Derivations of Generalized Quaternion Algebra
Derivations of Generalized Quaternion Algebra
Derivations of Generalized Quaternion Algebra
Abstract: The purpose of this paper is to determine derivations of the algebra Hα,β of generalized quaternions over the
reals and hence to obtain the algebra Der(Hα,β ) of derivations of Hα,β . Once we know derivations we might decompose
Der(Hα,β ) in terms of its inner and/or central derivations whenever they exist. Apart from Der(Hα,β ) we would also
be able to obtain generalized derivations, which have been studied by analysts in the context of algebras of some normed
spaces, and of prime and semiprime rings.
1. Introduction
Derivations of an algebra give interesting insights for studying its algebraic structure. In particular, derivations
of Lie algebras have been used in a control theoretical setting since they are intimately related with linear vector
fields. By a linear vector field on a Lie group G we mean that its flow forms a 1-parameter subgroup of the group
of G -automorphisms. Such a vector field (together with right invariant vector fields, called control vectors) is
considered to define the dynamics of linear control systems on Lie groups. Moreover, it is well known that when
G is a simply connected and nilpotent Lie group with the Lie algebra g, any derivation of g induces a linear
vector field on G . We refer the reader to [1, 2, 4, 5] for further information regarding the use of derivations in
control theory contexts and also [3], in which we provided a simple computational algorithm for finding explicit
derivations of Lie algebras.
In this paper, we consider exclusively the derivations of generalized quaternion algebra over R as a class
of Lie algebra. We first state the conditions that a linear map should obey to become a derivation for the
generalized quaternion algebra under consideration and then we obtain a typical derivation in its matrix form.
Once we have on hand the explicit derivations, we also determine the algebra of derivations in terms of inner
and/or central derivations whenever they exist. We recall that this was already done for Lie algebras in the
literature and we find it convenient to present in this article an appropriate variant for quaternion algebra. We
end our work with a brief section on generalized derivations since they are natural generalizations of classical
derivations and there has been ongoing interest about this subject. However, we constrain ourselves only to the
task of determining generalized derivations of quaternion algebra in the matrix form.
This paper is organized as follows: Section 2 only gives minimal information on quaternion algebra.
Section 3 considers their derivations, which is the main subject of our work. We then find a typical derivation of
∗Correspondence:
kizil@yildiz.edu.tr
2010 AMS Mathematics Subject Classification: 11R52, 15A99
2649
This work is licensed under a Creative Commons Attribution 4.0 International License.
KIZIL and ALAGÖZ/Turk J Math
quaternion algebra in its matrix form. In Section 4, we attempt to extend the derivations obtained to generalized
derivations since such derivations are often of interest. We explicitly get a generalized derivation by means of
its matrix representation.
2. Quaternion algebra
By an algebra A over a field F we mean a vector space over F provided with a bilinear map · : A × A → A .
We will be only considering the case F = R and A being quaternion algebra over R. We start by giving the
definition of generalized quaternions from [7] as follows:
Definition 2.1 A generalized quaternion x is of the form x = x0 e0 +x1 e1 +x2 e2 +x3 e3 where x0 , x1 , x2 , x3 ∈ R
and the quaternionic units e0 , e1 , e2 , and e3 obey the following equations:
for some α, β ∈ R .
βe1 αe2 e3
u ∧ v = u1 u2 u3 .
v1 v2 v3
Note that e1 ∧ e2 = e3 , e2 ∧ e3 = βe1 , and e3 ∧ e1 = αe2 . The multiplicative product “ · ” for Hα,β is defined
by
x · y = (x0 y0 − g(x, y))e0 + x0 y+y0 x + x ∧ y.
Addition (and hence subtraction) is defined component-wise: x+y = (x0 +y0 )e0 +x+y . In particular, any scalar
c ∈ R can be thought of as a quaternion c = c+0 and hence multiplication of x = x0 e0 +x1 e1 +x2 e2 +x3 e3 ∈ Hα,β
by c is given by
c · x = cx0 e0 + cx.
It follows that the multiplication rule is anticommutative but associative and distributive over addition.
2650
KIZIL and ALAGÖZ/Turk J Math
for all x, y ∈ A.
It is clear that the set of all derivations of an algebra A forms a vector space, which we denote by Der(A) .
Recall that gl(A) is a Lie algebra with Lie bracket given by [f, g] = f ◦ g − g ◦ f for all f, g ∈ gl(A) . Note that
D1 D2 may fail to be a derivation of A, in general. However, the commutator [D1 , D2 ] is always a derivation
since Der(A) ⊂ gl(A) and
[D1 , D2 ](x · y) = [D1 , D2 ](x) · y + x · [D1 , D2 ](y)
for every D1 , D2 ∈ Der(A) and x, y ∈ A. We also recall that any associative algebra A can be made into a Lie
algebra, say L(A), by taking the commutator as the Lie bracket [x, y] = x · y − y · x for all x, y ∈ A. It follows
that if D ∈ Der(A) then D is also a derivation of the corresponding Lie algebra, which means
for all x, y ∈ L(A). Nonetheless, it should be noted that there may exist an associative algebra A and a
derivation of the corresponding Lie algebra L(A), which is not a derivation of A . In this paper, we deal mainly
with A-derivations, where by A we simply mean the quaternion algebra Hα,β . Once we determine the algebra
Der(Hα,β ) we will be in a position to obtain derivations of some quaternions by attributing either ±1 or 0 to
α and/or β .
A particular class of derivations are the so-called inner derivations.
Definition 3.2 (Inner derivation) Given x ∈ A , by an inner derivation associated to x we mean the map
for every y ∈ A .
Let ad(Hα,β ) denote the set of all inner derivations of Hα,β as a subset of Der(Hα,β ) . By bilinearity,
ad(Hα,β ) can be generated by the maps
where ei ∈ B(Hα,β ). That is, any inner derivation D = ad(x), x ∈ Hα,β , is a linear combination of ad(ei )s.
It should be noted that for any ei ∈ B(Hα,β ) the map ad(ei ) is the zero-map if and only if ei ∈ Z(Hα,β ).
This simply means that for any quaternion algebra we always have ad(e0 ) = 0 since e0 acts for all as a global
identity.
Also note that we do not have x · (y · z) + y · (z · x) + z · (x · y) = 0 for all x, y, z ∈ Hα,β , which seems at
first glance necessary to guarantee ad(x) ∈ Der(Hα,β ). This is mainly because one needs the Jacobi identity
for a given Lie algebra g to say ad(x) ∈ Der(g) . Thanks to the associativity of Hα,β , it is easy to see for every
y, z ∈ Hα,β and every D = ad(x), x ∈ Hα,β , that D(y · z) = D(y) · z + y · D(z) . Hence, every inner derivation
is indeed a derivation in the sense of Definition 3.1.
2651
KIZIL and ALAGÖZ/Turk J Math
Hα,β
The set ad(Hα,β ) can be determined from the equation ad(Hα,β ) ' Z(Hα,β ) and thus it is easy to
determine its dimension. From Der(Hα,β )/ad(Hα,β ) we obtain outer (=noninner) derivations. We also define
for further purposes central derivations as follows:
Definition 3.3 (Central derivation) A derivation D of an algebra A mapping A into its center Z(A) is called
a central derivation.
Once we explicitly have all the derivations we are able to determine (if existing) the inner and/or central
derivations. Let D denote a derivation of Hα,β . Thus, D admits a matrix representation with respect to the
basis B(Hα,β ) , which is the 4 × 4 matrix [D] = (dij )T whose entries are defined by the following equations:
X
4
D(ei−1 ) = dij ej−1 , 1 ≤ i ≤ 4.
j=1
Each column of the matrix [D] is, of course, an element of Hα,β . In order to obtain D in its matrix form it
suffices to know the Leibnitz rule in (3.1) for the products ei · ej with 1 ≤ i ≤ j ≤ 3 . First, we state for later
purposes the following simple remark.
Remark 3.4 Since e0 is a central idempotent, it follows that D(e0 ) = 0 for every D ∈ Der(Hα,β ) . Moreover,
if we are given a derivation D , then the first column of [D] consists of only zeros. In fact,
2652
KIZIL and ALAGÖZ/Turk J Math
from which we obtain −βd11 = −2βd33 and −βd13 = 2d31 and read d31 = 0, and
6 0
0 if β =
d33 = .
0 6= d33 if β = 0
Also,
which gives −αβd11 = −2αβd44 and −αβd14 = 2d41 . That is, d41 = 0 (since d14 = 0 ) and
6 0
0 if β =
d44 = .
0 6= d44 if β = 0
Thus, from D(e1 · e2 ) = D(e1 ) · e2 + e1 · D(e2 ) we obtain d41 = (−βd23 − αd32 ), d42 = (d31 − βd24 ),
d43 = (d21 − αd34 ) , and d44 = (d22 + d33 ). Therefore, we have
β
− d23 = d32 ,
α
−βd24 = d42 ,
d43 = −αd34 ,
d33 = d44 .
Theorem 3.5 The algebra Der(Hα,β ) of derivations for Hα,β is generated by the following matrices:
0 0 0 0
0 0 −α
β
a −βb
D=
0
∈ Der(Hα,β ), (3.3)
a d −αc
0 b c d
2653
KIZIL and ALAGÖZ/Turk J Math
It follows at once from the preceding theorem that the dimension of Der(Hα,β ) is at most 4 . Moreover,
there exists a unique noninner derivation and no central derivation arises since Z(Hα,β ) = R · e0 . One might
also conclude from ad(Hα,β ) ' Hα,β /Z(Hα,β ) that the algebra ad(Hα,β ) of inner derivations is generated by
the following matrices:
0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 2β
ad(e1 ) = , ad(e2 ) = ,
0 0 0 −2α 0 0 0 0
0 0 2 0 0 −2 0 0
0 0 0 0
0 0 −2β 0
ad(e3 ) = .
0 2α 0 0
0 0 0 0
O O
If a = b = c = β = 0, then D = , where O and I2 are 2×2 zero and identity matrices, respectively,
O I2
is the only noninner derivation of Hα,β , i.e. D ∈ Der(Hα,β )/ad(Hα,β ) . It is clear that D does not commute
with any inner derivation. If we put α = β = 1 then Hα,β is the classical real quaternion algebra, say HR , for
which the automorphism group Aut(HR ) of HR consists entirely of inner automorphisms
ix : y → x · y · x−1
for invertible x ∈ HR . It is well known that Aut(HR ) is isomorphic to the group of rotations
where GL(3, R) denotes the general linear group of 3 × 3 invertible matrices with real entries. Hence, HR has
only inner derivations (as antisymmetric matrices), which might also be seen from (3.3) if we pick α = β = 1 :
0 0 0 0
0 0 −a −b
D=
0
.
a 0 −c
0 b c 0
Der(HR ) = ad(HR )
Corollary 3.6 (Semiquaternions) Let α = 1 and β = 0 , for which Hα,β is the algebra of semiquaternions
Hs . Then dim Der(Hs ) = 4 and any D ∈ Der(Hs ) is of the form
0 0 0 0
0 0 0 0
D=
0
.
a d −c
0 b c d
2654
KIZIL and ALAGÖZ/Turk J Math
Proof Since D does not commute with any inner derivation ad(ei ) , i = 1, 2, 3 , there is no central derivation.
This might be also seen from the fact that D4 (Hs ) ⊂ Re2 + Re3 while the center Z(Hs ) = Re0 . 2
Corollary 3.7 (Split quaternions) If α = −β = 1 then Hα,β is the algebra of split-quaternions Hsp and any
D ∈ Der(Hsp ) is of the form
0 0 0 0
0 0 a b
D=
0
.
a 0 −c
0 b c 0
Remark 3.8 It is seen that the algebras HR and Hsp of real and split quaternions, respectively, consist
of entirely inner derivations. Recall that HR possesses derivations that are purely antisymmetric matrices
and the exponential of an antisymmetric matrix gives an orthogonal matrix. It is well known that rotations in
quaternions are represented by conjugations (i.e. inner automorphisms). However, we have for split quaternions
both symmetric and antisymmetric derivations. For example, if a = b = 0 we have only one antisymmetric
derivation (as a generator), while for c = 0 we get two symmetric derivations. Thus, for D1 with a = 1
and b = c = 0 or D2 with b = 1 and a = c = 0 the exponential exp(Di ) is a symmetric matrix since
exp(AT ) = (exp A)T for any square matrix A.
Corollary 3.9 (Split semiquaternions) If α = −1 and β = 0 then Hα,β is the algebra of split semiquaternions
Hsps and any D ∈ Der(Hsps ) is of the form
0 0 0 0
0 0 0 0
D=
0
.
a d c
0 b c d
O O
Remark 3.10 Note that here there exists a unique (generic) noninner derivation D = and the
O I2
only derivation that makes a difference when we compare Der(Hs ) and Der(Hsps ) is the derivation D with c.
It is known that there exist Lie algebras g such that Der(g) = ad(g) and also Lie algebras such that
Der(g) = ad(g)+ C(g), where C(g) means the set of central derivations of g . In the latter case, we understand
that there exist as few derivations as possible. Hence, our theorem allows us to determine for which quaternion
algebra H one might have Der(H) = ad(H) and Der(H) = C(H) or even Der(H) = ad(H) + C(H).
4. Generalized derivations
Generalized derivation is a natural extension of classical derivation and has a wide range of applications in the
literature since it is a quite useful tool in algebraic and geometric classification of algebras. For example, a
number of papers studied generalized derivations in the context of prime and semiprime rings. In particular,
there has been interest regarding the relationship between the commutativity of a ring and the existence of
2655
KIZIL and ALAGÖZ/Turk J Math
certain derivations. For Lie algebras, a generalized derivation is defined as a linear transformation f of a Lie
algebra g such that there exists a derivation d ∈ Der(g) with the property f [x, y] = [f (x), y] + [x, d(y)] for
all x, y ∈ g . In a still more general setting, Leger and Lucks considered in [6] an endomorphism f of a Lie
algebra g as a generalized derivation if there exist g, h ∈ End(g, g) such that h[x, y] = [f (x), y] + [x, g(y)] for
all x, y ∈ g.
In this final section we deal with the notion of generalized derivations of quaternion algebra. The
motivation is that we have already obtained the algebra of derivations for generalized quaternions and hence
we can also determine generalized derivations for quaternions of various types.
Definition 4.1 (Generalized derivation) A linear map F of an algebra A (resp. a ring R ) into itself is said
to be a generalized derivation of A if there exists a nonzero derivation D ∈ Der(A) such that
for all x, y ∈ A.
Clearly, any derivation is a generalized derivation but the converse is, in general, not true. Hence, it is
clear by the above definition that a generalized derivation composes a derivation and a left multiplier, which is a
linear map f such that f (xy) = f (x)y . A simple example of generalized derivations is F (x) = ax + xb , where
a and b are fixed elements of A. We will call such maps generalized inner derivations since they generalize
inner derivations x → [a, x] = ax − xa. On the other hand, since Hα,β is a ring with the identity e0 , it follows
that by putting x = e0 in (4.1) we get
F (y) = F (e0 ) · y + D(y)
for all y ∈ Hα,β . This means every generalized derivation of Hα,β is an inner generalized derivation if and only if
Der(Hα,β ) = ad(Hα,β ), i.e. every derivation of Hα,β is inner. Hence, we can extract using inner derivations of
Hα,β all generalized inner derivations and determine in which case the quaternion algebra in question possesses
a generalized derivation that is not inner.
Generalized inner derivations have been primarily studied on operator algebras. It follows at once from
the preceding definition that knowing Der(A) is a crucial step in order to determine generalized derivations.
Since we already know Der(Hα,β ) we can find linear maps satisfying the equation in (4.1) to reveal generalized
derivations.
We would like to note that unlike the procedure for finding derivations of Hα,β there is no need to
verify (4.1) for the products ei ej between quaternionic units to determine the matrix representation of F . See
Theorem 4.2 below. Hence, generalized derivations for quaternion algebra are easier to obtain once we know
derivations of the associative algebra Hα,β .
Theorem 4.2 Let F : Hα,β → Hα,β be a generalized derivation of Hα,β and B(Hα,β ) = {e0 , e1 , e2 , e3 } denote
its standard basis. Then the matrix representation [F ] of F is as follows:
f11 −αf12 −βf13 −αβf14 0 0 0 0
f12 f11 −βf14 βf13 0 0 −α
β
a −βb
[F ] =
f13
+ ,
αf14 f11 −αf12 0 a d −αc
f14 −f13 f12 f11 0 b c d
′
where fij s and a, b, c, d are all real numbers such that d = d(β) 6= 0 if β = 0 and d = 0 otherwise.
2656
KIZIL and ALAGÖZ/Turk J Math
Proof Let [F ] = (fij )T be the 4 × 4 matrix whose entries are defined by the following equations:
It is clear that each column of the above matrix is an element of Hα,β . A simple observation yields that
for some derivation D ∈ Der(Hα,β ) . Hence, we have F (e0 ) = F (e0 ) + D(e0 ), from which we cannot deduce
explicitly the entries located in the first column of [F ] . Nonetheless, we obtain from equation (4.2) with i = 1, 2 ,
and 3 the following relations:
f21 = −αf12 , f22 = f11 , f23 = αf14 + d23 , f24 = −f13 + d24 ,
β
f31 = −βf13 , f32 = −βf14 − d23 , f33 = f11 + d33 , f34 = f12 + d34 ,
α
f41 = −αβf14 , f42 = βf13 − βd24 , f43 = −αf12 − αd34 , f44 = f11 + d33 .
2
Example 4.3 Let D denote the derivation of HR such that α = β/α = 1 , d23 = −d32 = 1 , and dij = 0
otherwise. Select f11 = 1 and fij = 0 otherwise. Then
References
[1] Ayala V, Da Silva A. On the characterization of the controllability property for linear control systems on non-
nilpotent, solvable three-dimensional Lie groups. Journal of Differential Equations 2019; 12 (5): 8233-8257. doi:
10.1016/j.jde.2018.12.027
[2] Ayala V, Da Silva A, Zsigmond G. Control sets of linear systems on Lie groups. Nonlinear Differential Equations and
Applications 2017; 24: 8. doi: 10.1007/s00030-017-0430-5
[3] Ayala V, Kizil E, Tribuzy I. On an algorithm for finding derivations of Lie algebras. Proyecciones 2012; 31 (1): 81-90.
doi: 10.4067/S0716-09172012000100008
[4] Ayala V, Tirao J. Linear control systems on Lie groups and Controllability. Proceedings of Symposia in Pure
Mathematic 1999; 66: 47-64.
[5] Jouan P. Controllability of linear systems on Lie groups. Journal of Dynamical and Control Systems 2011; 17 (4):
591-616. doi: 10.1007/s10883-011-9131-2
[6] Leger G, Luks E. Generalized derivations of Lie algebras. Journal of Algebra 2000; 228: 165-203. doi:
10.1006/jabr.1999.8250
[7] Mamagani A, Jafari M. On properties of generalized quaternion algebra. Journal of Novel Applied Sciences 2013; 2:
683-689.
[8] Togo S. Derivations of Lie algebras. Journal of Science of Hiroshima University Series A-I 1964; 28 (2): 133-158.
2657