Beruflich Dokumente
Kultur Dokumente
Marco Matone
Dipartimento di Fisica e Astronomia G. Galilei
Istituto Nazionale di Fisica Nucleare
Universit`a di Padova, Via Marzolo, 8-35131 Padova, Italy
Abstract
We show that a simple algorithm, that can be generalized by iteration, provides
a tool for extending the simplified BCH formula, recently derived by Van-Brunt
and M. Visser. In particular, we show that if [X, Y ] = uX + vY + cI, [Y, Z] =
wY + zZ + dI, [X, Z] = mX + nY + pZ + eI then
eX eY eZ = eaX+bY +cZ+dI .
where a, b, c and d are solutions of four equations. This result may have interesting
applications both in mathematics and physics. As a specific example we derive the
SL2 (C) relation
ln e1 L1 e0 L0 e1 L1 =
e +
[1 L1 + (e+ + e 2)L0 + 1 L1 ] ,
e
that follows straightforwardly from the formulation. We also note that a similar
formula arises in considering the exponentiation of the Virasoro algebra.
Introduction
Very recently Van-Brunt and Visser [1][2] found a remarkable relation that simplifies, in
particular cases, the BCH formula. Namely, if the commutator between X and Y has the
form
[X, Y ] = uX + vY + cI,
(1)
with u, v and c, c-numbers, then
eX eY = eX+Y +f (u,v)[X,Y ] ,
where
f (u, v) = f (v, u) =
(2)
In the following we will show that a simple algorithm leads to an extension of the VanBrunt and Visser formula (2) that includes commutators containing other elements of the
algebra than those in the commutator. In particular, we show that if
[X, Y ] = uX + vY + cI ,
[Y, Z] = wY + zZ + dI ,
[X, Z] = mX + nY + pZ + eI ,
c
(m3 m)m+n .
12
We also note that our formula can be extended to more general situations, such as
eX1 eXn .
The algorithm
There is a simple algorithm that generalizes the formula by Van-Brunt and Visser. Consider the decomposition
eX eY eZ = eX e Y e+ Y eZ ,
(3)
with
+ + = 1 .
If
[X, Y ] = uX + vY + cI ,
[Y, Z] = wY + zZ + dI ,
then, by (2),
eX eY eZ = eX eY ,
where
= g (u, v)X + h (u, v)Y + l (u, v)cI ,
X
Y = h+ (z, w)Y + g+ (z, w)Z + l+ (z, w)dI ,
with
g (u, v) = 1 + uf (u, v) =
u
,
1 eu
[X, Z] = mX + nY + pZ + eI .
Note that if
Y ] = uX
+ vY + cI ,
[X,
(4)
eX eY eZ = eX+Y +f (u,v)[X,Y ] .
and Y with their expressions in terms of X, Y and I, in the RHS of (4),
Replacing X
Y ], yields
and comparing the result with the direct computation of [X,
u = h+ (z, w)u + g+ (z, w)m ,
uh (u, v) + vh+ (z, w) = g (u, v)h+ (z, w)v + g (u, v)g (w, z)n
+ h (u, v)g+ (z, w)w ,
v = g (u, v)p + h (u, v)z ,
ul (u, v)c + vl+ (z, w)d + c = g (u, v)h+ (z, w)c + g (u, v)g+ (z, w)e
+ h (u, v)g+ (z, w)d ,
The first two equations fix u and v in terms of + . The third one fixes and the last one
fixes c. Also note that the first two equations provide an expression for v that compared
with the third equation leads to the equation for +
h (u, v)h+ (z, w)u + h (u, v)g+ (z, w)m + (g (u, v)p + h (u, v)z)h+ (z, w)
g (u, v)h+ (z, w)v g (u, v)g+ (z, w)n h (u, v)g+ (z, w)w = 0 .
(5)
The above algorithm can be extended to more general cases, e.g. considering decompositions like (3) for
Xn
eX
.
1 e
3
SL2(C)
(6)
where the parameters k s are arbitrary complex numbers. In this case, X, Y and Z in
(3) are X = 1L1 , Y = 0 L0 and Z = 1 L1 . One has to solve Eq.(5) with
u = 0
v=0
c=0
w=0
z = 0
d=0
21 1
0
m=0 n=
p=0 e=0
(7)
0 1
L1 ,
1 e0
and
ln(e1 L1 e0 L0 ) = 0 L0 +
By
0 1
L1 .
1 e0
e1 L1 e0 L0 e1 L1 = e1 L1 e L0 e+ L0 e1 L1 ,
where + + = 0 , so that = 0 . We get
L0 +
e1 L1 e0 L0 e1 L1 = e
1e
L1 + L0 +
+ 1
+
1e
L1
(8)
Fixing and + in such a way that the commutator between the two exponents in the
right hand side of (8) be of the form (1), that is
[ L0 +
1
+ 1
1
+ 1
L
,
L
+
L
]
=
u
(
L
+
L
)+
v
(
L
+
L1 ) ,
1
+
0
1
0
1
+
0
1e
1e +
1e
1 e+
gives
u = + ,
v = ,
4
and
(1 e )(1 e+ ) = 1 1 ,
that is
1 + e0 1 1
Therefore
1 L1 0 L0 1 L1
ln e
e
e
=
(1 + e0 1 1 )2 4e0
.
2
[1 L1 + (2 e+ e )L0 + 1 L1 ] .
+
e
(9)
Note that, in the case of SL2 (R), therefore for real k s, depending on the values of
L1 =
0 0
1 0
in the left hand side of (9) and comparing the result with yields, for D 6= 0,
A = e0 /2 (e0 1 1 ) ,
B = 1 e0 /2 ,
so that
C = 1 e0 /2 ,
D = e0 /2 ,
t2 1
,
(10)
D
where t := 12 tr . Since the eigenvalues of , solutions of the characteristic polynomial
2 2t + 1 = 0, are = t t2 1, we have
e =
so that
,
D
ln( / )
+
Note that when is parabolic, that is for |t| = 1, and therefore + = , we have
ln(+ / )
= 2sgn(t) .
+
Also, note that when is elliptic, that is for |t| < 1, one has + = = ei , so that
ln(+ / )
=
.
+
sin
5
(11)
Eq.(11) implies
A B
C D
=e
ln(+ / )
(A D)/2
B
C
(D A)/2
equivalently, using + = 1,
ln(t + t2 1)
= exp
( tI2 ) ,
t2 1
(12)
that can be iterated indefinitely by replacing some of the in the right hand side by its
exponential form.
The relation (12) can be also derived in an alternative way. Set = eX . Since X is
traceless, one has eX = aX + bI2 , and taking the trace gives b = t. Therefore = ec(tI2 )
for some c. Assuming distinct eigenvalues one can diagonalize both sides (with the same
matrix), to find
2 ln +
,
c=
+
that reproduces (12).
A particular case of (12) is when D = A1
ln A
=e
1
2C
AA1
2B
AA1
(A D)/2 t2 1
,
z =
C
where z denotes the PSL2 (R) Mobius transformation
z :=
Az + B
.
Cz + D
If B = 0, then
A 0
C D
=e
ln(+ / )
z+ z
z+ + z
0
1
z+ z
A 1
C D
=e
ln(+ / )
z+ z
z z
z+ + z
+
1
z+ z
Acknowledgements
It is a pleasure to thank Piealberto Marchetti, Paolo Pasti, Dima Sorokin and Roberto
Volpato for interesting discussions.
References
[1] A. Van-Brunt and M. Visser, arXiv:1501.02506.
[2] A. Van-Brunt and M. Visser, arXiv:1501.05034.