Perfect and multiple state transfer in oriented Cayley graphs

First observe that U(t) \in \mathrm{span}\ \mathcal{A}, for any time t. If PST occurs from a to b at time \tau then the (b,a)-entry is the only non-zero entry in column a of U(\tau). We see the U(\tau) = \lambda A_j for some A_j \in \mathcal{A} with k_j=1. Since (A_j)_{a,a}=0, j\neq 0 and A_j is a fixed-point-free permutation.

Now A is a sum of matrices of the form A_i - A_i^T. Since A_iJ=A_i^TJ, the vector of all ones {\bf 1} is an eigenvector of A with eigenvalue 0. Whence

and \lambda=1. Since A_j commutes with A, A_j is the permutation matrix of an automorphism of X.

(a) Let \mathcal{S}_a =\{u_0, u_1, \ldots, u_m\} where u_0=a. For j=1,\ldots, m, let \tau_j be the minimum PST time from a to u_j, and let \tau_0=0. Without loss of generality, we assume

Note that \tau_m is less than the minimum period of a.

For j=1,\ldots,m, PST from u_{j-1} to u_j occurs at time (\tau_j-\tau_{j-1}), hence

for some fixed-point-free permutation matrix P_j \in \mathcal{A} satisfying P_ju_{j-1}=u_j and phase \lambda_j.

Suppose there exists k<m such that \tau_1=\tau_j-\tau_{j-1}, for j=1,\ldots,k, and \tau_1\neq \tau_{k+1}-\tau_k. Then P_1=\cdots=P_k and P_{k+1} are distinct permutation matrices in \mathcal{A} and P_1 u_j \neq P_{k+1}u_j, for j=0,\ldots,m.

If \tau_{k+1}-\tau_k<\tau_1 then

So PST occurs from a to P_{k+1}a at time strictly less than \tau_1, which contradicts our choice of u_1.

If \tau_{k+1}-\tau_k > \tau_1 then

and PST occurs from a to P_1u_k\neq P_{k+1}u_k (= u_{k+1}) at time (\tau_1+\tau_k). Since \tau_1+\tau_k < \tau_{k+1} is less than the minimum period of a, P_1u_k \neq a. Further

because P_1 is a permutation matrix satisfying P_1u_{j-1}=u_j, for j=0,\ldots,k. The occurrence of PST from a to P_1u_k \not \in \{u_0,\ldots, u_{k+1}\} contradicts our choice of u_{k+1}.

We conclude that P_1=\cdots = P_m, and when restricted to \mathcal{S}_a, this permutation is the cycle \left(u_0, u_1, \ldots, u_m\right).

(b) Let P = U(\tau). We have seen previously that P \in \mathcal{A} is a permutation matrix. Let k be the multiplicative order of P. When restricted onto \mathcal{S}_a, we know from (a) that the permutation corresponding to P is an |\mathcal{S}_a|-cycle. Hence, P^{|\mathcal{S}_a|} has diagonal entries equal to 1. Since P \in \mathcal{A}, we must have that P^{|\mathcal{S}_a|} = I. In other words, the multiplicative order of P is equal to |\mathcal{S}_a|, for a\in V(X). (c) follows immediately from the latter.

Let A=\sum_{r} \theta_r E_r be the spectral decomposition of A. Since A\in \mathrm{span}\ \mathcal{A}, the eigenvalue support of a, \Phi_a, contains all eigenvalues of A. By Theorem 6.1 of [godsil2020perfect], there exists a square-free positive integer \Delta such that \Phi_a \subset {\mathbb{Z}}(i\sqrt{\Delta}).

From Theorem 2.2, U(\tau)=P is a permutation matrix of order m=\vert \mathcal{S}_a\vert. For r=0,\ldots,d, PE_r = \omega_r E_r for some m-th root of unity \omega_r. Since P is a permutation matrix, it follows from [lato2019quantum, Lemma 2.10.3] that \omega_r \in \mathbb{Q}(i\sqrt{\Delta}). Hence m \in \{2,3,4,6\}.

Suppose m\in \{3, 6\}. There exists r such that \omega_r is a primitive m-th root of unity, so \Delta=3. It follows from PE_r = \omega_rE_r and Theorem 2.10.3 of [lato2019quantum] that \sqrt{3}i \in \mathbb{Q}(\theta_r). Hence \theta_r \in \sqrt{3}i \mathbb{Z}, and \tau \in \frac{\pi}{\sqrt{3}} \mathbb{Q} by Corollary 6.1.3 of [lato2019quantum].

Similarly, when m=4, there exists r such that PE_r = i E_r which implies i\in \mathbb{Q}(\theta_r). Thus \theta_r \in i\mathbb{Z} and \tau \in \pi\mathbb{Q}.

Let m be the order of the permutation matrix P=U(\tau) as in the proof of Theorem 2.2. Since P \in \mathcal{A}, up to re-ordering of vertices, P is a block diagonal matrix in which every block is a m\times m permutation matrix of order m. Let a_j be a vertex in the j-th block and the result follows.

Let P be the permutation matrix of the automorphism of X that maps g to ga^{-1}. Then PU(\tau)=U(\tau)P, and U(\tau)a=b if and only if U(\tau)e=ba^{-1}. By Theorem 2.2, ba^{-1} \in \mathcal{S}_e \subset Z(G).

By Theorem 2.2, if PST occurs from e to z then z\in Z(G). From (?), U(\tau)e = z if and only if

which is equivalent to

Since z \in Z(G), z acts on every irreducible module as a scalar matrix and for g\in G

Condition (?) follows immediately from (Proof).

By Corollary 3.4, it suffices to show that there is no normal Cayley graph on G admits PST from e to z. We assume the contrary, that is, there is a normal Cayley graph \mathrm{Cay}(G,C) that has PST from e to z, at a time \tau. Using Condition (?) of Theorem 3.3, it follows that

for all \chi \in Y. Using (i) of the hypothesis, since z\notin \mathrm{ker}(\chi), for all \chi \in Y, it follows from the above equation that \theta_{\chi}\neq 0 for all \chi \in Y. By [godsil2020perfect, Theorem 7.1], the vertex e is periodic, and thus by [godsil2020perfect, Theorem 6.1], it follows that for all \chi,\psi \in Y, we have, \theta_{\chi}/\theta_{\psi} \in \mathbb{Q}. Thus, since \theta_{\psi}\in \mathbb{Q}(\psi), we have \theta_{\chi} \in \bigcap\limits_{\psi \in Y} \mathbb{Q}(\psi) =\mathbb{Q}. Since \theta_{\chi} is purely imaginary, it follows that \theta_{\chi}=0, which is a contradiction.

Assume the contrary. Then, by Theorem 3.1, there is a solvable group G of order 6n and C \subset G such that \mathrm{Cay}(G, C) is an oriented normal Cayley graph, with \mathcal{S}_{e} equal to a cyclic subgroup Z of order 6 in Z(G).

We shall apply the theory of p-special characters, as developed (in far greater generality) in [Gajendragadkar], for the primes p=2 and 3. For this purpose we recall some definitions. Given a character \phi of a group H, we define \det(\phi)(h) to be the determinant of T(h), where T is any matrix representation affording \phi. This is a well-defined linear character of H. Let p be a prime. An irreducible character \chi is p-special if \chi(1) is a power of p and the following further conditions hold: for every subnormal subgroup H of G and every irreducible constituent \phi of \chi_H, the order of \det(\phi) is a power of p.

Let \alpha and \beta be characters of Z of order 2 and 3 respectively, By [Gajendragadkar, Cor. 4.8], G has a 2-special irreducible character \chi such that \chi_Z is a multiple of \alpha. Likewise, G has a 3-special irreducible character \psi such that \psi_Z is a multiple of \beta. Hence by Condition (?) in Theorem 3.3 we have \theta_\chi\neq 0 and \theta_\psi\neq 0.

By [Gajendragadkar, Prop. 6.3(a)], \mathbb Q(\chi) lies in a cyclotomic field of order equal to the 2-power in the exponent of G, and \mathbb Q(\psi) lies in a cyclotomic field of order equal to the 3-power in the exponent of G. The intersection of these cyclotomic fields is \mathbb Q. Hence Corollary 3.5 applies.

From (?), we have

and

The above two equalities imply that

for all v \in \mathbb{Z}_{3}^{n}. Now, the result follows from Theorem 3.3 and Corollary 3.4.

We have

and

The above two equalities imply that

for all v \in \mathbb{Z}_{4}^{n}. Now, the result follows from Theorem 3.3.

Let \langle z\rangle be a cyclic subgroup of order 4. Let \lambda be the character of \langle z\rangle such that \lambda(z)=-1. There is a subgroup A with {\mathbb Z_4}^n=\langle z\rangle\times A. Thus, \lambda can be extended to a character of {\mathbb Z_4}^n whose values are rational. The result now follows Corollary 3.5.

By Clifford theory ([isaacs1994character, Corollary 6.19], since [M_{2}(n): \langle x\rangle]=2, given any non-linear irreducible character \chi, its restriction \chi|_{\langle x \rangle} is a sum of distinct conjugates of a linear character of \psi \in \mathrm{Irr}(\langle x \rangle). The set of conjugates of \psi is \{\psi,\ \psi^{\sigma}\} and thus \chi|_{\langle x \rangle}=\psi +\psi^{\sigma}. Since \psi \neq \psi^{\sigma}, we must have o(\psi)= 2^{n-1}. Thus, \psi(x) must be a primitive 2^{n-1}th root of unity. Therefore, (ii) follows from the fact that \chi|_{\langle x \rangle}=\psi +\psi^{\sigma}.

Given a non-central element g, let h \in M_{2}(n) be such that hgh^{-1}g^{-1} \neq e. Since \{e,\ x^{2^{n-2}}\} is the commutator, it follows that hgh^{-1}g^{-1}=x^{2^{n-2}}, and thus hgh^{-1}=x^{2^{n-2}}g. Therefore,

with the last equality following from the fact that x^{2^{n-2}} is a central element. As x^{2^{n-2}} is an element of order 2 and \psi is faithful, it follows that \psi(x^{2^{n-2}})=-1, and thus \chi(x^{2^{n-2}})=-2. Using (?), it follows that \chi(g)=-\chi(g), and thus (i) is true.

Using Corollary 3.4, it suffices to show that PST occurs from e to x^{2^{n-3}}, at time \tau =\pi/4. In other words, using Theorem 3.3, it suffices to show that

for all \phi \in \mathrm{Irr}(M_{2}(n)).

We first show that the above is true for linear characters. As we mentioned at the start of this section, we have [M_{2}(n),\ M_{2}(n)] =\langle x^{2^{n-2}}\rangle, and thus given a linear character \chi of M_{2}(n), we have \chi^{2^{n-2}}=1. By o(\chi), we denote the order of \chi, as an element of the dual group \widehat{M_{2}(n)}.

Case 1: Assume that o(\chi)=2^{r}, for some r\geq 3. Then, we have \chi(x^{2^{r-1}})=-1 and thus \chi(x^{4(k+2^{r-3})+1})=-\chi(x^{4k+1}), for all 0\leq k \leq 2^{n-3}-1. We note that the map x^{4k+1} \mapsto x^{4(k+2^{r-3})+1} is a bijection on \{x^{4k+1}\ :\ 0\leq k \leq 2^{n-3}-1 \}. Therefore, we have

with the last equality following from the fact that \chi(x^{4(k+2^{r-3})+1})=-\chi(x^{4k+1}), for all k. Thus, \sum\limits_{k=0}^{2^{n-3}-1} \chi(x^{4k+1})=0. We now have

Since x^{2^{n-2}} \in [M_{2}(n), M_{2}(n)] \leq \mathrm{ker}(\chi), we have \chi(x^{2^{n-2}})=1, we have \chi(x^{2^{n-3}})=\pm 1, and \chi(x^{2^{n-4}} \sigma)= \chi(\sigma) \chi(x^{2^{n-4}})=\pm\chi(\sigma)\sqrt{\chi(x^{2^{n-3}})}. Since \sigma is an element of order 2, it follows that \chi(\sigma)=\pm 1, and thus \chi(x^{2^{n-4}} \sigma)= \pm \sqrt{\chi(x^{2^{n-3}})}. Therefore, we have

and thus (Proof) is satisfied by all linear characters of order at least 8.

Case 2: Assume that o(\chi) =4. In this case, have \chi(x)=\pm i, and thus,

We also have \chi(x^{2^{n-3}}), \chi(x^{2^{n-4}} \sigma) \in \mathbb{Q}. Therefore, we have \theta_{\chi}= \pm 2^{n-2} i. As n\geq 5, we observe that (a) \chi(x^{2^{n-3}})=1 and (b) (\pi/4)\theta_{\chi} must be an even multiple of \pi. Therefore (Proof) is satisfied for all linear characters of order 4.

Case 3: We now consider the case o(\chi) =2. In this case, must have \chi(x)=\pm 1 and \chi(\sigma)=\pm1. Therefore, \chi(g) \in \mathbb{Q}, and thus, \theta_{\chi}=0. Since \chi(x)=\pm 1, we have \chi(x^{2^{n-3}})=1, and thus (Proof) holds for linear characters of order 2 as well.

We now move on to non-linear irreducible characters. Using Lemma 8.1, it follows that for a non-linear irreducible \chi, we have \chi(x^{2^{n-3}})/\chi(1)= \pm i and thus \theta_{\chi}= 2\chi(x^{2^{n-3}})/\chi(1)= \pm 2i, with the same sign. It follows that (Proof) holds for all non-linear irreducible characters as well.

We recall that the set \{\chi_{f\circ \pi}\ :\ \pi \in S_{n}\} is the set of conjugates of \chi_{f}. Using Clifford theory([isaacs1994character, Theorems 6.2 and 6.11]), we have

By a conjugate of f, we mean an element of the set \{f\circ \pi\ :\ \pi \in S_{n}\}. Given i\in [n] and \mu \in \mathrm{Irr}(G), let \Vert f \Vert_{\mu} denote the number of conjugates h of f, with h(1)=\mu. Elementary counting arguments yield that

Using (Proof), we have

The result follows by noting that

We recall from the definition (?) that \psi_{f,\Lambda} is a character induced from the subgroup G\wr S_{\mathcal{B}_{f}}, where S_{\mathcal{B}_{f}} is the subgroup of S_{n} preserving the partition \mathcal{B}_{f}=\{f^{-1}(\mu)\ : \mu \in \mathrm{Irr}(G)\} of [n]. We note that S_{\mathcal{B}_{f}} contains an n-cycle if and only if |f([n])|=1. Therefore, in case |f([n])|>1, no element of G \wr S_{n} is conjugate to (\mathbf{x}; \pi). Now, the Frobenius formula for induced characters implies part (1).

We now assume that f([n])=\{\phi\}. In this case, we have S_{\mathcal{B}_{f}}=S_{n}; \lambda:=\Lambda(\phi) \vdash n and \Lambda(\mu)=[0], for all \mu \in \mathrm{Irr}(G)\setminus \{\phi\}; and \psi_{f,\Lambda}= \widetilde{\phi^{n}_{\lambda}}. Using the well-known formula (see [James_1984, Lemma 4.3.9] or [macdonald1998symmetric, (8.2), Appendix B]) for values of \widetilde{\phi^{n}_{\lambda}}, we have

It is well known that \frac{|G|}{\phi(1)} \equiv 0 \pmod{|Z(G)|}. (see [isaacs1994character, 3.12]). There are exactly (n-1)! number of n-cycles in S_{n}. Using a well-known result [isaacs1994character, Theorem 3.7] on integrality of character values and the fact that irreducible characters of S_{n} take on integer values, it follows that \dfrac{\rho_{\lambda}(\pi)\times (n-1)!}{\rho_{\lambda}(1)} is an integer. Thus, the quantity k_{f,\Lambda}= \dfrac{|G|^{n-1}}{\phi^{n-1}(1) |Z(G)|^{n-1}} \times \dfrac{\rho_{\lambda}(\pi)\times (n-1)!}{{\rho_{\lambda}(1)}}. This concludes the proof of part (2).

By Theorem 3.3, it suffices to show that

for all \psi \in \mathrm{Irr}(G \wr S_{n}). Here,

Consider some \psi \in \mathrm{Irr}(G \wr S_{n}). Following the description of \mathrm{Irr}(G \wr S_{n}) given at the start of this section, given \psi \in \mathrm{Irr}(G\wr S_{n}), there exists f:[n] \to \mathrm{Irr}{G} and a partition-valued function \Lambda on \mathrm{Irr}(G) with |\Lambda(\phi)|=|f^{-1}(\phi)| (for all \phi \in \mathrm{Irr}(G)) such that \psi=\psi_{f,\Lambda} (as defined in (?)). We first compute \theta_{\psi_{f,\Lambda}}.

Case 1: Assume that |f([n])|>1. Using Lemma 9.2, it follows that

We note that every element of \tilde{C}_{1} is conjugate to an element of the form ((x,e,\ldots ,e) ; (1)) for some x\in C, and that the number of S_{n}-conjugates in \tilde{C}_{1} of the element ((x,e,\ldots ,e) ; (1)) (with x \in C) is n . It now follows from Lemma 9.1 that

and therefore, if \theta_{\phi} is the eigenvalue associated with \phi, of \mathrm{Cay}(G,C), we have

Case 2: Now, we assume that f([n])=\{\phi\}. Set \lambda:=\Lambda(\phi). Fix an n-cycle, \sigma \in S_{n}. We note that every element in \tilde{C}_{2} is conjugate to an element of the form (\mathbf{x}; \sigma) with \mathbf{x}_{\sigma} \in C. We count that for each c \in C, that X_{c}:=\{(\mathbf{x}; \sigma) \ : \mathbf{x}_{\sigma} = c \} is a set of size |G|^{n-1}. Since every element of X_{c} lies in the same conjugacy class, using Lemma 9.2, we have

for some k_{f,\Lambda}\in \mathbb{Z}. We recall that there are exactly (n-1)! number of n-cycles in S_{n}. Thus, we have

Arguing as in Case 1, using Lemma 9.1, we have

and therefore, we have

Using, (Proof), (?) and (?), we have

Since \mathrm{Cay}(G, C) admits PST from e to z at time \tau, by Theorem 3.3, we have

for all \phi \in \mathrm{Irr}(G). Applying the above in (?), yields

Since z \in Z(G) and since \dfrac{\overline{\phi(z)}}{\phi(1)} is a |z|-th root of unity, it follows that \dfrac{\overline{\phi(z)}}{\phi(1)}^{|Z(G)|}=1. Therefore, we have

The result now follows from Theorem 3.3.