More on the Isomorphism SU(2) SU(2) = SO(4) ∼ ⊗ Kazuyuki FUJII , Hiroshi OIKE and Tatsuo SUZUKI ∗ † ‡ Department of Mathematical Sciences ∗ Yokohama City University 7 0 0 Yokohama, 236–0027 2 n a Japan J 2 Takado 85–5, Yamagata, 990–2464 1 † 2 Japan v 6 8 Department of Mathematical Sciences 1 ‡ 8 0 Waseda University 6 0 / Tokyo, 169–8555 h p - Japan t n a u q : v Abstract i X r a In this paper we revisit the isomorphism SU(2) SU(2) = SO(4) to apply to some ⊗ ∼ subjects in Quantum Computation and Mathematical Physics. The unitary matrix Q by Makhlin giving the isomorphism as an adjoint action is studied and generalized from a different point of view. Some problems are also presented. In particular, the homogeneous manifold SU(2n)/SO(2n) which characterizes entan- glements in the case of n = 2 is studied, and a clear–cut calculation of the universal ∗E-mail address : [email protected] †E-mail address : [email protected] ‡E-mail address : [email protected] 1 Yang–Mills action in (hep-th/0602204) is given for the abelian case. 1 Introduction ThepurposeofthispaperistoreconsidertheMakhlin’stheoremandtogeneralizeitinanyqubit system, and moreover to apply to some subjects in Quantum Computation and Mathematical Physics. The isomorphism SU(2) SU(2) = SO(4) ∼ ⊗ isoneofwell–known theoremsinelementaryrepresentationtheoryandisatypicalcharacteristic of four dimensional euclidean space. However, it is usually abstract, see for example [1] 1. In [2] Makhlin gave an interesting expression to the theorem. That is, F : SU(2) SU(2) SO(4), F(A B) = Q†(A B)Q ⊗ −→ ⊗ ⊗ with some unitary matrix Q U(4). As far as we know this is the first that the map was ∈ given by the adjoint action. The construction gave and will give many applications to both Quantum Computation and Mathematical Physics, see for example [3] or [4]. In this paper we reconsider the construction (namely, Q) from a different point of view. Its construction is based on the Bell bases of 2–qubit system, so we treat a more general unitary matrix R based on them. Our method may be clear and fresh. Next we consider the problem whether or not it is possible to construct an inclusion F : SU(2) SU(2) SU(2) SO(8), F(A B C) = R†(A B C)R ⊗ ⊗ −→ ⊗ ⊗ ⊗ ⊗ with some unitary matrix R U(8). A trial is made. ∈ Since SU(2) SU(2) = SO(4) entangled states in 2–qubit system are characterized by ∼ ⊗ the homogeneous space SU(4)/SO(4) called the Lagrangean Grassmannian. In Geometry it is generalized to SU(n)/SO(n). We moreover enlarge it to U(n)/O(n), which is isomorphic to 1This is the most standard textbook in Japan on elementary representation theory 2 the product space U(1) SU(n)/SO(n). Here U(1) is a kind of phase of SU(n)/SO(n). We × give an interesting coordinate system to U(n)/O(n), which is not known as far as we know. In last we apply to so–called universal Yang–Mills action being developed by us, [5]. This is another non–linear generalization of usual Yang–Mills action [6], which is different from the Born–Infeld one [7]. To write down the action form explicitly is not so easy due to its nonlinearity. We give a clear–cut derivation to it for the abelian case. 2 The Isomorphism Revisited In this section we review the result in [2] from a different point of view. The 1–qubit space is C2 = VectC 0 , 1 where {| i | i} 1 0 0 = , 1 = . (1) | i   | i   0 1     Let σ ,σ ,σ be the Pauli matrices : 1 2 3 { } 0 1 0 i 1 0 σ = , σ = − , σ = . (2) 1 2 3       1 0 i 0 0 1 −       These ones act on the 1–qubit space. Let us prepare some notations for the latter convenience. By H(n;C) (resp. H (n;C)) the 0 set of all (resp. all traceless) hermite matrices in M(n;C) H(n;C) = A M(n;C) A† = A H (n;C) = A H(n;C) trA = 0 . 0 { ∈ | } ⊃ { ∈ | } In particular, we have H (2;C) = a a σ +a σ +a σ a ,a ,a R . 0 1 1 2 2 3 3 1 2 3 { ≡ | ∈ } Here H(n;C) H(n,R) is of course the set of all real symmetric matrices. ⊃ Next, let us consider the 2–qubit space C2 C2 = C4, which is ∼ ⊗ C2 C2 = VectC b00 , 01 , 10 , 11 ⊗ {| i | i | i | i} b 3 where ab = a b (a,b 0,1 ). | i | i⊗| i ∈ { } A comment is in order. We in the following use notations on tensor product which are different from usual ones 2. That is, C2 C2 = a b a,b C2 , ⊗ { ⊗ | ∈ } while k C2 C2 = c a b a ,b C2, c C, k N = C4. j j j j j j ∼ ⊗ ⊗ | ∈ ∈ ∈ ( ) j=1 X b We consider the Bell bases Ψ , Ψ , Ψ , Ψ which are defined as 1 2 3 4 {| i | i | i | i} 1 1 Ψ = ( 00 + 11 ), Ψ = ( 01 + 10 ), 1 2 | i √2 | i | i | i √2 | i | i 1 1 Ψ = ( 01 10 ), Ψ = ( 00 11 ). (3) 3 4 | i √2 | i−| i | i √2 | i−| i By making use of the bases the unitary matrix R is defined as R = eiθ1 Ψ ,eiθ2 Ψ ,eiθ3 Ψ ,eiθ4 Ψ U(4). 1 2 3 4 | i | i | i | i ∈ (cid:0) (cid:1) In the matrix form eiθ1 0 0 eiθ4   1 0 eiθ2 eiθ3 0 R = . (4)   √2  0 eiθ2 eiθ3 0   −     eiθ1 0 0 eiθ4     −    It is well–known the isomorphism F : SU(2) SU(2) = SO(4). ∼ ⊗ To realize it as an adjoint action by R (if it is possible) F(A B) = R†(A B)R SO(4), (5) ⊗ ⊗ ∈ we have only to determine eiθ1,eiθ2,eiθ3,eiθ4 the coefficients of R. Let us consider this problem { } in a Lie algebra level because it is in general not easy to treat it in a Lie group level. 2We believe that our notations are clearer than usual ones 4 f L(SU(2) SU(2)) L(SO(4)) ⊗ exp exp SU(2) SU(2) SO(4) ⊗ F Since the Lie algebra of SU(2) SU(2) is ⊗ L(SU(2) SU(2)) = i(a 1 +1 b) a,b H (2;C) , 2 2 0 ⊗ { ⊗ ⊗ | ∈ } we have only to examine f(i(a 1 +1 b)) = iR†(a 1 +1 b)R L(SO(4)). (6) 2 2 2 2 ⊗ ⊗ ⊗ ⊗ ∈ By setting a = 3 a σ and b = 3 b σ let us calculate the right hand side of (6). The j=1 j j j=1 j j result is P P iR†(a 1 +1 b)R = 2 2 ⊗ ⊗ 0 ie−i(θ1−θ2)(a +b ) e−i(θ1−θ3)(a b ) ie−i(θ1−θ4)(a +b ) 1 1 2 2 3 3 − −   iei(θ1−θ2)(a +b ) 0 ie−i(θ2−θ3)(a b ) e−i(θ2−θ4)(a +b ) 1 1 3 3 2 2 − − . (7)    ei(θ1−θ3)(a b ) iei(θ2−θ3)(a b ) 0 ie−i(θ3−θ4)(a b )   2 2 3 3 1 1  − − − −    iei(θ1−θ4)(a +b ) ei(θ2−θ4)(a +b ) iei(θ3−θ4)(a b ) 0   3 3 2 2 1 1   − −    Here if we set ie−i(θ1−θ2) = 1, ie−i(θ1−θ4) = 1, ie−i(θ2−θ3) = 1, ie−i(θ3−θ4) = 1, − from which e−i(θ1−θ3) = 1 and e−i(θ2−θ4) = 1 automatically, then we have − − − eiθ1 = 1, eiθ2 = i, eiθ3 = 1, eiθ4 = i. − − − Therefore our R becomes 1 0 0 i −   1 0 i 1 0 R = − − . (8)   √2  0 i 1 0   −      1 0 0 i       5 We used the notation R again for simplicity. Note that the unitary matrix R is a bit different from Q in [2]. For the latter convenience let us rewrite. If we set 0 a +b a b a +b 1 1 2 2 3 3 −   (a +b ) 0 a b (a +b ) iR†(a 1 +1 b)R = − 1 1 3 − 3 − 2 2 2 2   ⊗ ⊗  (a b ) (a b ) 0 a b   2 2 3 3 1 1  − − − − −     (a +b ) a +b (a b ) 0  3 3 2 2 1 1   − − −    0 f f f 12 13 14   f 0 f f − 12 23 24 L(SO(4)) (9)   ≡ ∈  f f 0 f   13 23 34  − −     f f f 0  14 24 34   − − −    then we obtain f +f f f f +f 12 34 13 24 14 23 a = a σ +a σ +a σ = σ + − σ + σ , (10) 1 1 2 2 3 3 1 2 3 2 2 2 f f f +f f f 12 34 13 24 14 23 b = b σ +b σ +b σ = − σ σ + − σ . (11) 1 1 2 2 3 3 1 2 3 2 − 2 2 3 A Trial toward Generalization We would like to generalize the result SU(2) SU(2) = SO(4) in the preceding section. Of ∼ ⊗ course it is not true that SU(2) SU(2) SU(2) = SO(8). Our question is as follows : is it ∼ ⊗ ⊗ possible to find an inclusion F : SU(2) SU(2) SU(2) SO(8) ⊗ ⊗ −→ with the form F(A B C) = R†(A B C)R SO(8) (12) ⊗ ⊗ ⊗ ⊗ ∈ by finding a unitary matrix R U(8) ? ∈ Let us make a trial in the following. In the preceding section we used the Bell bases to construct the unitary matrix R, so in this case we trace the same line 3. 3We believe the way of thinking natural 6 In 3–qudit system the generalized “Bell bases” are known to be Ψ , Ψ , Ψ , Ψ , Ψ , Ψ , Ψ , Ψ where 1 2 3 4 5 6 7 8 {| i | i | i | i | i | i | i | i} 1 1 Ψ = ( 000 + 111 ), Ψ = ( 001 + 110 ), 1 2 | i √2 | i | i | i √2 | i | i 1 1 Ψ = ( 010 + 101 ), Ψ = ( 011 + 100 ), 3 4 | i √2 | i | i | i √2 | i | i 1 1 Ψ = ( 011 100 ), Ψ = ( 010 101 ), 5 6 | i √2 | i−| i | i √2 | i−| i 1 1 Ψ = ( 001 110 ), Ψ = ( 000 111 ), (13) 7 8 | i √2 | i−| i | i √2 | i−| i see for example [8]. Then the unitary matrix R corresponding to (4) is given by R = eiθ1 Ψ ,eiθ2 Ψ ,eiθ3 Ψ ,eiθ4 Ψ ,eiθ5 Ψ ,eiθ6 Ψ ,eiθ7 Ψ ,eiθ8 Ψ U(8), 1 2 3 4 5 6 7 8 | i | i | i | i | i | i | i | i ∈ (cid:0) (cid:1) or in the matrix form eiθ1 0 0 0 0 0 0 eiθ8   0 eiθ2 0 0 0 0 eiθ7 0    0 0 eiθ3 0 0 eiθ6 0 0      1  0 0 0 eiθ4 eiθ5 0 0 0  R =  . (14) √2  0 0 0 eiθ4 eiθ5 0 0 0   −     0 0 eiθ3 0 0 eiθ6 0 0   −     0 eiθ2 0 0 0 0 eiθ7 0   −     eiθ1 0 0 0 0 0 0 eiθ8     −    We must check whether or not it is possible to construct F(A B C) = R†(A B C)R SO(8) (15) ⊗ ⊗ ⊗ ⊗ ∈ by determining the coefficients eiθ1,eiθ2,eiθ3,eiθ4,eiθ5,eiθ6,eiθ7,eiθ8 . { } Similarly in the preceding section we have only to check it in a Lie algebra level : f(i(a 1 1 +1 b 1 +1 1 c)) = iR†(a 1 1 +1 b 1 +1 1 c)R L(SO(8)). 2 2 2 2 2 2 2 2 2 2 2 2 ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ⊗ ∈ The result is negative. That is, the coefficients eiθ1,eiθ2,eiθ3,eiθ4,eiθ5,eiθ6,eiθ7,eiθ8 satisfy- { } ing the above equation don’t exist (a long calculation like (7) is omitted). Therefore we again propose 7 Problem Does a unitary matrix R SU(8) exist giving an inclusion ∈ F : SU(2) SU(2) SU(2) SO(8), F(A B C) = R†(A B C)R ? ⊗ ⊗ −→ ⊗ ⊗ ⊗ ⊗ A comment is in order. If we can find such an inclusion then we have the following fiber bundle SO(8) SU(8) SU(8) . F(SU(2) SU(2) SU(2)) −→ F(SU(2) SU(2) SU(2)) −→ SO(8) ⊗ ⊗ ⊗ ⊗ That is, entangled states for 3–qubit system are characterized by the homogeneous space SU(8)/F(SU(2) SU(2) SU(2)) and this space is understood by the fiber bundle. ⊗ ⊗ 4 General R Matrix In this section we treat the general n–qubit system. We would like to generalize the unitary matrix R in the preceding sections. Generalized “Bell bases” are constructed as follows. For 0 k 2n−1 1, since k can be written as ≤ ≤ − k = a 2n−2 +a 2n−3 + +a 2+a , a 0,1 0 1 n−3 n−2 j ··· ∈ { } we set 1 Ψ = 0a a a + 1a˘ a˘ a˘ , k+1 0 1 n−2 0 1 n−2 | i √2 {| ··· i | ··· i} 1 Ψ2n−k = 0a0a1 an−2 1a˘0a˘1 a˘n−2 (16) | i √2 {| ··· i−| ··· i} where a˘ = 1 a . j j − We define the unitary matrix R as R = eiθ1 Ψ1 ,eiθ2 Ψ2 , ,eiθ2n−1 Ψ2n−1 ,eiθ2n−1+1 Ψ2n−1+1 , ,eiθ2n−1 Ψ2n−1 ,eiθ2n Ψ2n . | i | i ··· | i | i ··· | i | i (cid:0) (cid:1) 8 In the matrix form eiθ1 eiθ2n   eiθ2 eiθ2n−1  ... ...      1  eiθ2n−1 eiθ2n−1+1  R =  . (17) √2  eiθ2n−1 eiθ2n−1+1   −   ... ...       eiθ2 eiθ2n−1   −     eiθ1 eiθ2n     −    For n = 2 and 3 the matrix R in the preceding sections is recovered. Now we give a characterization to R. That is, R satisfies the equation R(σ 1 1 )R† = σ σ σ . (18) 3 2 2 1 1 1 ⊗ ⊗···⊗ ⊗ ⊗···⊗ The proof is left to readers (check this for (4) and (14)). If we define W = W 1 1 2 2 ⊗ ⊗···⊗ where W is the usual Walsh–Hadamfard matrix (Wσ W = σ ), then the product RW gives 3 1 RW(σ 1 1 )(RW)† = σ σ σ . f (19) 1 2 2 1 1 1 ⊗ ⊗···⊗ ⊗ ⊗···⊗ f f That is, RW is a copy operation for σ : 1 f σ σ 1 1 σ 1 σ 1 We believe that R will play an important role in Quantum Computation. 9 5 Application to Elementary Represetation Theory To construct a representation from SU(2) to SO(3) is very well–known. In this section we point out a relation between the representation ρ and the isomorphism F in the section 2. First of all let us make a review of constructing the representation, ρ : SU(2) SO(3), (20) −→ see for example [9], Appendix F. For the matrix a+ib c+id g = a,b,c,d R   ∈ c+id a ib − −   it is easy to see g SU(2) a2 +b2 +c2 +d2 = 1. ∈ ⇐⇒ For the Pauli matrices σ ,σ ,σ in (2) we set 1 2 3 { } 1 τ = σ for j = 1,2,3. j j 2 Then the representation ρ is constructed as follows : since gτ g−1 = (a2 b2 c2 +d2)τ 2(ab cd)τ +2(ac+bd)τ , 1 1 2 3 − − − − gτ g−1 = 2(ab+cd)τ +(a2 b2 +c2 d2)τ 2(ad bc)τ , 2 1 2 3 − − − − gτ g−1 = 2(ac bd)τ +2(ad+bc)τ +(a2 +b2 c2 d2)τ , 3 1 2 3 − − − − we have gτ g−1,gτ g−1,gτ g−1 = (τ ,τ ,τ )ρ(g) 1 2 3 1 2 3 (cid:0) (cid:1) where a2 b2 c2 +d2 2(ab+cd) 2(ac bd) − − − − ρ(g) =  2(ab cd) a2 b2 +c2 d2 2(ad+bc) . (21) − − − −    2(ac+bd) 2(ad bc) a2 +b2 c2 d2   − − − −    Here let us rewrite the above. Noting that a2 b2 c2 +d2 = 1 2(b2 +c2), a2 b2 +c2 d2 = 1 2(b2 +d2), − − − − − − a2 +b2 c2 d2 = 1 2(c2 +d2), − − − 10