Hindawi Publishing Corporation The Scientific World Journal Volume 2013, Article ID 292787, 9 pages http://dx.doi.org/10.1155/2013/292787
Research Article Riemannian Means on Special Euclidean Group and Unipotent Matrices Group
Xiaomin Duan,1 Huafei Sun,1 and Linyu Peng2
1 School of Mathematics, Beijing Institute of Technology, Beijing 100081, China 2 Department of Mathematics, University of Surrey, Guildford, Surrey GU2 7XH, UK
Correspondence should be addressed to Huafei Sun; [email protected]
Received 1 August 2013; Accepted 16 September 2013
Academic Editors: R. Abu-Saris and P. Bracken
Copyright Β© 2013 Xiaomin Duan et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Among the noncompact matrix Lie groups, the special Euclidean group and the unipotent matrix group play important roles in both theoretic and applied studies. The Riemannian means of a finite set of the given points on the two matrix groups are investigated, respectively. Based on the left invariant metric on the matrix Lie groups, the geodesic between any two points is gotten. And the sum of the geodesic distances is taken as the cost function, whose minimizer is the Riemannian mean. Moreover, a Riemannian gradient algorithm for computing the Riemannian mean on the special Euclidean group and an iterative formula for that on the unipotent matrix group are proposed, respectively. Finally, several numerical simulations in the 3-dimensional case are given to illustrate our results.
1. Introduction exponential mapping from the arithmetic mean of points on the Lie algebra se(3) to the Lie group SE(3) was constructed A matrix Lie group, which is also a differentiable manifold to give the Riemannian mean in order to get a mean filter. simultaneously, attracts more and more researchersβ atten- In this paper, the Riemannian means on SE(π) and those tion from both theoretic interest and its applications [1β5]. on UP(π), which are both important noncompact matrix The Riemannian mean on the matrix Lie groups is widely Lie groups [13, 14], are considered, respectively. Especially, studied for its varied applications in biomedicine, signal SE(3) is the spacial rigid body motion, and UP(3) is the processing, and robotics control [6β9]. Fiori and Tanaka 3-dimensional Heisenberg group π»(3).Basedontheleft [10] suggested a general-purpose algorithm to compute the invariant metric on the matrix Lie groups, we get the geodesic average element of a finite set of matrices belonging to distance between any two points and take their sum as a any matrix Lie group. In [11], the author investigated the cost function. And the Riemannian mean will minimize Riemannian mean on the compact Lie groups and proposed a it. Furthermore, the Riemannian mean on SE(π) is gotten globally convergent Riemannian gradient descent algorithm. using the Riemannian gradient algorithm, rather than the Different invariant notions of mean and average rotations on approximate mean. An iterative formula for computing the SO(3) (it is compact) are given in [9]. Recently, Fiori [12]dealt Riemannian mean on UP(π) is proposed according to the with computing averages over the group of real symplectic Jacobi field. Finally, we give some numerical simulations on matrices, which found applications in diverse areas such as SE(3) and those on π»(3) to illustrate our results. optics and particle physics. However, the Riemannian mean on the special Euclidean group SE(π) and the unipotent matrix group UP(π),whichare 2. Overview of Matrix Lie Groups the noncompact matrix Lie groups, has not been well studied. Fletcher et al. [6]proposedaniterativealgorithmtoobtain In this section, we briefly introduce the Riemannian frame- the approximate solution of the Riemannian mean on SE(3) work of the matrix Lie groups [15, 16], which forms the by use of the Baker-Cambell-Hausdorff formula. In [7], the foundation of our study of the Riemannian mean on them. 2 The Scientific World Journal
2.1. The Riemannian Structures of Matrix Lie Groups. Agroup where πΏ denotes the left translation, π denotes the right πΊ is called a Lie group if it has differentiable structure: the translation, and (πΏπ΄)β and (π π΄β1 )β are the tangent mappings group operators, that is, πΊ Γ πΊ β πΊ, (π₯,π¦) σ³¨βπ₯β π¦and associated with πΏπ΄ and π π΄β1 , respectively. The adjoint action β1 πΊβπΊ,π₯σ³¨βπ₯ , are differentiable, π₯, π¦.AmatrixLie βπΊ Adπ΄ : s β s is group is a Lie group with all elements matrices. The tangent β1 space of πΊ at identity is the Lie algebra g,wheretheLiebracket Adπ΄π=π΄ππ΄ . (5) is defined. The exponential map, denoted by exp, is a map from It is also easy to see the formula that the Lie algebra g to the group πΊ. Generally, the exponential =πΏ π . map is neither surjective nor injective. Nevertheless, it is Adπ΄ π΄ π΄ (6) a diffeomorphism between a neighborhood of the identity πΌ on πΊ and a neighborhood of the identity 0 on g.The Then, the left invariant metric on π is given by (local) inverse of the exponential map is the logarithmic map, β¨π, πβ© =β¨(πΏ β1 ) π, (πΏ β1 ) πβ© denoted by log. π΄ π΄ β π΄ β πΌ The most general matrix Lie group is the general linear π (7) (π, R) πΓπ =β¨π΄β1π, π΄β1πβ© := ((π΄β1π) π΄β1π) group GL consisting of the invertible matrices πΌ tr with real entries. As the inverse image of R β{0}under the π΄ σ³¨β (π΄) (π, R) continuous map det ,GL is an open subset with π, ππ΄ βπ π and tr denoting the trace of the matrix. of the set of πΓπreal matrices, denoted by ππΓπ,whichis π πΓπ Similarly, we can define the right invariant metric on as well. isomorphic to R , it has a differentiable manifold structure It has been shown that there exist the left invariant metrics on (submanifold). The group multiplication of GL(π, R) is the all matrix Lie groups. usual matrix multiplication, the inverse map takes a matrix π΄ (π, R) π΄β1 on GL to its inverse , and the identity element is the 2.2. Compact Matrix Lie Group. A Lie group is compact if identity matrix πΌ.TheLiealgebragl(π, R) of GL(π, R) turns π(π) π its differential structure is compact. The unitary group , out to be πΓπ with the Lie bracket defined by the matrix the special unitary group SU(π), the orthogonal group π(π), commutator the special orthogonal group SO(π),andthesymplecticgroup Sp(π) are the examples of the compact matrix Lie groups [17]. [π, π] =ππβππ, glβπ,πβ (π, R) . (1) Denote a compact Lie group by π1 anditsLiealgebrabys1. β¨β , β β© π All other real matrix Lie groups are subgroups of There exists an adjoint invariant metric on 1 such that GL(π, R), and their group operators are subgroup restrictions β¨Adπ΄π, Adπ΄πβ© = β¨π, πβ© (8) of the ones on GL(π, R). The Lie bracket on their Lie algebras is still the matrix commutator. with π, π β s1. Notice the fact that the left invariant metric Let π denote a matrix Lie group and s its Lie algebra. of any adjoint invariant metric is also right invariant; namely, The exponential map for π turns out to be just the matrix it is a bi-invariant metric; so all compact Lie groups have exponential; that is, given an element πβs, the exponential bi-invariant metrics. Furthermore, if the left invariant and map is the adjoint invariant metrics on π1 deduce a Riemannian β β ππ connection , then the following properties are valid: exp (π) = β . (2) π! 1 π=0 β π= [π, π] , π 2 (9) Theinversemap,thatis,thelogarithmicmap,isdefinedas 1 follows: β¨R (π, π) π, πβ© =β β¨[π, π] , [π, π]β© , 4 β π π+1 (π΄βπΌ) log (π΄) = β (β1) , (3) where R(π, π) is a curvature operator about the smooth π=1 π tangent vector field on the Riemannian manifold (π1,β). Therefore, the section curvature K is given by for π΄ in a neighborhood of the identity πΌ of π. The exponential ofamatrixplaysacrucialroleinthetheoryoftheLiegroups, β¨[π, π] , [π, π]β© which can be used to obtain the Lie algebra of a matrix Lie K (π, π) = β₯0, 4(β¨π, πβ©β¨π, πβ© β β¨π, πβ©2) (10) group, and it transfers information from the Lie algebra to the Lie group. K The matrix Lie group also has the structure of a Rieman- which means that is nonnegative on the compact Lie nian manifold. For any π΄, π΅ βπ and πβππ΄π,thetangent group. space of π at π΄,wehavethemapsthat In addition, according to the Hopf-Rinow theorem, a compact connected Lie group is geodesically complete. It πΏπ΄π΅ = π΄π΅, (πΏπ΄)βπ=π΄π, means that, for any given two points, there exists a geodesic curve connecting them and the geodesic curve can extend β1 (4) π π΄π΅=π΅π΄ ,(π π΄β1 )βπ=ππ΄, infinitely. The Scientific World Journal 3
3 2.3. The Riemannian Mean on Matrix Lie Group. Let πΎ: with π=(ππ₯,ππ¦,ππ§)βR . βπβπΉ gives the amount of [0, 1] β π be a sufficiently smooth curve on π. We define rotation with respect to the unit vector along π,whereββ βπΉ the length of πΎ(π‘) by denotes the Frobenius norm. Physically, π represents the 1 angularvelocityoftherigidbody,whereasV corresponds to Μ Μ β(πΎ):=β« ββ¨πΎ (π‘) , πΎ (π‘)β©πΎ(π‘)ππ‘ the linear velocity [19]. In [18], the author presents a closed- 0 form expression of the exponential map se(3) β SE(3) by (11) 1 π 1β π πβ π = β« β {(πΎ(π‘)β1πΎΜ (π‘)) πΎ(π‘)β1πΎΜ (π‘)}ππ‘, cos ( ) 2 sin ( ) 3 tr exp (π) =πΌ4 +π+ π + π (18) 0 π2 π3 where π denotes the transpose of the matrix. The geodesic 2 2 2 2 with πβse(3) and π =ππ₯ +ππ¦ +ππ§.Notethatitcan distance between two matrices π΄ and π΅ on π considered as be regarded as an extension of the well-known Rodrigues a differentiable manifold is the infimum of the lengths ofthe formula on SO(3).ThelogarithmicmapSE(3) β se(3) is curves connecting them; that is, yielded as π (π΄,) π΅ := inf {β(πΎ)|πΎ:[0, 1] π =π (π πΌ βπ π+π π2 βπ π3), (12) log ( ) 1 2 4 3 4 5 (19) σ³¨β π with πΎ (0) =π΄,πΎ(1) =π΅}. where According to the Euclidean analogue (mean on Euclidean 1 3 π π space), a definition of the mean of π matrices π 1,...,π π is π = ( ) ( ), 1 8csc 2 sec 2 the minimizer of the sum of the squared distances from any matrix to the given matrices π 1,...,π π on π. Now, we define π2 =πcos (2π) β sin (π) , theRiemannianmeanbasedonthegeodesicdistance(12). (20) π3 =πcos (π) +2πcos (2π) β sin (π) β sin (2π) , Definition 1. The mean of π given matrices π 1,...,π π on π in the Riemannian sense corresponding to the metric (7)is π4 =2πcos (π) +πcos (2π) β sin (π) β sin (2π) , defined as π =π (π) β (π) , π 5 cos sin 1 2 π =arg min βπ(π π,π ) . (13) (π) = 2 + 2 (π) βπ<π<π π βπ 2π tr cos ,for . π=1 (π) (π) 3.2. Algorithm for the Riemannian Mean on SE . Denote 3. The Riemannian Mean on SE π, π β SE(π) by In this section, we discuss the Riemannian mean on the π΄ π π΄ π π=( 1 1), π=( 2 2). special Euclidean group SE(π),whichisasubgroupofGL(π+ 01 01 (21) 1, R).Moreover,thespecialrigidbodymotiongroupSE(3) is taken as an illustrating example. Taking the corresponding exponential mappings on man- π ifolds SO(π) and R into consideration, the geodesic πΎπ,π π π π (π) 3.1. About SE(π). The special Euclidean group SE(π) in R is between and on the Lie group SE is given by the semidirect product of the special orthogonal group SO(π) π πΌπ΄ ,π΄ (π‘) π½π ,π (π‘) R πΎ (π‘) =( 1 2 1 2 ) with itself [18]; that is, π,π 01 (π) = (π) β Rπ. SE SO (14) π‘ (22) π΄ (π΄ππ΄ ) π +(π βπ)π‘ (π) =( 1 1 2 1 2 1 ), The matrix representation of elements of SE is 01 π΄π π (π) ={( )|π΄β (π) ,πβR }. π SE 01 SO (15) where πΌ:[0,1]β SO(π) and π½:[0,1]β R are the geodesics expressed, respectively, by An element of SE(π) physically represents a displacement, π΄ π π π‘ where corresponds to the orientation, or attitude, of the πΌπ΄,π΅ (π‘) = expπ΄ (π‘ log (π΄ π΅)) = π΄(π΄ π΅) ,π΄,π΅βSO (π) , rigid body and π encodes the translation. The Lie algebra se(π) of SE(π) canbedenotedby π½ (π‘) = (π‘ (π βπ)) π1,π2 expπ1 2 1 Ξ© V (π) ={( )|Ξ©π =βΞ©,V β Rπ}. =π +(π βπ)π‘, π ,π β Rπ. se 00 (16) 1 2 1 1 2 (23) Specially, when π=3, the skew-symmetric matrix Ξ© can π π be uniquely expressed as Then, the midpoint of and is defined by π 1/2 1 0βππ§ ππ¦ π΄ (π΄ π΄ ) (π +π) 1 1 2 2 1 1 Ξ©=(ππ§ 0βππ₯) (17) πβπ=( ). (24) βππ¦ ππ₯ 0 01 4 The Scientific World Journal
Before the geodesic distance on SE(π) is given, we first In addition, it is valuable to mention that the distance π introduce a lemma which is a known conclusion in linear βlog(π΄1π΄2)βπΉ, induced by the standard bi-invariant metric algebra [20]. on SO(π), stands for the rotation motion from the point π to π and the distance βπ2 βπ1βπΉ stands for the translation πΓπ πΓπ π Lemma 2. If πΈβR and π»βR are invertible motion on R . Therefore, considering an object undergoing matrices, then the block matrix a rigid body Euclidean motion, then, this motion can be decomposed into a rotation with respect to the center of mass πΈπΉ ( ) of the object and a translation of the center of mass of the 0π» (25) object. πΓπ is invertible, where πΉβR .Furthermore, Theorem 4. For π given points on SE(π)
β1 β1 β1 β1 πΈπΉ πΈ βπΈ πΉπ» π΄π ππ ( ) =( ). ππ =( ), (32) 0π» 0π»β1 (26) 01 π where π΄π β SO(π) and ππ β R , π = 1,2,...,π,ifthe Now, we give the geodesic distance on SE(π) as follows. Riemannian mean of π΄1,π΄2,...,π΄π and the Riemannian π ,π ,...,π Lemma 3. The geodesic distance between two points π and π mean of 1 2 π (i.e., arithmetic mean) are denoted by π΄ π π on SE(π) induced by the scale-dependent left invariant metric and , respectively, then, one has the Riemannian mean of π ,π,...,π β (π) (7) is given by 1 2 π SE by π΄ π σ΅© σ΅©2 σ΅© σ΅©2 1/2 π=( ). π (π, π) =(σ΅© (π΄ππ΄ )σ΅© + σ΅©π βπσ΅© ) . (27) (33) σ΅©log 1 2 σ΅©πΉ σ΅© 2 1σ΅©πΉ 01
Proof. As mentioned above, the geodesic distance between Proof. In the Riemannian sense, by (13), the mean π is defined two matrices π and π on SE(π) is achieved by the length of as geodesics connecting them; thus, we will compute it through π 1 2 substituting (22)into(11). π=arg min βπ(ππ,π) πβ (π) 2π From Lemma 2,weget SE π=1
β1 π 1 σ΅© σ΅©2 σ΅© σ΅©2 πΎπ,π (π‘) σ΅© π σ΅© σ΅© σ΅© = arg min β (σ΅©log (π΄ππ΄)σ΅© + σ΅©ππ βπσ΅© ) πβ (π) 2π σ΅© σ΅©πΉ πΉ SE π=1 π βπ‘ π π βπ‘ π (π΄1π΄2) π΄1 β(π΄1π΄2) π΄1 (π1 +(π2 βπ1)π‘) (34) π =( ). 1 σ΅© σ΅©2 σ΅© π σ΅© 01 = arg min βσ΅©log (π΄ππ΄)σ΅© π΄β (π) 2π σ΅© σ΅©πΉ SO π=1 (28) π 1 σ΅© σ΅©2 Then, according to the principle about the derivatives ofthe + arg min βσ΅©ππ βπσ΅© . πβRπ 2π πΉ matrix-valued functions, the following formula is valid: π=1
π‘ From [9], the geodesic distance between π΄π and π΄ on SO(π) π΄ (π΄ππ΄ ) (π΄ππ΄ )πβπ πΎΜ (π‘) =( 1 1 2 log 1 2 2 1). is given by π,π 00(29) σ΅© σ΅©2 π(π΄ ,π΄)2 = σ΅© (π΄ππ΄)σ΅© , π σ΅©log π σ΅©πΉ (35) Moreover, we have that so we have that β1 π β1 tr ((πΎ (π‘) πΎπ,πΜ (π‘)) πΎ (π‘) πΎπ,πΜ (π‘)) π π,π π,π 1 σ΅© σ΅©2 σ΅© π σ΅© 2 (30) arg min βσ΅©log (π΄ππ΄)σ΅© = arg min π(π΄π,π΄) = π΄. 2 π π π΄β (π) 2π σ΅© σ΅©πΉ π΄β (π) SO π=1 SO =βlog (π΄1π΄2)+(π2 βπ1) (π2 βπ1). (36) Therefore, the geodesic distance on SE(π) between π and π On the other hand, for ππ β R , π = 1,2,...,π,itiseasyto π is given by see that 1 1/2 π π π (π, π) = β« (β 2 (π΄ππ΄ )+(π βπ)2) ππ‘ 1 σ΅© σ΅©2 1 log 1 2 2 1 arg min βσ΅©ππ βπσ΅© = βππ = π. (37) 0 πβRπ 2π πΉ π (31) π=1 π=1 σ΅© σ΅©2 σ΅© σ΅©2 1/2 =(σ΅© (π΄ππ΄ )σ΅© + σ΅©π βπσ΅© ) . σ΅©log 1 2 σ΅©πΉ σ΅© 2 1σ΅©πΉ Therefore, the fact is shown that the Riemannian mean π of {ππ} is equivalent to the arithmetic mean. This completes the proof of Lemma 2. Consequently, we prove that equality (33)isvalid. The Scientific World Journal 5
πΏ In addition, let denote the cost function of the mini- 3 (π) mization problem (34)onSE ;thatis, Q πΏ (π) =πΏ (π΄) +πΏ (π) 2.5 rota trans π π 2 1 σ΅© π σ΅©2 1 σ΅© σ΅©2 (38) = βσ΅© (π΄ π΄)σ΅© + βσ΅©π βπσ΅© , 2π σ΅©log π σ΅©πΉ 2π σ΅© π σ΅©πΉ π=1 π=1 z 1.5 πΏ πΏ where rota and trans standfortherotationandthetransla- 1 tion components of the cost function πΏ,respectively.Wehave πΏ (π΄) π΄β (π) the gradient of rota for SO as follows [21]: 0.5
π π 0 P grad (πΏ )=βπ΄β log (π΄ π΄π). (39) 4 rota β1 π=1 2 0 0 2 1 β2 4 3 x y Consequently, the Riemannian gradient descent algorithm is applied to calculate π΄, taking the geodesic on SO(π) as Figure 1: The rigid motion π·(π‘) from π to π. the trajectory and the negative gradient (39) as the descent direction. Finally, we achieve the following algorithm for computing 3 the Riemannian mean π on SE(π). Q 2.5 Algorithm 5. Given π matrices ππ, π = 1,2,...,π,on SE(π), their Riemannian mean π is computed by the following 2 iterative method. π z 1.5 (1) Store (1/π) βπ=1 ππ to π. P β Q
(2) Set π΄=π΄1 as an initial input, and choose a desired 1 tolerance π>0. π π 0.5 (3) If ββπ=1 log(π΄ π΄π)βπΉ <π,thenstop. P π π΄=π΄ {βπ βπ (π΄ π΄ )} 0 (4) Otherwise, update exp π=1 log π , 4 andgotostep(3). 2 0 β1 0 2 1 x β2 4 3 y 3.3. Simulations on SE(3). Let us consider a rigid object π in the Euclidean space undergoing a rigid body Euclidean Figure 2: The Riemannian mean πβπ. motion SE(3). Suppose that the coordinate of the center of 3 gravity in π is ππ β R ;then,theoptimaltrajectoryfrom the configuration π to π is the curve π·(π‘) such that 4.1. About UP(π). The set of all of the uppertriangular πΓ π· (π‘) π π matrices with diagonal elements that are all one is called ( )=πΎ (π‘) ( π), 1 π,π 1 (40) unipotent matrices group, denoted by UP(π). In fact, given an invertible matrix πΆβUP(π),thereis π‘ β [0, 1] πΎ (π‘) a neighborhood π of πΆ such that every matrix in π is also where and π,π denotes the geodesic connecting πΓπ π and π on SE(3) (see Figure 1). For the configuration of two in UP(π),soUP(π) is an open subset of R .Furthermore, points π and π,asshowninFigure 2,givenbytheangular the matrix product πβ πis clearly a smooth function of π π πβ1 velocity ππ, ππ of the rigid body and the linear velocity the entries of and ,and is a smooth function of π (π) Vπ, Vπ,wechooseππ = (π/2)(0, 1, 1), Vπ = (0, 0, 0), ππ = the entries of .Thus,UP is a Lie group. On the other (π) π(π β 1)/2 π(1/4, 0, β1/2),andVπ = (4.380, β1.348, 3.690);then,we hand, it can be verified that UP is of dimension obtain their Riemannian mean according to Algorithm 5, andisnilpotent.Sincewecanusethenonzeroelements which is just the middle point πβπfrom (24). πΆππ ,π<π, directly as global coordinate functions for UP(π), π(πβ1)/2 the manifold underlying UP(π) is diffeomorphic to R . 4. The Riemannian Mean on UP(π) Therefore, UP(π) is not compact, but simply connected. The Lie algebra up(π) of UP(π) consists of uppertriangu- In this section, the Riemannian mean of π given points on lar matrices π with diagonal elements πππ = 0, π = 1,...,π. the unipotent matrix group UP(π) is considered. UP(π) is It is an indispensable tool which gives a realization of the anoncompactmatrixLiegroupaswell.Moreover,inthe Heisenberg commutation relations of quantum mechanics in special case π=3,itistheHeisenberggroupπ»(3). the 3-dimensional case [17]. 6 The Scientific World Journal
π Moreover, it is the fact that both πΆβπΌand π are all so the Riemannian mean π΄ of the π matrices {π΅ } should nilpotent matrices, for any πΆβUP(π) and πβup(π).Thus, satisfy from (2)and(3), the infinite series representations of the (π) exponential mapping in up and the logarithm mapping in π (π) β1 π UP canbegiven,respectively,by β log (π΄ π΅ )=0. (49) π=1 π (πΆβπΌ)π (πΆ) = β (β1)π+1 , log π (41) π=1 From the logarithm of the matrices on UP(π) given by (41), we can rewrite (49)as where πΆβUP(π), βπΆπΉ βπΌβ <1,and
π π π β1 π π (π) = β π π (π΄ π΅ βπΌ) exp π! (42) β β (β1)π+1 =0. (50) π=0 π π=1π=1 with πβup(π). (π) Notice that UP is connected, which means that, for any Therefore, the Riemannian mean π΄ of the π given matrices π΄, π΅ πΎ(π‘) π given pair , we can find a geodesic curve such that {π΅ } can be given explicitly by solving (50). πΎ(0) = π΄ and πΎ(1) = π΅, namely, by taking the initial velocity π=2 β1 For the case of ,from(50), it is shown that the as πΎ(0)Μ = log(π΄ π΅). Let the geodesic curve πΎ(π‘) be π Riemannian mean π΄2 of π given matrices {π΅2} in UP(2) is β1 their arithmetic mean; that is, πΎ (π‘) =π΄exp (π‘ log (π΄ π΅)) β UP (π) (43)
πΎ(0) = π΄, πΎ(1) =π΅ πΎσΈ (0) = (π΄β1π΅) β (π) 1 π with ,and log up . π΄ = βπ΅π. π΄ π΅ 2 π 2 (51) Then, the midpoint of and is given by π=1
1 β1 π΄βπ΅=π΄exp ( log (π΄ π΅)) , (44) 2 Next, for π=3, we obtain the Riemannian mean on UP(3) (π»(3)) as follows. and from (11) the geodesic distance π(π΄, π΅) canbecomputed π explicitly by Theorem 6. Given π matrices {π΅3} on the Heisenberg group σ΅© σ΅© π»(3) by π (π΄,) π΅ = σ΅© (π΄β1π΅)σ΅© . σ΅©log σ΅©πΉ (45) 1ππ ππ 4.2. Algorithm for the Riemannian Mean on UP(π). For π 12 13 1 2 π given points π΅ ,π΅ ,...,π΅ in UP(π), πΏ denotes the cost π π π΅3 =(01π), (52) function of the minimization problem (13); that is, 23 00 1 π 1 π 2 min πΏ (π΄) = min βπ(π΅ ,π΄) . (46) π΄β (π) π΄β (π) 2π UP UP π=1 where π=1,2,...,π, then, one has the Riemannian mean π΄3 Following [22, 23], it has been shown that the Jacobi field is on the Heisenberg group π»(3) such that equal to zero at the Riemannian mean. The Jacobi field for the Riemannian mean is equal to the sum of tangent vectors to all 1 1 π π β (π ,π ) geodesics (from mean to each point). Noticing the fact that 12 13 2 cov 12 23 the geodesic between two points π΄ and π΅ has already been π΄3 =(01 π ), (53) given by (43),wecanthencomputetheJacobifieldatpoint 23 π΄ π π΅π π‘=0 to points (at )suchthat 00 1 β1 π π‘ β1 π πΎπ (π‘) =π΄(π΄ π΅ ) =π΄exp (π‘ log (π΄ π΅ )) , π π σ΅¨ where πππ := (1/π) βπ=1 πππ , π, π = 1, 2, 3 (π <π),andcov(π12, ππΎ (π‘)σ΅¨ (47) π σ΅¨ =π΄ (π΄β1π΅π), π=1,...,π. π π π σ΅¨ log π23):=(1/π)βπ=1(π12 βπ12)(π23 βπ23). ππ‘ σ΅¨π‘=0 π΄ Then, we suppose that the summation of all these vectors Proof. First, let us denote the Riemannian mean 3 by should be equal to zero; that is,
π σ΅¨ π 1π12 π13 ππΎπ (π‘)σ΅¨ β1 π πΏ = β σ΅¨ =π΄β (π΄ π΅ )=0, π΄3 =(01π23). (54) π΄ ππ‘ σ΅¨ log (48) π=1 σ΅¨π‘=0 π=1 00 1 The Scientific World Journal 7
π Then, note that, for the given matrices {π΅3} on π»(3),their B2 Riemannian mean π΄3 has to satisfy (50), so we get the 20 following solutions: 15 1 π π = βππ , 12 π 12 10 π=1 B1 23 π a 5 1 π π23 = βπ , (55) π 23 0 π=1
3 1 π β5 B π = β (π βππ )(π βππ ). 0 13 π 12 12 23 23 π=1 5 0 π a 10 10 12 Asamatterofconvenience,supposingthatπππ := (1/π)π=1 β 20 15 30 π π 40 a13 πππ , π, π = 1, 2, 3 (π <π),andcov(π12,π23):=(1/π)βπ=1(π12 β π π π12)(π23 βπ23),weshowthat(54)isvalid. Figure 3: The geodesic triangle on π»(3). This completes the proof of Theorem 6. More generally, while π>1, we can get the Riemannian which means that (58)isvalidandπ΄πβ1 satisfies the equation mean on UP(π) given by the following theorem. π π πβ1 π (β1) π Theorem 7. Take π>1.Forπ given matrices {π΅ } in UP(π), β β (π΄β1 π΅π βπΌ) =0. π π πβ1 πβ1 (61) one assumes that they are in the form of π=1π=1 π΅π ππ π΅π =( πβ1 πβ1) Moreover, from (41), we have that π 01 (56) β1 π log (π΄πβ1π΅πβ1) =0. (62) π π πβ1 with π΅πβ1 βUP(πβ1)and ππβ1 β R ; then, the Riemannian π π΄ π π΅ Furthermore, it is shown that π΄πβ1 is the Riemannian mean mean π of the matrices π is given by π of {π΅ }.Atlast,wewriteπ΄πβ1 as π΄πβ1,sotheproofof π΄ π πβ1 π΄ =( πβ1 πβ1), Theorem 7 is completed. π 01 (57) Asshownabove,wegivetheiterativeformulaforcom- π΄ {π΅π } π where πβ1 is the Riemannian mean of πβ1 and πβ1 is given puting the Riemannian mean for any dimension π>1.Either by the formula that (51)or(54)canbechosenastheinitialformula. β1 π πβ1 π π (β1) β1 π π = π΄ (β β (π΄ π΅ βπΌ) ) 4.3. Simulations on π»(3). In this section, we take two πβ1 πβ1 π+1 πβ1 πβ1 π=1π=0 examplestoillustratetheresultsabouttheRiemannianmean (58) π»(3) π πβ1 π on the Heisenberg group , which is the 3-dimensional (β1) β1 π β1 Γ(β β (π΄ π΅π βπΌ) π΄ ππ ). space. π+1 πβ1 πβ1 πβ1 πβ1 π=1π=0 Example 8. Consider the Riemannian mean of three points 1 2 3 Proof. For simplicity of exposition, we suppose that the π΅ , π΅ , π΅ on the Heisenberg group π»(3).Using(43), we Riemannian mean π΄π is the block matrix in the form of can get the geodesics of three points on π»(3),whichforma π΄ π geodesic triangle. In Figure 3, all of the curves are geodesics. π΄ =( πβ1 πβ1) π 01 (59) Moreover, as shown in Figure 4, the midpoint of each geodesic is easy to be obtained by (44). Thus, each centerline π πβ1 with π΄πβ1 β UP(π β 1) and ππβ1 β R . Since the Rie- connects a vertex to the midpoint of its opposing side. On π π»(3), these centerlines always meet in a single point which mannian mean π΄π of the π matrices {π΅π} should satisfy (50), we substitute the block matrix forms (59)and(57)into is coincident with the Riemannian mean computed by (54), (50). Then, we obtain the following matrix equation for the denoted by a red dot as shown in Figure 4. Riemannian mean π΄π: 1 2 3 4 Example 9. Given four points π΅ , π΅ , π΅ , π΅ on the Heisen- π π (β1)πβ1 π»(3) β β berg group , we can get a geodesic tetrahedron from π=1π=1 π (43)(seeFigure 5), where all curves are geodesics. Moreover, similar to Example 8, the Riemannian means of three vertexes β1 π π β1 π πβ1 β1 π Γ((π΄πβ1π΅πβ1 βπΌ) (π΄πβ1π΅πβ1 βπΌ) π΄πβ1 (ππβ1 βππβ1))=0, on each curved face are obtained, denoted by red circles 00 (see Figure 6). Then, we plot each centerline which connects (60) a vertex to the Riemannian mean of its opposing side. It 8 The Scientific World Journal
B2 20 B2
15
10 10 B1 23 a 5 1 4 5 B B 0 20 23 a 3 β5 B 0 15 0
13 5 10 a 0 β5 B3 10 10 a 20 6 12 30 4 2 5 15 40 a13 0 β2 a12 β4 Figure 4: The Riemannian mean of three points. Figure 6: The Riemannian mean of four points.
Conflict of Interests
B2 The authors declare that there is no conflict of interests regarding the publication of this paper.
10 Acknowledgments 4 B The authors wish to express their appreciation to the review- 5 B1 20 ers for their helpful suggestions which greatly improved the 23
a presentation of this paper. This paper is supported by the 0 15 National Natural Science Foundation of China (nos. 61179031 and 10932002).
13 β5 B3 10 a 6 4 2 5 References 0 β2 a12 β4 [1] G. Cigler and R. DrnovΛsek, βFrom local to global similarity of Figure 5: The geodesic tetrahedron on π»(3). matrix groups,β Linear Algebra and Its Applications,vol.435,no. 6, pp. 1285β1295, 2011. [2]A.L.SilvestreandT.Takahashi,βOnthemotionofarigidbody with a cavity filled with a viscous liquid,β Proceedings of the Royal Society of Edinburgh A,vol.142,no.2,pp.391β423,2012. is shown that these centerlines still meet in a single point, [3] H. A. Ardakani and T. J. Bridges, βShallow-water sloshing denoted by a red pentacle. In fact, the point is the Riemannian 1 2 3 4 in vessels undergoing prescribed rigid-body motion in three mean of π΅ , π΅ , π΅ , π΅ applying (54). dimensions,β Journal of Fluid Mechanics, vol. 667, pp. 474β519, 2011. [4]F.Li,K.Wang,J.Guo,andJ.Ma,βSuborbitsofapointstabilizer 5. Conclusion in the orthogonal group on the last subconstituent of orthogo- nal dual polar graphs,β Linear Algebra and Its Applications,vol. In this paper, we consider the Riemannian means on the 436, no. 5, pp. 1297β1311, 2012. (π) special Euclidean group SE and the unipotent matrix [5] K. Zhu, βInvariance of Fock spaces under the action of the (π) group UP ,respectively.Basedontheleftinvariantmetric Heisenberg group,β BulletindesSciencesMathematiques,vol. on the matrix Lie groups, we get the geodesic distance 135, no. 5, pp. 467β474, 2011. between any two points and take their sum as a cost function. [6] P.T.Fletcher,C.Lu,andS.Joshi,βStatisticsofshapeviaprincipal Furthermore, we get the Riemannian mean on SE(π) using geodesic analysis on Lie groups,β in Proceedings of the IEEE the Riemannian gradient algorithm. Moreover, we give an Computer Society Conference on Computer Vision and Pattern iterative formula for computing the Riemannian mean on Recognition (CVPR β03),vol.1,pp.95β101,June2003. UP(π) according to its Jacobi field. Finally, we make advan- [7] L. Merckel and T. Nishida, βChange-point detection on the tages of several numerical simulations on SE(3) and π»(3) to Lie group SE(3),β in Computer Vision, Imaging and Computer illustrate our results. Graphics.TheoryandApplications,vol.229ofCommunications The Scientific World Journal 9
in Computer and Information Science, pp. 230β245, Springer, Berlin, Germany, 2011. [8] K.O.Johnson,R.K.Robison,andJ.G.Pipe,βRigidbodymotion compensation for spiral projection imaging,β IEEE Transactions on Medical Imaging,vol.30,no.3,pp.655β665,2011. [9]M.Moakher,βMeansandaveraginginthegroupofrotations,β SIAM Journal on Matrix Analysis and Applications,vol.24,no. 1, pp. 1β16, 2002. [10] S. Fiori and T. Tanaka, βAn algorithm to compute averages on matrix Lie groups,β IEEE Transactions on Signal Processing,vol. 57, no. 12, pp. 4734β4743, 2009. [11] J. H. Manton, βA globally convergent numerical algorithm for computing the centre of mass on compact Lie groups,β in Proceedings of the 8th International Conference on Control, Automation, Robotics and Vision (ICARCV β04),vol.3,pp.2211β 2216, Kunming, China, December 2004. [12] S. Fiori, βSolving minimal-distance problems over the manifold of real-symplectic matrices,β SIAM Journal on Matrix Analysis and Applications,vol.32,no.3,pp.938β968,2011. [13] E. Marberg, βSuperclasses and supercharacters of normal pat- tern subgroups of the unipotent upper triangular matrix group,β Journal of Algebraic Combinatorics,vol.35,no.1,pp.61β92,2012. [14] N. Thiem, βBranching rules in the ring of superclass functions of unipotent upper-triangular matrices,β Journal of Algebraic Combinatorics,vol.31,no.2,pp.267β298,2010. [15] A. Kirillov, An Introduction to Lie Groups and Lie Algebras, Cambridge University Press, Cambridge, UK, 2008. [16] W. Chen and X. Li, An Introduction to Riemannian Geometry, Peking University Press, Beijing, China, 2002. [17] B. C. Hall, Lie Groups, Lie Algebras, and Representations: An Elementary Introduction, Springer, New York, NY, USA, 2003. [18] J. M. Selig, Geometric Fundamentals of Robotics, Springer, New York, NY, USA, 2nd edition, 2005. [19] M. Zefran,Λ V. Kumar, and C. B. Croke, βOn the generation of smooth three-dimensional rigid body motions,β IEEE Transac- tionsonRoboticsandAutomation,vol.14,no.4,pp.576β589, 1998. [20] X. Zhang, Matrix Analysis and Applications,TsinghuaUniver- sity Press, Beijing, China, 2004. [21] H. Karcher, βRiemannian center of mass and mollifier smooth- ing,β Communications on Pure and Applied Mathematics,vol.30, no. 5, pp. 509β541, 1977. [22] M. Arnaudon and X. Li, βBarycenters of measures transported by stochastic flows,β Annals of Probability,vol.33,no.4,pp. 1509β1543, 2005. [23] F. Barbaresco, βInteractions between symmetric cones and information geometrics: bruhat-tits and siegel spaces models for high resolution autoregressive doppler imagery,β in Emerg- ing Trends in Visual Computing,vol.5416ofLecture Notes in Computer Science,pp.124β163,Springer,Berlin,Germany,2009. Advances in Advances in Journal of Journal of Operations Research Decision Sciences Applied Mathematics Algebra Probability and Statistics Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014
The Scientific International Journal of World Journal Differential Equations Hindawi Publishing Corporation Hindawi Publishing Corporation http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014
Submit your manuscripts at http://www.hindawi.com
International Journal of Advances in Combinatorics Mathematical Physics Hindawi Publishing Corporation Hindawi Publishing Corporation http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014
Journal of Journal of Mathematical Problems Abstract and Discrete Dynamics in Complex Analysis Mathematics in Engineering Applied Analysis Nature and Society Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014
International Journal of Journal of Mathematics and Mathematical Discrete Mathematics Sciences
Journal of International Journal of Journal of Function Spaces Stochastic Analysis Optimization Hindawi Publishing Corporation Hindawi Publishing Corporation Volume 2014 Hindawi Publishing Corporation Hindawi Publishing Corporation Hindawi Publishing Corporation http://www.hindawi.com Volume 2014 http://www.hindawi.com http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014 http://www.hindawi.com Volume 2014