column is repeated for every α∈J\alpha\in\mathcal{J}, and each θα\theta_{\alpha} is a root of unity. Zero entries are omitted.

1AAB01B_{01}B02B_{02}B03B_{03}B12B_{12}B13B_{13}P1P_1Q1Q_1B23B_{23}P2P_2Q2Q_2P3P_3Q3Q_3LαL_{\alpha}
111111
gg11111
g2g^211111
g3g^311111
ρ\rhottttttttttttppqqttrrssppqqγα\gamma_{\alpha}
θ\theta11111−1-1η\eta−η-\eta−η-\eta−1-111−η-\eta−η-\etaθα\theta_{\alpha}

Table 1.

Proof. By Proposition 2.1, I(gj)I(g^j) are multiplicity-free and each have five simple summands, and meet pairwise in exactly one summand. The decomposition of I(1)I(1) is obtained in Lemma 2.2, and I(gj)I(g^j) contains a simple summand whose twist is 1. The twist-power constraint from Section 2 gives

θY2=1(Y⊂I(g2)),θY4=1(Y⊂I(g) or I(g3)).\theta_Y^2=1 \qquad(Y\subset I(g^2)), \qquad\theta_Y^4=1 \qquad(Y\subset I(g)\text{ or }I(g^3)).

The three pairwise intersections among I(g),I(g2),I(g3)I(g),I(g^2),I(g^3) are either three distinct simple objects or a single simple object TT common to all three inductions. In the latter case, the invertible summands of F(T)F(T) are exactly g,g2,g3g,g^2,g^3. Moreover, I(g)I(g) consists of B01B_{01}, TT, and three further simple objects whose restrictions contain gg as their only invertible summand. Denote their twists by ξ1,ξ2,ξ3\xi_1,\xi_2,\xi_3. Since θT2=1\theta_T^2=1 and d∉Q(i)d\notin\mathbb{Q}(i), comparison of coefficients in Tr⁡(θI(g)2)=0\operatorname{Tr}(\theta_{I(g)}^2)=0 gives 2+3+ξ12+ξ22+ξ32=02+3+\xi_1^2+\xi_2^2+\xi_3^2=0, ξℓ2∈{1,−1}\xi_\ell^2\in\{1,-1\}, which is impossible. Thus the three intersections are distinct. Denote the common summand of I(gi)I(g^i) and I(gj)I(g^j) by BijB_{ij} for 1≤i<j≤31\leq i<j\leq3. The invertible summands of F(Bij)F(B_{ij}) are exactly gi,gjg^i,g^j. Each I(gj)I(g^j) therefore has exactly two remaining summands, denoted Pj,QjP_j,Q_j, whose restrictions contain gjg^j as their only invertible summand. This proves the first four rows of the table.

Set ε=θB12∈{1,−1}\varepsilon=\theta_{B_{12}}\in\{1,-1\}, η=θB13\eta=\theta_{B_{13}}, ξ=θP1\xi=\theta_{P_1}, and ξ′=θQ1\xi'=\theta_{Q_1}. The coefficients of 1 in the second and first twist traces of I(g)I(g) give, respectively, 4+2η2+ξ2+(ξ′)2=04+2\eta^2+\xi^2+(\xi')^2=0, 2+2ε+2η+ξ+ξ′=02+2\varepsilon+2\eta+\xi+\xi'=0. The first equation forces η2=ξ2=(ξ′)2=−1\eta^2=\xi^2=(\xi')^2=-1. The real part of the second then gives ε=−1\varepsilon=-1, and its imaginary part gives ξ=ξ′=−η\xi=\xi'=-\eta. The same argument for I(g3)I(g^3) yields θB23=−1\theta_{B_{23}}=-1 and θP3=θQ3=−η\theta_{P_3}=\theta_{Q_3}=-\eta. Finally, the coefficient of 1 in the first twist trace of I(g2)I(g^2) is 2−2−2+θP2+θQ2=02-2-2+\theta_{P_2}+\theta_{Q_2}=0, so θP2=θQ2=1\theta_{P_2}=\theta_{Q_2}=1.

Let c12,c13,p,qc_{12},c_{13},p,q be the multiplicities of ρ\rho in F(B12),F(B13),F(P1),F(Q1)F(B_{12}),F(B_{13}),F(P_1),F(Q_1), respectively. The ρ\rho coefficient in F(I(g))F(I(g)) and the dd-coefficients in the second and first twist traces give

t+c12+c13+p+q=4t,t+c12−c13−p−q=0,t−c12+η(c13−p−q)=0.\begin{aligned} t+c_{12}+c_{13}+p+q&=4t,\\ t+c_{12}-c_{13}-p-q&=0,\\ t-c_{12}+\eta(c_{13}-p-q)&=0. \end{aligned}

Hence, c12=c13=p+q=tc_{12}=c_{13}=p+q=t. Duality identifies B23=B12∗B_{23}=B_{12}^{*} and allows the labelling P3=P1∗,Q3=Q1∗P_3=P_1^{*},Q_3=Q_1^{*}, proving the corresponding entries in the ρ\rho row. Writing r,sr,s for the ρ\rho multiplicities of F(P2),F(Q2)F(P_2),F(Q_2), the ρ\rho coefficient of F(I(g2))F(I(g^2)) gives 3t+r+s=4t3t+r+s=4t. Thus

The restriction of the remaining simple objects LαL_{\alpha} (α∈J\alpha\in\mathcal{J}) do not contain invertible summands, so F(Lα)=γαρF(L_{\alpha})=\gamma_{\alpha}\rho for some γα∈Z>0\gamma_{\alpha}\in\mathbb{Z}_{>0}. Its twist θα\theta_{\alpha} is a root of unity. This completes the table.

Notice that ρ\rho is self-dual, so the second Frobenius–Schur indicator ν2(ρ)=±1\nu_2(\rho)=\pm1 [6].

Proposition 2.4. Let ν=ν2(ρ)∈{1,−1}\nu= \nu_{2}(\rho) \in\{1,-1\}, u=pqu = pq, v=rsv = rs, w=t2+1w = \sqrt{t^{2}+1}, N=∑αγα2N = \sum_{\alpha}\gamma_{\alpha}^{2}, Sj=∑αγα2θαjS_{j} = \sum_{\alpha}\gamma_{\alpha}^{2}\theta_{\alpha}^{j}, j=1,2j=1,2. Then we have

N=8+6t2+4u+2v.(7)N = 8 + 6t^{2} + 4u + 2v. \tag*{(7)}
S1=−t2+2v−2tw+η(t2−4u),(8)S_{1} = -t^{2} + 2v - 2tw + \eta(t^{2} - 4u), \tag*{(8)}
S2=−4(t−ν)w−4u+2v.(9)S_{2} = -4(t-\nu)w - 4u + 2v. \tag*{(9)}

In particular,

0≤u,v≤t24,N≤8+152t2.(10)0 \leq u,v \leq\frac{t^{2}}{4}, \qquad N \leq8 + \frac{15}{2}t^{2}. \tag*{(10)}

Proof. By Proposition 2.3, I(ρ)=tA⊕i<jtBi,j⊕p(P1⊕P3)⊕q(Q1⊕Q3)⊕rP2⊕sQ2⊕αγαLαI(\rho)=tA\oplus_{i<j}tB_{i,j}\oplus p(P_{1}\oplus P_{3})\oplus q(Q_{1}\oplus Q_{3})\oplus rP_{2}\oplus sQ_{2}\oplus_{\alpha}\gamma_{\alpha}L_{\alpha}. Equations (2) implies F(I(ρ))=4t(⊕j=03gj)⊕(8+16t2)ρF(I(\rho))=4t(\oplus_{j=0}^{3}g^{j})\oplus(8+16t^{2})\rho, which then gives

8+16t2=[F(I(ρ)),ρ]=[I(ρ),I(ρ)]=7t2+2(p2+q2)+r2+s2+N.8+16t^{2}=[F(I(\rho)),\rho]=[I(\rho),I(\rho)]=7t^{2}+2(p^{2}+q^{2})+r^{2}+s^{2}+N.

Using p2+q2=t2−2up^{2}+q^{2}=t^{2}-2u and r2+s2=t2−2vr^{2}+s^{2}=t^{2}-2v, we obtain Equation (7).

Substituting the twists from Proposition 2.3 into the identities Tr⁡(θI(ρ))\operatorname{Tr}(\theta I(\rho)) and Tr⁡(θ2I(ρ))\operatorname{Tr}(\theta^{2}I(\rho)) yields

0=4t+d(3t2−2v+η(4u−t2)+S1),νD=8t+d(4t2+4u−2v+S2).0=4t+d(3t^{2}-2v+\eta(4u-t^{2})+S_{1}), \qquad\nu D=8t+d(4t^{2}+4u-2v+S_{2}).

After simplification, we prove Equations (8) and (9). The Inequalities (10) follow from p+q=r+s=tp+q=r+s=t.

Bounds for sums of roots of unity

We establish the arithmetic estimates for the identities from § 2. We write φ\varphi for Euler’s function and μ\mu for the Möbius function, φ(1)=μ(1)=1\varphi(1)=\mu(1)=1, and let ζm=exp⁡(2πi/m)\zeta_m=\exp(2\pi i/m). For a prime pp and a positive integer mm, let vp(m)v_p(m) denote the exponent of pp in the prime factorization of mm.

Normalized Galois averages

Let c>1c>1 be square-free and set K=Q(c)K=\mathbb{Q}(\sqrt{c}). Its conductor is f=cf=c if c≡1(mod4)c\equiv1\pmod4, and f=4cf=4c if c≡2,3(mod4)c\equiv2,3\pmod4. In particular, K⊆Q(ζm)K\subseteq\mathbb{Q}(\zeta_m) if and only if f∣mf\mid m.

Choose a cyclotomic field EE containing KK and all roots under consideration. For w∈Ew\in E, define its normalized relative trace by

TK(w)=Tr⁡E/K(w)[E:K]=1∣G∣∑σ∈Gσ(w),G=Gal⁡(E/K).T_K(w)=\frac{\operatorname{Tr}_{E/K}(w)}{[E:K]}=\frac{1}{|G|}\sum_{\sigma\in G}\sigma(w), \qquad G=\operatorname{Gal}(E/K).

Obviously, TK(−)T_K(-) is KK-linear and satisfies TK(a)=aT_K(a)=a for a∈Ka\in K. Moreover, for every τ∈G\tau\in G, τ(TK(w))=1∣G∣∑σ∈G(τσ)(w)=TK(w)\tau(T_K(w))=\frac{1}{|G|}\sum_{\sigma\in G}(\tau\sigma)(w)=T_K(w) and TK(τ(w))=TK(w)T_K(\tau(w))=T_K(w), so TK(w)∈EG=KT_K(w)\in E^{G}=K.

The normalization makes TK(w)T_K(w) independent of the choice of cyclotomic field containing KK and ww. Indeed, if E⊆E′E\subseteq E' and w∈Ew\in E, transitivity of the trace gives Tr⁡E′/K(w)[E′:K]=Tr⁡E/K([E′:E]w)[E′:E][E:K]=Tr⁡E/K(w)[E:K]\frac{\operatorname{Tr}_{E'/K}(w)}{[E':K]}=\frac{\operatorname{Tr}_{E/K}([E':E]w)}{[E':E][E:K]}=\frac{\operatorname{Tr}_{E/K}(w)}{[E:K]}. Thus we may compute TK(w)T_K(w) in any cyclotomic field containing KK and ww. Since K⊂RK\subset\mathbb{R}, complex conjugation belongs to GG. Consequently

TK(Re⁡w)=TK(w+w‾2)=TK(w)+TK(w‾)2=TK(w).T_K(\operatorname{Re}w)=T_K\left(\frac{w+\overline{w}}{2}\right)=\frac{T_K(w)+T_K(\overline{w})}{2}=T_K(w).

If β\beta is a primitive rrth root of unity, then

Tr⁡Q(β)/Q(β)φ(r)=1φ(r)∑1≤j≤r(j,r)=1βj=μ(r)φ(r).\frac{\operatorname{Tr}_{\mathbb{Q}(\beta)/\mathbb{Q}}(\beta)}{\varphi(r)}=\frac{1}{\varphi(r)}\sum_{\substack{1\le j\le r\\(j,r)=1}}\beta^j=\frac{\mu(r)}{\varphi(r)}.

Since β\beta is a primitive rrth root of unity, the classical evaluation of Ramanujan sums [9 Section 1] gives ∑1≤j≤r(j,r)=1βj=μ(r)\sum_{\substack{1\le j\le r\\(j,r)=1}}\beta^j=\mu(r). For every positive integer rr, we have ∣μ(r)∣≤1|\mu(r)|\le1 and φ(r)≥1\varphi(r)\ge1. Hence μ(r)φ(r)≤∣μ(r)∣φ(r)≤1\frac{\mu(r)}{\varphi(r)}\le\frac{|\mu(r)|}{\varphi(r)}\le1, with equality if and only if r=1r=1.

Let ℓ,r\ell,r be positive integers such that f∣ℓf\mid\ell and (ℓ,r)=1(\ell,r)=1, and put K1=Q(ζℓ)K_1=\mathbb{Q}(\zeta_\ell) and K2=Q(ζr)K_2=\mathbb{Q}(\zeta_r). Let K1K2K_1K_2 denote the compositum of K1K_1 and K2K_2, namely the smallest subfield of C\mathbb{C} containing both. By the Chinese remainder theorem, Gal⁡(K1K2/Q)≅Gal⁡(K1/Q)×Gal⁡(K2/Q)\operatorname{Gal}(K_1K_2/\mathbb{Q})\cong\operatorname{Gal}(K_1/\mathbb{Q})\times\operatorname{Gal}(K_2/\mathbb{Q}). Since K⊆K1K\subseteq K_1, an automorphism of K1K2K_1K_2 fixes KK if and only if its restriction to K1K_1 fixes KK. Its restriction to K2K_2 can be any element of Gal⁡(K2/Q)\operatorname{Gal}(K_2/\mathbb{Q}). Hence Gal⁡(K1K2/K)≅Gal⁡(K1/K)×Gal⁡(K2/Q)\operatorname{Gal}(K_1K_2/K)\cong\operatorname{Gal}(K_1/K)\times\operatorname{Gal}(K_2/\mathbb{Q}). Writing these two factors as G1G_1 and G2G_2, respectively, for u∈K1u\in K_1 and v∈K2v\in K_2, we obtain the formula

TK(uv)=1∣G1∣∣G2∣∑σ∈G1∑τ∈G2σ(u)τ(v)=(1∣G1∣∑σ∈G1σ(u))(1∣G2∣∑τ∈G2τ(v))=TK(u)Tr⁡K2/Q(v)[K2:Q].\begin{aligned} T_K(uv)=\frac{1}{|G_1||G_2|}\sum_{\sigma\in G_1}\sum_{\tau\in G_2}\sigma(u)\tau(v) \\ &=\left(\frac{1}{|G_1|}\sum_{\sigma\in G_1}\sigma(u)\right)\left(\frac{1}{|G_2|}\sum_{\tau\in G_2}\tau(v)\right)=T_K(u)\frac{\operatorname{Tr}_{K_2/\mathbb{Q}}(v)}{[K_2:\mathbb{Q}]}. \end{aligned}

We will use this factorization only for coprime cyclotomic factors.

For a root of unity ξ\xi, write TK(ξ)=xξ+yξcT_K(\xi)=x_\xi+y_\xi\sqrt{c}, xξ,yξ∈Qx_\xi,y_\xi\in\mathbb{Q}, and let m=ord⁡(ξ)m=\operatorname{ord}(\xi). If f∤mf\nmid m, then K⊈Q(ξ)K\nsubseteq\mathbb{Q}(\xi) and K∩Q(ξ)=QK\cap\mathbb{Q}(\xi)=\mathbb{Q}. Restriction identifies Gal⁡(KQ(ξ)/K)\operatorname{Gal}(K\mathbb{Q}(\xi)/K) with Gal⁡(Q(ξ)/Q)\operatorname{Gal}(\mathbb{Q}(\xi)/\mathbb{Q}), so the KK-orbit of ξ\xi contains all primitive mmth roots. Therefore TK(ξ)=μ(m)φ(m)∈QT_K(\xi)=\frac{\mu(m)}{\varphi(m)}\in\mathbb{Q}.

Next assume that f∣mf\mid m and that vp(m)>vp(f)v_p(m)>v_p(f) for a prime factor pp of ff. These assumptions imply p2∣mp^2\mid m and f∣m/pf\mid m/p. Every prime divisor of mm divides m/pm/p, so a=1+m/pa=1+m/p is relatively prime to mm. The automorphism σ(ξ)=ξa\sigma(\xi)=\xi^a of Q(ξ)\mathbb{Q}(\xi) fixes KK, because a≡1(modf)a\equiv1\pmod f. Let ω=ξm/p\omega=\xi^{m/p}. Then σ(ξ)=ωξ\sigma(\xi)=\omega\xi and σ(ω)=ω\sigma(\omega)=\omega, since p∣m/pp\mid m/p. It follows by induction that σj(ξk)=ωjkξk\sigma^j(\xi^k)=\omega^{jk}\xi^k. For every positive integer kk with p∤kp\nmid k, trace invariance gives

pTK(ξk)=TK(pξk)=∑j=0p−1TK(σj(ξk))=TK(ξk∑j=0p−1ωjk)=0.(11)pT_K(\xi^k)=T_K(p\xi^k)=\sum_{j=0}^{p-1}T_K(\sigma^j(\xi^k))=T_K\left(\xi^k\sum_{j=0}^{p-1}\omega^{jk}\right)=0. \tag*{(11)}

In particular, TK(ξ)=0T_K(\xi)=0. Combining rationality with this cancellation proves the necessary conditions: if yξ≠0y_\xi\ne0, then f∣mf\mid m and vp(m)=vp(f)v_p(m)=v_p(f) for every prime p∣fp\mid f.

We use the following theorem without proof.

Theorem 3.1 ([5 Theorem 6.2]). Let c>1c>1 be square-free and let a,b∈Za,b\in\mathbb{Z}. If roots of unity ξ1,…,ξn\xi_1,\ldots,\xi_n satisfy ∑j=1nξj=a+bc\sum_{j=1}^{n}\xi_j=a+b\sqrt{c}, then n≥∣b∣φ(2c)n\ge|b|\varphi(2c).

The following lemma is a direct consequence of [1 Lemma 3.11].

Lemma 3.2. Let c>1c>1 be even and square-free. Suppose that roots of unity ξ1,…,ξn\xi_1,\ldots,\xi_n satisfy

Re⁡∑j=1nξj=x+yc,∑j=1nξj2=z+qc,x,y,z,q∈Z.\operatorname{Re}\sum_{j=1}^{n}\xi_j=x+y\sqrt{c},\qquad\sum_{j=1}^{n}\xi_j^2=z+q\sqrt{c},\qquad x,y,z,q\in\mathbb{Z}.

Then n≥φ(2c)(∣y∣+∣q∣)n\ge\varphi(2c)(|y|+|q|).

A joint estimate over Q(5)\mathbb{Q}(\sqrt{5})

Throughout this subsection, put K=Q(5)K = \mathbb{Q}(\sqrt{5}) and T=TKT = T_K. Let ζ=ζ5\zeta= \zeta_5. The extension Q(ζ)/K\mathbb{Q}(\zeta)/K has degree 2, with complex conjugation as its nontrivial automorphism. Its two orbits of primitive fifth roots are therefore {ζ,ζ−1}\{\zeta,\zeta^{-1}\} and {ζ2,ζ−2}\{\zeta^2,\zeta^{-2}\}. Squaring interchanges them, so for every primitive fifth root β\beta there is a sign ε∈{1,−1}\varepsilon\in\{1,-1\} such that T(β)=−1+ε54T(\beta) = \frac{-1+\varepsilon\sqrt{5}}{4}, T(β2)=−1−ε54T(\beta^2) = \frac{-1-\varepsilon\sqrt{5}}{4}. The two signs are linked by the choice of β\beta.

Lemma 3.3. Let ξ1,…,ξn\xi_1,\ldots,\xi_n be roots of unity such that

Re⁡∑j=1nξj=x+y5,∑j=1nξj2=z+q5,x,y,z,q∈Z.\operatorname{Re}\sum_{j=1}^{n}\xi_j=x+y\sqrt{5},\qquad\sum_{j=1}^{n}\xi_j^2=z+q\sqrt{5},\qquad x,y,z,q\in\mathbb{Z}.

Then n≥−x−y−4qn\ge-x-y-4q.

Proof. For a root of unity ξ\xi, write T(ξ):=xξ+yξ5T(\xi):=x_\xi+y_\xi\sqrt{5}, T(ξ2):=zξ+qξ5T(\xi^2):=z_\xi+q_\xi\sqrt{5}, ψ(ξ):=−xξ−yξ−4qξ\psi(\xi):=-x_\xi-y_\xi-4q_\xi. We first prove ψ(ξ)≤1\psi(\xi)\le1 by considering s=ord⁡(ξ)s=\operatorname{ord}(\xi).

If 5∤s5\nmid s, then neither ss nor ord⁡(ξ2)\operatorname{ord}(\xi^2) is divisible by the conductor 5. Both averages are rational, so yξ=qξ=0y_\xi=q_\xi=0. Moreover, xξx_\xi is the real part of an average of roots of unity, and hence xξ≥−1x_\xi\ge-1. Thus ψ(ξ)=−xξ≤1\psi(\xi)=-x_\xi\le1. If 25∣s25\mid s, the cancellation argument in Subsection 3.1 and Equation (11), with p=5p=5 and k=1,2k=1,2, gives T(ξ)=T(ξ2)=0T(\xi)=T(\xi^2)=0. Hence ψ(ξ)=0\psi(\xi)=0 in this case.

In the remaining case 5∣s5\mid s but 25∤s25\nmid s. Removing the power of 2 and the single factor 5 gives s=2a5rs=2^a5^r, a=v2(s)≥0a=v_2(s)\ge0, (r,10)=1(r,10)=1. Suppose first that a≥3a\ge3. The integers b1=1+s/2b_1=1+s/2 and b2=1+s/4b_2=1+s/4 are odd and are 1 modulo every odd prime divisor of ss. Thus they are relatively prime to ss. They are also 1 modulo 5, so the automorphisms σi(ξ)=ξbi\sigma_i(\xi)=\xi^{b_i} fix KK. Since ξs/2=−1\xi^{s/2}=-1, we have σ1(ξ)=−ξ\sigma_1(\xi)=-\xi, σ2(ξ2)=−ξ2\sigma_2(\xi^2)=-\xi^2. Since T(σ(w))=T(w)T(\sigma(w))=T(w) for every automorphism σ\sigma fixing KK, we obtain

T(ξ)=T(σ1(ξ))=T(−ξ)=−T(ξ),T(ξ2)=T(σ2(ξ2))=T(−ξ2)=−T(ξ2).T(\xi)=T(\sigma_1(\xi))=T(-\xi)=-T(\xi),\qquad T(\xi^2)=T(\sigma_2(\xi^2))=T(-\xi^2)=-T(\xi^2).

Thus T(ξ)=T(ξ2)=0T(\xi)=T(\xi^2)=0.

It remains to treat a∈{0,1,2}a\in\{0,1,2\}. Let ℓ=2a5\ell=2^a5. Since (ℓ,r)=1(\ell,r)=1, choose integers b,eb,e satisfying br+eℓ=1br+e\ell=1 and set δ=ξbr\delta=\xi^{br}, γ=ξeℓ\gamma=\xi^{e\ell}. Then ξ=δγ\xi=\delta\gamma, and the congruences br≡1(modℓ)br\equiv1\pmod{\ell} and eℓ≡1(modr)e\ell\equiv1\pmod r show that δ\delta and γ\gamma are primitive roots of orders ℓ\ell and rr, respectively. For r=1r=1, this means γ=1\gamma=1. Because rr is odd, squaring permutes its primitive roots. Thus γ\gamma and γ2\gamma^2 have the same rational average λ=μ(r)/φ(r)\lambda=\mu(r)/\varphi(r). The coprime factorization from Subsection 3.1 gives

T(ξ)=λT(δ),T(ξ2)=λT(δ2),ψ(ξ)=λψ(δ).T(\xi)=\lambda T(\delta),\qquad T(\xi^2)=\lambda T(\delta^2),\qquad\psi(\xi)=\lambda\psi(\delta).

The last identity follows by comparing the rational and 5\sqrt{5} coefficients in the first two identities. Furthermore, −12≤λ≤1-\frac{1}{2}\le\lambda\le1. Indeed, ∣μ(r)∣≤1|\mu(r)|\le1. If μ(r)=−1\mu(r)=-1, then the odd integer rr is greater than 1. Hence 1 and r−1r-1 are distinct and both coprime to rr, so φ(r)≥2\varphi(r)\ge2.

We now compute ψ(δ)\psi(\delta) for the three possible orders. For ℓ=5\ell=5, take δ=β\delta=\beta in the fifth-root formulas above. For ℓ=10\ell=10, write δ=−β\delta=-\beta with β\beta primitive of order 5; then T(δ)=−T(β)T(\delta)=-T(\beta) and T(δ2)=T(β2)T(\delta^2)=T(\beta^2). For ℓ=20\ell=20, put η=δ5\eta=\delta^5 and β=δ−4\beta=\delta^{-4}. These roots have orders 4 and 5, satisfy δ=ηβ\delta=\eta\beta, and have η2=−1\eta^2=-1. The rational average of η\eta is zero, so coprime factorization yields

T(δ)=0,T(δ2)=−T(β2)=1+ε54.T(\delta)=0,\qquad T(\delta^2)=-T(\beta^2)=\frac{1+\varepsilon\sqrt{5}}{4}.

Substitution into the definition of ψ\psi gives the following table:

ℓ\ellxδx_\deltayδy_\deltaqδq_\deltaψ(δ)\psi(\delta)
5−14-\frac{1}{4}ε4\frac{\varepsilon}{4}−ε4-\frac{\varepsilon}{4}1+3ε4\frac{1+3\varepsilon}{4}
1014\frac{1}{4}−ε4-\frac{\varepsilon}{4}−ε4-\frac{\varepsilon}{4}−1+5ε4\frac{-1+5\varepsilon}{4}
2000ε4\frac{\varepsilon}{4}−ε-\varepsilon

Table 2.

( ε∈{1,−1}\varepsilon\in\{1,-1\}).

In each row the sign is determined by its associated fifth root β\beta. These values give −3/2≤ψ(δ)≤1-3/2 \le\psi(\delta) \le1. If λ≥0\lambda\ge0, then ψ(ξ)=λψ(δ)≤λ≤1\psi(\xi) = \lambda\psi(\delta) \le\lambda\le1. If λ<0\lambda< 0, multiplication reverses the lower bound on ψ(δ)\psi(\delta), and hence ψ(ξ)=λψ(δ)≤−32λ≤34<1\psi(\xi) = \lambda\psi(\delta) \le-\frac{3}{2}\lambda\le\frac{3}{4} < 1. This proves ψ(ξ)≤1\psi(\xi) \le1 for every root of unity.

Finally, use one common cyclotomic field for all the given roots. The identity T(w)=T(Re⁡w)T(w) = T(\operatorname{Re} w) and the hypotheses give

∑j=1nT(ξj)=T(Re⁡∑j=1nξj)=T(x+y5)=x+y5,∑j=1nT(ξj2)=T(z+q5)=z+q5.\sum_{j=1}^{n} T(\xi_j) = T\left(\operatorname{Re}\sum_{j=1}^{n}\xi_j\right) = T(x+y\sqrt{5}) = x+y\sqrt{5}, \qquad\sum_{j=1}^{n} T(\xi_j^2) = T(z+q\sqrt{5}) = z+q\sqrt{5}.

Since 1,51,\sqrt{5} are linearly independent over Q\mathbb{Q}, comparison of coefficients gives ∑jxξj=x\sum_j x_{\xi_j}=x, ∑jyξj=y\sum_j y_{\xi_j}=y, ∑jqξj=q\sum_j q_{\xi_j}=q. Therefore, −x−y−4q=∑j=1n(−xξj−yξj−4qξj)=∑j=1nψ(ξj)≤n-x-y-4q=\sum_{j=1}^{n}(-x_{\xi_j}-y_{\xi_j}-4q_{\xi_j})=\sum_{j=1}^{n}\psi(\xi_j)\le n.

Lemma 3.4. If there are roots of unity ξj\xi_j (1≤j≤161\le j\le16) such that ∑j=116ξj=−4−45\sum_{j=1}^{16}\xi_j=-4-4\sqrt{5}, then their multiset consists of eight copies of ζ52\zeta_5^2 and eight copies of ζ53\zeta_5^3. Consequently, ∑j=116ξj2=−4+45\sum_{j=1}^{16}\xi_j^2=-4+4\sqrt{5}.

Proof. Let T(ξ)=xξ+yξ5T(\xi)=x_\xi+y_\xi\sqrt{5}. We first identify all roots for which yξ=−1/4y_\xi=-1/4. If 5∤ord⁡(ξ)5\nmid\operatorname{ord}(\xi), then yξ=0y_\xi=0; if 25∣ord⁡(ξ)25\mid\operatorname{ord}(\xi), then T(ξ)=0T(\xi)=0. Otherwise ord⁡(ξ)=5r\operatorname{ord}(\xi)=5r with (r,5)=1(r,5)=1. Choose integers u,vu,v such that ur+5v=1ur+5v=1 and let α=ξur\alpha=\xi^{ur}, β=ξ5v\beta=\xi^{5v}. Then ξ=αβ\xi=\alpha\beta, with α\alpha and β\beta primitive of orders 55 and rr. The coprime factorization in Subsection 3.1 and the fifth-root calculation in Subsection 3.2 give

T(ξ)=T(α)μ(r)φ(r)=μ(r)φ(r)(−1+ε54),ε∈{1,−1}.T(\xi)=T(\alpha)\frac{\mu(r)}{\varphi(r)}=\frac{\mu(r)}{\varphi(r)}\left(\frac{-1+\varepsilon\sqrt{5}}{4}\right),\qquad\varepsilon\in\{1,-1\}.

Here rr need not be odd. In particular, yξ=εμ(r)4φ(r)≥−14y_\xi=\frac{\varepsilon\mu(r)}{4\varphi(r)}\ge-\frac{1}{4}. Equality requires ∣μ(r)∣=φ(r)=1|\mu(r)|=\varphi(r)=1, which is possible only when r=1r=1 or r=2r=2. For r=1r=1, equality requires ε=−1\varepsilon=-1, giving the roots ζ52,ζ53\zeta_5^2,\zeta_5^3 and rational coefficient xξ=−1/4x_\xi=-1/4. For r=2r=2, we have μ(2)=−1\mu(2)=-1, so equality requires ε=1\varepsilon=1, giving the roots −ζ5,−ζ54-\zeta_5,-\zeta_5^4 and rational coefficient xξ=1/4x_\xi=1/4.

Averaging the assumed identity and comparing coefficients yields ∑j=116yξj=−4\sum_{j=1}^{16}y_{\xi_j}=-4, ∑j=116xξj=−4\sum_{j=1}^{16}x_{\xi_j}=-4. The first equality is a sum of sixteen terms each at least −1/4-1/4, so every term equals −1/4-1/4. Let aa and bb count the roots in {ζ52,ζ53}\{\zeta_5^2,\zeta_5^3\} and {−ζ5,−ζ54}\{-\zeta_5,-\zeta_5^4\}, respectively, including multiplicities. Then the number of roots and the rational coefficients give a+b=16a+b=16, −a4+b4=−4-\frac{a}{4}+\frac{b}{4}=-4. Thus a=16a=16, b=0b=0, and every original root is either ζ52\zeta_5^2 or ζ53\zeta_5^3. Let their multiplicities be a2,a3a_2,a_3. The original sum is real, and conjugation exchanges these two roots. Subtracting the conjugate sum gives (a2−a3)(ζ52−ζ53)=0(a_2-a_3)(\zeta_5^2-\zeta_5^3)=0. Since the roots are distinct, a2=a3=8a_2=a_3=8. Finally, ∑j=116ξj2=8(ζ54+ζ56)=8(ζ54+ζ5−1)=−4+45\sum_{j=1}^{16}\xi_j^2=8(\zeta_5^4+\zeta_5^6)=8(\zeta_5^4+\zeta_5^{-1})=-4+4\sqrt{5}.

Proof of the main theorem

Proposition 4.1. Let CC be a near-group fusion category of type Z/4Z+k\mathbb{Z}/4\mathbb{Z}+k, k=4tk=4t. Then t≤2t\le2.

Proof. Write t2+1=m2ct^2+1=m^2c, where m∈Z≥1m\in\mathbb{Z}_{\ge1} and cc is square-free. For every odd prime divisor pp of cc, we have t2≡−1(modp)t^2\equiv-1\pmod p and p∤tp\nmid t. Fermat’s little theorem gives 1≡tp−1=(−1)(p−1)/2(modp)1\equiv t^{p-1}=(-1)^{(p-1)/2}\pmod p. Thus (p−1)/2(p-1)/2 is even, so p≡1(mod4)p\equiv1\pmod4. If cc is odd, it is a product of such primes, and hence c≡1(mod4)c\equiv1\pmod4. By Proposition 2.4, we have

Re⁡S1=−t2+2v−2tmc,S2=−4u+2v−4(t−v)mc.(12)\operatorname{Re}S_1=-t^2+2v-2tm\sqrt{c},\qquad S_2=-4u+2v-4(t-v)m\sqrt{c}. \tag*{(12)}

Case 1: cc is odd and c≠5c\ne5. Every prime divisor of cc is congruent to 11 modulo 44. Since c>1c>1 is square-free and c≠5c\ne5, it has a prime divisor p0≠5p_0\ne5, and hence p0≥13p_0\ge13. Moreover, (p−1)/p>1(p-1)/\sqrt{p}>1 for every prime divisor pp of cc. Therefore, φ(2c)c=∏p∣cp−1p≥p0−1p0≥1213\frac{\varphi(2c)}{\sqrt{c}}=\prod_{p\mid c}\frac{p-1}{\sqrt{p}}\ge\frac{p_0-1}{\sqrt{p_0}}\ge\frac{12}{\sqrt{13}}. Note ν≤1\nu\le1, applying Theorem 3.1 to S2S_2, we get

N≥4(t−1)mφ(2c)≥4813(t−1)t2+1>13t(t−1).N\ge4(t-1)m\varphi(2c)\ge\frac{48}{\sqrt{13}}(t-1)\sqrt{t^2+1}>13t(t-1).

An elementary calculation shows 13t(t−1)>8+15t2/213t(t-1)>8+15t^2/2 for t≥3t\ge3, which contradicts Inequality (10).

Case 2: cc is even and c≠2c\ne2. Write c=2hc=2h with h>1h>1. Then the smallest possible odd prime divisor of hh is 55, so φ(2c)c=2∏p∣hp−1p≥810>52\frac{\varphi(2c)}{\sqrt{c}}=\sqrt{2}\prod_{p\mid h}\frac{p-1}{\sqrt{p}}\ge\frac{8}{\sqrt{10}}>\frac{5}{2}. Apply Lemma 3.2 to (12), we obtain

N≥(6t−4ν)mφ(2c)≥(6t−4)mφ(2c).(13)N\ge(6t-4\nu)m\varphi(2c)\ge(6t-4)m\varphi(2c). \tag*{(13)}

Consequently, N>52(6t−4)t=15t2−10tN>\frac{5}{2}(6t-4)t=15t^2-10t. For t≥3t\ge3, this is larger than 8+15t2/28+15t^2/2, impossible.

Case 3: c=2c=2. Inequality (13) implies N≥2(6t−4)t2+1>2(6t−4)tN\ge\sqrt{2}(6t-4)\sqrt{t^2+1}>\sqrt{2}(6t-4)t. For every t≥7t\ge7, 2(6t−4)t>8+15t2/2\sqrt{2}(6t-4)t>8+15t^2/2. Indeed, the difference is (62−15/2)t2−42t−8(6\sqrt{2}-15/2)t^2-4\sqrt{2}t-8; at t=7t=7 it is 2662−751/2>0266\sqrt{2}-751/2>0, and its derivative is positive for t≥7t\ge7. Thus t≤6t\le6. None of t=3,4,5,6t=3,4,5,6 has t2+1t^2+1 equal to twice a square.

Case 4: c=5c=5. Lemma 3.3 gives N≥t2−2ν+(18t−16ν)mN\ge t^2-2\nu+(18t-16\nu)m. Using ν≤1\nu\le1, the formula for NN, and the bounds on u,vu,v, we obtain

(18t−16)m≤8+5t2+4u+4v≤8+7t2=35m2+1.(14)(18t-16)m\le8+5t^2+4u+4v\le8+7t^2=35m^2+1. \tag*{(14)}

If m≥2m\ge2, then t2=5m2−1>4m2t^2=5m^2-1>4m^2. Since t,mt,m are integers, this implies t≥2m+1t\ge2m+1. But then

(18t−16)m≥(18(2m+1)−16)m=36m2+2m>35m2+1,(18t-16)m\ge(18(2m+1)-16)m=36m^2+2m>35m^2+1,

contradicting Inequality (14). Thus m=1m=1 and t=2t=2.

In summary, we conclude that t≤2t\le2.

We show that k=8k=8 can be excluded in the following proposition.

Proposition 4.2. The case k=8k=8 is impossible.

Proof. Suppose k=8k=8, and retain the notations of § 2. Set d=4+25d=4+2\sqrt{5}, a=1+2da=1+2d, b=1+db=1+d, D=8bD=8b. Then d2=4ad^2=4a, b2=5ab^2=5a, D2=320aD^2=320a, and a+a−1=18a+a^{-1}=18. Also u=pqu=pq, v=rs∈{0,1}v=rs\in\{0,1\}, since p+q=r+s=2p+q=r+s=2. Let ss denote the normalized SS-matrix of the center Z(C)\mathcal{Z}(\mathcal{C}), with s1,X=(dim⁡X)/Ds_{1,X}=(\dim X)/D.

Case 1: v=1v=1. If u=0u=0, there is a central invertible object ZZ with F(Z)=gF(Z)=g, and hence F(Z2)=g2F(Z^2)=g^2. Since Z2Z^2 is simple, the restriction table forces it to be either P2P_2 or Q2Q_2. However, r=s=1r=s=1 gives F(P2)=F(Q2)=g2⊕φF(P_2)=F(Q_2)=g^2\oplus\varphi, a contradiction. Thus u=v=1u=v=1.

For the nontrivial automorphism σ\sigma of Q(5)\mathbb{Q}(\sqrt{5}), the dimension-row formula [11 Lemma 2.1] gives sA,X=σ(dim⁡X/D)s_{A,X}=\sigma(\dim X/D). Apart from 1,A1,A, the dimensions of simple objects of Z(C)\mathcal{Z}(\mathcal{C}) are bb, 2b2b, or γαd\gamma\alpha d. Since σ(b/D)=b/D\sigma(b/D)=b/D and σ(d/D)=−d/D\sigma(d/D)=-d/D, it follows that sA,A=1Ds_{A,A}=\frac{1}{D}, sA,X2=s1,X2s_{A,X}^2=s_{1,X}^2 (X∉{1,A})(X\notin\{1,A\}). Let PP be a simple object of dimension bb. Then sP,1=sP,A=1/8s_{P,1}=s_{P,A}=1/8. The Verlinde formula and orthogonality of the s1,−s_{1,-} and sP,−s_{P,-} give NA,AP=∑XsA,X2−s1,X2s1,XsP,X=a2−1+a−1−a8D=16(a+1)8D=12N^P_{A,A}=\sum_X\frac{s_{A,X}^2-s_{1,X}^2}{s_{1,X}}s_{P,X}=\frac{a^2-1+a^{-1}-a}{8D}=\frac{16(a+1)}{8D}=\frac{1}{2}, where the last equalities use a+a−1=18a+a^{-1}=18 and D=4(a+1)D=4(a+1), it is a contradiction.

Case 2: v=0v=0. Relabel the simple objects so that r=0r=0, and let h=P2h=P_2. Then F(h)=g2F(h)=g^2, θh=1\theta_h=1, and h2=1h^2=1, since the unit is the only central simple object restricting to 11. For each simple object XX, define εX\varepsilon_X by cX,h∘ch,X=εXid⁡h⊗Xc_{X,h}\circ c_{h,X}=\varepsilon_X\operatorname{id}_{h\otimes X}, where cc denotes the braiding. The hexagon and balancing equation give εX∈{1,−1}\varepsilon_X\in\{1,-1\}, θh⊗X=εXθX\theta_{h\otimes X}=\varepsilon_X\theta_X, sh,X=εXdim⁡XDs_{h,X}=\varepsilon_X\frac{\dim X}{D}. Since h≠1h\ne1, orthogonality of sh,−s_{h,-} and s1,−s_{1,-} implies ∑εX=−1dim⁡(X)2=D22=160a\sum_{\varepsilon_X=-1}\dim(X)^2=\frac{D^2}{2}=160a. Tensoring by hh maps each invertible summand gjg^j of F(X)F(X) to gj+2g^{j+2}, with exponents taken modulo 4. The table of Proposition 2.3 shows that it preserves the twists of objects whose restrictions contain exactly one invertible summand. It also fixes B02B_{02} and B13B_{13}, since each is uniquely determined by the invertible summands of its restriction. All these objects have positive sign. Tensoring by hh exchanges B01B_{01} with B23B_{23} and B03B_{03} with B12B_{12}, and the two objects in each pair have opposite twists. Thus, among the previously determined objects, precisely these four have negative sign, and each has dimension 2b2b.

Let

N±=∑εLα=±1γα2,Sj,±=∑εLα=±1γα2θαj(j=1,2).N_{\pm}=\sum_{\varepsilon_{L_\alpha}=\pm1}\gamma_{\alpha}^{2},\qquad S_{j,\pm}=\sum_{\varepsilon_{L_\alpha}=\pm1}\gamma_{\alpha}^{2}\theta_{\alpha}^{j}\quad(j=1,2).

The preceding dimension count and the identity N=32+4uN=32+4u of Equation (7) give

N−=160a−4(2b)2d2=20,N+=12+4u≤16.(15)N_{-}=\frac{160a-4(2b)^{2}}{d^{2}}=20,\qquad N_{+}=12+4u\leq16. \tag*{(15)}

Tensoring by hh pairs the negative-sign LαL_\alpha with equal multiplicities γα\gamma_\alpha and opposite twists. Indeed, g2ρ=ρg^{2}\rho=\rho, and a fixed point would have θLα=−θLα\theta_{L_\alpha}=-\theta_{L_\alpha}. Hence S1,−=0S_{1,-}=0, and the formula for S1S_{1} gives

S1,+=S1=−4−45+4η(1−u).S_{1,+}=S_{1}=-4-4\sqrt{5}+4\eta(1-u).

Applying Theorem 3.1 to S1,++S1,+‾=−8−85S_{1,+}+\overline{S_{1,+}}=-8-8\sqrt{5} gives 2N+≥8φ(10)=322N_{+}\geq8\varphi(10)=32. Thus N+=16N_{+}=16 and u=1u=1. Lemma 3.4 now applies to S1,+=−4−45S_{1,+}=-4-4\sqrt{5} and gives S2,+=−4+45S_{2,+}=-4+4\sqrt{5}. Using the formula for S2S_{2} in Proposition 2.4 with u=1u=1 and v=0v=0, and subtracting the preceding expression for S2,+S_{2,+}, we obtain S2,−=S2−S2,+=−4(3−ν)5S_{2,-}=S_{2}-S_{2,+}=-4(3-\nu)\sqrt{5}. This is a sum of N−=20N_{-}=20 roots of unity, counted with multiplicity. Theorem 3.1 therefore implies 20≥4(3−ν)φ(10)=16(3−ν)≥3220\geq4(3-\nu)\varphi(10)=16(3-\nu)\geq32, a contradiction. □

Proposition 4.1 and Proposition 4.2 together imply the following theorem.

Theorem 4.3. There is a near-group fusion category of type Z/4Z+k\mathbb{Z}/4\mathbb{Z}+k if and only if k∈{0,3,4}k\in\{0,3,4\}.

Acknowledgements

Z. Yu is supported by NSFC (no.12571041) and Qinglan Project of Yangzhou University. The authors acknowledge the use of AI as an auxiliary tool during the preparation of this manuscript, including for mathematical discussion and editorial assistance. All mathematical content was independently verified by the authors, who take full responsibility for the final manuscript.

References

Yue Meng Email: [email protected] School of Mathematics, Yangzhou University, Yangzhou 225002, China

Zhiqiang Yu Email: [email protected] School of Mathematics, Yangzhou University, Yangzhou 225002, China

References

  1. [1]J. Dong, S. Palcoux and K. Wang, Grothendieck rings of near-Tambara–Yamagami fusion categories, arXiv:2610.35574, 2026.arxiv.org/abs/2610.35574
  2. [2]P. Etingof, S. Gelaki, D. Nikshych and V. Ostrik, Tensor categories, Mathematical Surveys and Monographs, vol. 205, American Mathematical Society, Providence, RI, 2015.
  3. [3]D. Evans and T. Gannon, Near-group fusion categories and their doubles, Adv. Math. 255 (2014), 586–640.arxiv.org/abs/1208.1500
  4. [4]M. Izumi, A Cuntz algebra approach to the classification of near-group categories, in Proceedings of the 2014 Maui and 2015 Qinhuangdao Conferences in Honour of Vaughan F. R. Jones’ 60th Birthday, Proc. Centre Math. Appl. Austral. Nat. Univ., vol. 46, Australian National University, Canberra, 2017, pp. 222–343.
  5. [5]H. Larson, Pseudo-unitary non-self-dual fusion categories of rank 4, J. Algebra 415 (2014), 184–213.arxiv.org/abs/1401.1879
  6. [6]S.-H. Ng and P. Schauenburg, Frobenius–Schur indicators and exponents of spherical categories, Adv. Math. 211 (2007), no. 1, 34–71.DOI
  7. [7]V. Ostrik, Fusion categories of rank 2, Math. Res. Lett. 10 (2003), no. 2-3, 177–183.DOI
  8. [8]V. Ostrik, Pivotal fusion categories of rank 3, with an appendix jointly with D. Nikshych, Mosc. Math. J. 15 (2015), no. 2, 373–396.
  9. [9]N. Robles and A. Roy, Moments of averages of generalized Ramanujan sums, Monatsh. Math. 182 (2017), 433–461.DOI
  10. [10]A. Schopieray, Categorification of integral group rings extended by one dimension, J. Lond. Math. Soc. (2) 108 (2023), no. 4, 1617–1641.arxiv.org/abs/2208.07319
  11. [11]A. Schopieray, The dimension block of a spherical fusion category, preprint (2026), arXiv:2609.35287v1.arxiv.org/abs/2609.35287
  12. [12]D. Tambara and S. Yamagami, Tensor categories with fusion rules of self-duality for finite abelian groups, J. Algebra 209 (1998), no. 2, 692–707.DOI

Paper details

vlnc:W_buv7jppe
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC0-1.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

AI

View statement
Looking for related papers…
Contents