The concept of a contraction, or limit transition between two mathematical structures of the same type, was first introduced by I. Segal in 1951 [35 ] as a purely mathematical construction; a survey of further developments in the study of contractions can be found in [24 , 32 ] . Contractions gained widespread application following the work of Inönü and Wigner [15 ] , who formalized the notion of a limit transition between different physical theories by introducing contractions of Lie algebras associated with subalgebras (the so-called Inönü–Wigner contractions ). One of the best-known examples of a contraction is the relation between relativistic and classical mechanics, which are invariant under the Poincaré and Galilei groups, respectively; the corresponding limit transitions relate, in particular, the Klein–Gordon equation and the Schrödinger equation [15 ] . Another classical example [35 ] is the contraction of the three-dimensional special linear algebra s l ( 2 ) \mathfrak{sl}(2) sl ( 2 ) to the Weyl–Heisenberg algebra h 3 = g 3.1 \mathfrak{h}_3 = \mathfrak{g}_{3.1} h 3 = g 3.1 ; similar limit transitions are used to establish connections between quantum-mechanical models.
Thus, by studying contractions, one can find limit transitions between differential equations invariant under the corresponding Lie groups or Lie algebras. Examples of finding limit transitions between realizations of Lie algebras and invariant equations have already appeared in the literature [3 , 16 , 27 , 25 , 26 ] , where a number of concrete examples and certain theoretical aspects were considered. In the present article, we focus on formalizing the procedure for constructing parameterized realizations and the invariant systems associated with them.
The main goal of this work is to formalize a practical approach that allows one to relate invariant models by means of known limit transitions between their symmetry algebras, and to test this algorithm on the example of three-dimensional Lie algebras. The theoretical treatment is accompanied by examples: proper nontrivial contractions of the algebra s l ( 2 , C ) \mathfrak{sl}(2,\mathbb{C}) sl ( 2 , C ) are investigated, and for selected contractions, families of parameterized realizations and their limits are constructed. Lie algebras are considered over the complex field, and their local realizations are realized by holomorphic vector fields on complex manifolds. Equivalence of realizations is understood locally, using biholomorphic coordinate transformations. At the end of the paper, we also briefly outline the conceptual framework required to describe invariant equations via differential invariants. For one parameterized realization of the algebra s l ( 2 , C ) \mathfrak{sl}(2,\mathbb{C}) sl ( 2 , C ) , we consider an example of constructing differential invariants, invariant systems, and a limit transition between systems of two second-order ordinary differential equations with one independent and two dependent variables of the form
{ x t t = F ( t , x , y , x t , y t ) , y t t = G ( t , x , y , x t , y t ) . (1) \begin{cases}
x_{tt}=F(t,x,y,x_t,y_t),\\
y_{tt}=G(t,x,y,x_t,y_t). \tag*{(1)}
\end{cases} { x tt = F ( t , x , y , x t , y t ) , y tt = G ( t , x , y , x t , y t ) . (1)
Such equations belong to systems of Newtonian type, which are actively used to model the dynamics of classical and quantum systems in external force fields. In the real case, system (1) can be viewed as a generalized form of Newton’s second law for a system with two degrees of freedom. At the same time, the functions F F F and G G G play the role of components of an external force that may explicitly depend on time and velocities. Such forces are characteristic of problems with dissipation or of describing the motion of a charged particle under the Lorentz force in an electromagnetic field. Among important examples of such dissipative systems are the van der Pol equation (modeling self-oscillations in electrical circuits) and its generalization—the classical Liénard equation for systems with nonlinear friction (for more details on the symmetry properties of this equation, see [22 , 36 ] ). Another example of a nonlinear model is the Duffing oscillator, which is used, in particular, to describe nanoresonators [20 ] . These physical examples are formulated primarily in the real category; comparing them with the results of the present work requires an appropriate complexification.
In general, integrating nonlinear systems of Newtonian type is a difficult problem. However, for certain classes of functions F F F and G G G , it can be substantially simplified or reduced to quadratures using a group-theoretic approach. A fundamental problem in this context is the group classification of equations of this type [29 , 28 , 7 ] , i.e., identifying systems that admit nontrivial Lie algebras of point symmetries.
An important subclass of such equations consists of nonlinear systems of Ermakov type, which arise in many problems of applied physics, in particular in describing the evolution of charged-beam parameters in nonstationary electromagnetic fields [4 ] . In [14 ] , it was shown that the radial equation of motion of an ion in a Paul radio-frequency trap reduces to an Ermakov equation admitting the point symmetry group S L ( 2 ) \mathrm{SL}(2) SL ( 2 ) . The appeal of Ermakov systems stems from the existence of a special first integral—a quantity constant along solutions—traditionally called the Ermakov invariant. In this sense, the term “invariant” should not be automatically identified with a differential invariant of a group action. Such systems also admit the construction of solutions via a nonlinear superposition principle [33 , 2 ] . This integral is also related to the Lewis invariant for an oscillator with time-dependent frequency [17 ] .
The paper is organized as follows. Section 1 provides the necessary theoretical background. Section 2 is devoted to a detailed study of the chain of algebra contractions
s l ( 2 , C ) → g 3.4 − 1 → g 3.1 . \mathfrak{sl}(2,\mathbb{C})\to\mathfrak{g}_{3.4}^{-1}\to\mathfrak{g}_{3.1}. sl ( 2 , C ) → g 3.4 − 1 → g 3.1 .
In Section 3, parameterized families of local realizations are constructed by an algebraic method, and explicit transformations relating these realizations to canonical representatives of the corresponding algebras are found. In Section 4, for a single example, we compute a basis of first- and second-order differential invariants, construct invariant systems of differential equations, and investigate the limit transition. The resulting invariant system is also compared with the system obtained in [41 ] .
Lie algebras, their contractions and realizations
In this section, to make the exposition self-contained, we review the necessary theoretical background on Lie algebras, their contractions, and their realizations, which is well known and can be found in [24 , 31 , 39 , 42 ] and the references therein.
Consider an n n n -dimensional complex vector space V V V and a Lie algebra g \mathfrak{g} g on V V V , which is a pair ( V , [ ⋅ , ⋅ ] ) (V,[\cdot,\cdot]) ( V , [ ⋅ , ⋅ ]) , where [ ⋅ , ⋅ ] [\cdot,\cdot] [ ⋅ , ⋅ ] is a bilinear skew-symmetric operation satisfying the Jacobi identity. This operation is also called the Lie bracket or commutator.
Fix a basis e 1 , … , e n e_1,\ldots,e_n e 1 , … , e n of the vector space V V V ; then the Lie algebra g = ( V , [ ⋅ , ⋅ ] ) \mathfrak{g}=(V,[\cdot,\cdot]) g = ( V , [ ⋅ , ⋅ ]) is uniquely determined by the tensor of structure constants of type ( 1 , 2 ) (1,2) ( 1 , 2 ) , namely c i j k ∈ C c_{ij}^{k}\in\mathbb{C} c ij k ∈ C , i , j , k ∈ { 1 , … , n } i,j,k\in\{1,\ldots,n\} i , j , k ∈ { 1 , … , n } , which defines the commutators of the basis elements:
[ e i , e j ] = ∑ k = 1 n c i j k e k . [e_i,e_j]=\sum_{k=1}^{n}c_{ij}^{k}e_k. [ e i , e j ] = k = 1 ∑ n c ij k e k .
The group G L ( V ) \mathrm{GL}(V) GL ( V ) of nonsingular complex linear transformations acts on the vector space V V V , partitioning the variety of all complex Lie algebras of a given dimension into isomorphism classes. Under each transformation U ∈ G L ( V ) U\in\mathrm{GL}(V) U ∈ GL ( V ) with matrix ( U j i ) i , j = 1 n (U^{i}_{j})_{i,j=1}^{n} ( U j i ) i , j = 1 n , the tensor of structure constants transforms according to the rule
c ~ i ′ j ′ k ′ = ∑ i , j , k = 1 n ( U − 1 ) k k ′ U i ′ i U j ′ j c i j k , \tilde{c}_{i'j'}^{k'}=\sum_{i,j,k=1}^{n}(U^{-1})_{k}^{k'}U_{i'}^{i}U_{j'}^{j}c_{ij}^{k}, c ~ i ′ j ′ k ′ = i , j , k = 1 ∑ n ( U − 1 ) k k ′ U i ′ i U j ′ j c ij k ,
where i ′ , j ′ , k ′ ∈ { 1 , … , n } i',j',k'\in\{1,\ldots,n\} i ′ , j ′ , k ′ ∈ { 1 , … , n } .
The study of properties and applications of Lie algebras is closely tied to the classification of all inequivalent Lie algebras of a given dimension over a given field. In this paper, we do not dwell on a detailed analysis or survey of existing classifications, but we note that we adhere to the numbering consistent with the classification of three-dimensional real Lie algebras used in [31 ] . This choice of numbering is motivated by the need to apply our previous results and compare the systems and invariants obtained in the limit cases.
Below is the complete list of inequivalent complex three-dimensional Lie algebras together with their nonzero commutation relations:
g 1 \mathfrak{g}_{1} g 1 : (Abelian algebra);
g 2.1 ⊕ g 1 \mathfrak{g}_{2.1}\oplus\mathfrak{g}_{1} g 2.1 ⊕ g 1 : [ e 1 , e 2 ] = e 1 [e_1,e_2]=e_1 [ e 1 , e 2 ] = e 1 ;
g 3.1 \mathfrak{g}_{3.1} g 3.1 : [ e 2 , e 3 ] = e 1 [e_2,e_3]=e_1 [ e 2 , e 3 ] = e 1 (Heisenberg algebra and reduced classical Galilei algebra);
g 3.2 \mathfrak{g}_{3.2} g 3.2 : [ e 1 , e 3 ] = e 1 [e_1,e_3]=e_1 [ e 1 , e 3 ] = e 1 , [ e 2 , e 3 ] = e 1 + e 2 [e_2,e_3]=e_1+e_2 [ e 2 , e 3 ] = e 1 + e 2 ;
g 3.3 \mathfrak{g}_{3.3} g 3.3 : [ e 1 , e 3 ] = e 1 [e_1,e_3]=e_1 [ e 1 , e 3 ] = e 1 , [ e 2 , e 3 ] = e 2 [e_2,e_3]=e_2 [ e 2 , e 3 ] = e 2 ;
g 3.4 a \mathfrak{g}_{3.4}^{a} g 3.4 a : [ e 1 , e 3 ] = e 1 [e_1,e_3]=e_1 [ e 1 , e 3 ] = e 1 , [ e 2 , e 3 ] = a e 2 [e_2,e_3]=ae_2 [ e 2 , e 3 ] = a e 2 , a ∈ C ∖ { 0 , 1 , − 1 } a\in\mathbb{C}\setminus\{0,1,-1\} a ∈ C ∖ { 0 , 1 , − 1 } , a ∼ a − 1 a\sim a^{-1} a ∼ a − 1 ;
g 3.4 b \mathfrak{g}_{3.4}^{b} g 3.4 b : [ e 1 , e 3 ] = e 1 [e_1,e_3]=e_1 [ e 1 , e 3 ] = e 1 , [ e 2 , e 3 ] = − e 2 [e_2,e_3]=-e_2 [ e 2 , e 3 ] = − e 2 (three-dimensional Poincaré algebra);
s l ( 2 , C ) \mathfrak{sl}(2,\mathbb{C}) sl ( 2 , C ) : [ e 1 , e 2 ] = e 1 [e_1,e_2]=e_1 [ e 1 , e 2 ] = e 1 , [ e 1 , e 3 ] = 2 e 2 [e_1,e_3]=2e_2 [ e 1 , e 3 ] = 2 e 2 , [ e 2 , e 3 ] = e 3 [e_2,e_3]=e_3 [ e 2 , e 3 ] = e 3 (special linear algebra).
Over the field of real numbers, the classification differs from the complex one, since non-isomorphic real algebras may have isomorphic complexifications. In particular, the complexification of the rotation group algebra s o ( 3 ) \mathfrak{so}(3) so ( 3 ) is isomorphic to s l ( 2 , C ) \mathfrak{sl}(2,\mathbb{C}) sl ( 2 , C ) . The complexifications of the algebras in the family g 3.5 b \mathfrak{g}_{3.5}^{b} g 3.5 b , under an appropriate matching of parameters, belong to the family g 3.4 a \mathfrak{g}_{3.4}^{a} g 3.4 a .
The next natural question is the study of relations between different isomorphism classes of Lie algebras of a given dimension. One of the fundamental mechanisms describing transitions from one algebra to another is a Lie algebra contraction—a special limit of a parametric deformation of its structure constants under which the initial algebra degenerates into a new one, typically less complex and non-isomorphic to it. Studying contractions is important for several reasons: they establish a hierarchy among classes of Lie algebras, have direct applications in mathematical physics, and allow one to build a complete picture of orbit closures in the variety of algebras of a given dimension, which is an important problem of algebraic geometry in its own right.
Let U : ( 0 , 1 ] → G L ( V ) U:(0,1]\to\mathrm{GL}(V) U : ( 0 , 1 ] → GL ( V ) be a continuous matrix-valued function defining a parameterized family of nonsingular linear transformations U ( ε ) ∈ G L ( V ) U(\varepsilon)\in\mathrm{GL}(V) U ( ε ) ∈ GL ( V ) , ε ∈ ( 0 , 1 ] \varepsilon\in(0,1] ε ∈ ( 0 , 1 ] . Consider the one-parameter family of Lie algebras g ε = ( V , [ ⋅ , ⋅ ] ε ) \mathfrak{g}_{\varepsilon}=(V,[\cdot,\cdot]_{\varepsilon}) g ε = ( V , [ ⋅ , ⋅ ] ε ) with commutation relations
[ v , w ] ε : = U ( ε ) − 1 [ U ( ε ) v , U ( ε ) w ] , v , w ∈ g , ε ∈ ( 0 , 1 ] . (2) [v,w]_{\varepsilon}:=U(\varepsilon)^{-1}[U(\varepsilon)v,U(\varepsilon)w],\qquad v,w\in\mathfrak{g},\quad\varepsilon\in(0,1]. \tag*{(2)} [ v , w ] ε := U ( ε ) − 1 [ U ( ε ) v , U ( ε ) w ] , v , w ∈ g , ε ∈ ( 0 , 1 ] . (2)
Definition 1. If for all v , w ∈ V v,w\in V v , w ∈ V the limit [ v , w ] 0 : = lim ε → 0 + [ v , w ] ε [v,w]_0:=\lim_{\varepsilon\to0+}[v,w]_{\varepsilon} [ v , w ] 0 := lim ε → 0 + [ v , w ] ε exists, then the Lie bracket [ ⋅ , ⋅ ] 0 [\cdot,\cdot]_0 [ ⋅ , ⋅ ] 0 is well defined, and the corresponding Lie algebra g 0 = ( V , [ ⋅ , ⋅ ] 0 ) \mathfrak{g}_0=(V,[\cdot,\cdot]_0) g 0 = ( V , [ ⋅ , ⋅ ] 0 ) is called a contraction of
the algebra g \mathfrak{g} g , while the parameter ε \varepsilon ε is called the contraction parameter . By a contraction we also mean the limit transition between the algebras g → g 0 \mathfrak{g} \to\mathfrak{g}_0 g → g 0 .
If the algebra g 0 \mathfrak{g}_0 g 0 is isomorphic to an Abelian algebra, the contraction g → g 0 \mathfrak{g} \to\mathfrak{g}_0 g → g 0 is called trivial ; if g 0 \mathfrak{g}_0 g 0 is isomorphic to the original algebra g \mathfrak{g} g , the contraction is called improper .
In a fixed basis of the algebra g \mathfrak{g} g , the parameterized transformation U ( ε ) U(\varepsilon) U ( ε ) can be represented by a matrix called the contraction matrix .
In the basis e 1 , … , e n e_1,\ldots,e_n e 1 , … , e n , it follows immediately from (2) that
c ~ i ′ j ′ k ( ε ) = ∑ i , j , k U ( ε ) i i ′ U ( ε ) j j ′ ( U ( ε ) − 1 ) k ′ k c i j k . (3) \tilde{c}^{k}_{i'j'}(\varepsilon)=\sum_{i,j,k}U(\varepsilon)^{i'}_{i}U(\varepsilon)^{j'}_{j}\left(U(\varepsilon)^{-1}\right)^{k}_{k'}c^{k}_{ij}. \tag*{(3)} c ~ i ′ j ′ k ( ε ) = i , j , k ∑ U ( ε ) i i ′ U ( ε ) j j ′ ( U ( ε ) − 1 ) k ′ k c ij k . (3)
For each ε > 0 \varepsilon>0 ε > 0 , formula (3) coincides with the transformation rule for structure constants under the change to the basis U ( ε ) e 1 , … , U ( ε ) e n U(\varepsilon)e_1,\ldots,U(\varepsilon)e_n U ( ε ) e 1 , … , U ( ε ) e n of the initial algebra. Therefore, the algebra g ε \mathfrak{g}_{\varepsilon} g ε is isomorphic to g \mathfrak{g} g . At the same time, the limit algebra g 0 \mathfrak{g}_0 g 0 need not be isomorphic to g \mathfrak{g} g : the limit transition of structure constants does not necessarily correspond to a nonsingular change of basis.
In this case, the structure constants c ^ i j k \hat{c}^{k}_{ij} c ^ ij k of the resulting algebra g 0 \mathfrak{g}_0 g 0 are determined by the limit
c ^ i j k = lim ε → 0 + c ~ i j k ( ε ) . \hat{c}^{k}_{ij}=\lim_{\varepsilon\to0+}\tilde{c}^{k}_{ij}(\varepsilon). c ^ ij k = ε → 0 + lim c ~ ij k ( ε ) .
From the viewpoint of algebraic geometry, a Lie algebra contraction is a limit point of its orbit in the Zariski topology. Moreover, for orbits of the GL ( n , C ) \operatorname{GL}(n,\mathbb{C}) GL ( n , C ) -action on the variety L n \mathcal{L}_n L n of complex n n n -dimensional Lie algebras, the Zariski closure coincides with the metric closure. The distinction between these notions lies in the fact that a contraction is a specific mechanism of limit transition, whereas the orbit closure provides the complete topological-geometric picture of all possible degenerations. Rigid algebras, whose orbits are open in the Zariski topology, form the skeleton of the variety L n \mathcal{L}_n L n , while chains of contractions define a partial order relation stratifying the space by complexity.
In general, finding a contraction matrix for representatives of fixed Lie algebras is a technically challenging problem, although D. Popovych [42 ] proposed a constructive approach to solving it. In certain cases, including contractions of the algebra s l ( 2 , C ) \mathfrak{sl}(2,\mathbb{C}) sl ( 2 , C ) , one can restrict attention to matrices diagonal in the contraction parameter—the so-called Inönü–Wigner contractions.
Definition 2. Suppose a contraction g → g 0 \mathfrak{g} \to\mathfrak{g}_0 g → g 0 is defined by a matrix U ( ε ) U(\varepsilon) U ( ε ) that can be represented in the form U ( ε ) = I 0 W ( ε ) I U(\varepsilon)=I_0W(\varepsilon)I U ( ε ) = I 0 W ( ε ) I , where I I I and I 0 I_0 I 0 are constant nonsingular matrices and W ( ε ) = diag ( ε α 1 , … , ε α n ) W(\varepsilon)=\operatorname{diag}(\varepsilon^{\alpha_1},\ldots,\varepsilon^{\alpha_n}) W ( ε ) = diag ( ε α 1 , … , ε α n ) . The numbers α i ∈ Z \alpha_i\in\mathbb{Z} α i ∈ Z , i = 1 , … , n i=1,\ldots,n i = 1 , … , n , are called the signature of the contraction, and the contraction is called a generalized Inönü–Wigner contraction , or GIW contraction.
In the case α i ∈ { 0 , 1 } \alpha_i\in\{0,1\} α i ∈ { 0 , 1 } , i = 1 , … , n i=1,\ldots,n i = 1 , … , n , the contraction is called a simple Inönü–Wigner contraction , or IW contraction. The matrices I I I and I 0 I_0 I 0 do not affect the signature; see Remark 3.
For a generalized Inönü–Wigner contraction, the weights α i \alpha_i α i define a decomposition of the vector space into weight subspaces. In a basis in which the contraction matrix is diagonal, the corresponding structure constants have the form
c ~ i j k ( ε ) = ε α i + α j − α k c i j k . \tilde{c}^{k}_{ij}(\varepsilon)=\varepsilon^{\alpha_i+\alpha_j-\alpha_k}c^{k}_{ij}. c ~ ij k ( ε ) = ε α i + α j − α k c ij k .
The limit exists if and only if α i + α j − α k ≥ 0 \alpha_i+\alpha_j-\alpha_k \ge0 α i + α j − α k ≥ 0 for all c i j k ≠ 0 c_{ij}^{k} \ne0 c ij k = 0 . In the limit bracket, precisely those terms remain for which α i + α j = α k \alpha_i+\alpha_j=\alpha_k α i + α j = α k . The weight decomposition defines a filtration of the Lie algebra, and the limit algebra is the corresponding associated graded algebra.
For a simple Inönü–Wigner contraction, we have the decomposition V = V 0 ⊕ V 1 V=V_0\oplus V_1 V = V 0 ⊕ V 1 , where the indices correspond to weights 0 and 1. Here, V 0 V_0 V 0 is a subalgebra of the limit algebra, [ V 0 , V 1 ] 0 ⊆ V 1 [V_0,V_1]_0\subseteq V_1 [ V 0 , V 1 ] 0 ⊆ V 1 , and [ V 1 , V 1 ] 0 = 0 [V_1,V_1]_0=0 [ V 1 , V 1 ] 0 = 0 .
The primary tools that allow one to rule out pairs of algebras between which no contraction is possible are necessary criteria based on the semicontinuity of algebraic invariants (dimensions of derived series, ranks of adjoint operators, dimension of the derivation algebra, etc.), which exhibit monotonic behavior under degeneration. Precisely these approaches will be explicitly implemented in Section 2 to find all possible contractions of the algebra s l ( 2 , C ) \mathfrak{sl}(2,\mathbb{C}) sl ( 2 , C ) and to construct the corresponding matrices explicitly.
We also state one more proposition needed to justify contractions of realizations constructed by the algebraic method from a subalgebra.
Proposition 4. Let a Lie algebra g 0 \mathfrak{g}_0 g 0 be a contraction of a Lie algebra g \mathfrak{g} g , and suppose g \mathfrak{g} g contains a subalgebra h \mathfrak{h} h . Then g 0 \mathfrak{g}_0 g 0 contains a subalgebra h 0 \mathfrak{h}_0 h 0 of the same dimension, dim h 0 = dim h \dim\mathfrak{h}_0=\dim\mathfrak{h} dim h 0 = dim h .
The concepts presented above describe a Lie algebra as an abstract object—through the commutation relations among its basis elements—and relate different algebras to one another. However, for applications in the theory of differential equations, this abstract picture is not sufficient. Since our ultimate goal is for a Lie algebra to act as the symmetry algebra of concrete equations, its elements must be realized as infinitesimal generators of point transformations on the space of independent and dependent variables. Such a representation expresses elements of the Lie algebra as vector fields and is called a realization of the Lie algebra. It is precisely the realization that serves as a bridge between an abstract symmetry algebra and differential equations admitting that symmetry. Once the explicit form of the vector fields is found, one can construct invariants, differential invariants, and the corresponding invariant equations, thereby solving our main problem—the construction of equations with a prescribed symmetry algebra.
Consider a simply connected m m m -dimensional complex manifold M M M , defined as a Hausdorff topological space with a countable base of topology and an atlas with holomorphic transition functions between charts.
Let ( z 1 , … , z m ) (z_1,\ldots,z_m) ( z 1 , … , z m ) , where z j = x j + i y j z_j=x_j+iy_j z j = x j + i y j , j = 1 , … , m j=1,\ldots,m j = 1 , … , m , denote local complex coordinates on M M M . Then the complexified tangent space T p M ⊗ C T_pM\otimes\mathbb{C} T p M ⊗ C at a point p ∈ M p\in M p ∈ M has two natural bases: the real basis
{ ∂ ∂ x 1 , ∂ ∂ y 1 , … , ∂ ∂ x m , ∂ ∂ y m } \left\{\frac{\partial}{\partial x_1},\frac{\partial}{\partial y_1},\ldots,\frac{\partial}{\partial x_m},\frac{\partial}{\partial y_m}\right\} { ∂ x 1 ∂ , ∂ y 1 ∂ , … , ∂ x m ∂ , ∂ y m ∂ }
and the complex Wirtinger basis
∂ ∂ z j = 1 2 ( ∂ ∂ x j − i ∂ ∂ y j ) , ∂ ∂ z ‾ j = 1 2 ( ∂ ∂ x j + i ∂ ∂ y j ) . \frac{\partial}{\partial z_j}=\frac{1}{2}\left(\frac{\partial}{\partial x_j}-i\frac{\partial}{\partial y_j}\right),\qquad\frac{\partial}{\partial\overline{z}_j}=\frac{1}{2}\left(\frac{\partial}{\partial x_j}+i\frac{\partial}{\partial y_j}\right). ∂ z j ∂ = 2 1 ( ∂ x j ∂ − i ∂ y j ∂ ) , ∂ z j ∂ = 2 1 ( ∂ x j ∂ + i ∂ y j ∂ ) .
Locally, any complex vector field Ξ \Xi Ξ can be written with smooth complex-valued coefficients ξ j \xi_j ξ j and ξ ˉ j \bar{\xi}_j ξ ˉ j in the form Ξ = ∑ j = 1 m ( ξ j ( z , z ˉ ) ∂ ∂ z j + ξ ˉ j ( z , z ˉ ) ∂ ∂ z ˉ j ) \Xi=\sum_{j=1}^{m}\left(\xi_j(z,\bar{z})\frac{\partial}{\partial z_j}+\bar{\xi}_j(z,\bar{z})\frac{\partial}{\partial\bar{z}_j}\right) Ξ = ∑ j = 1 m ( ξ j ( z , z ˉ ) ∂ z j ∂ + ξ ˉ j ( z , z ˉ ) ∂ z ˉ j ∂ ) .
Vector fields are said to be equivalent if they are related by a change to another system of holomorphic coordinates (we consider only holomorphic vector fields, i.e., fields of type ( 1 , 0 ) (1,0) ( 1 , 0 ) whose antiholomorphic part vanishes and whose coefficients in the holomorphic basis are holomorphic functions). In an arbitrary holomorphic coordinate chart, a holomorphic vector field can be written as
Ξ = ∑ j = 1 m ξ j ( z ) ∂ ∂ z j , where ξ j ( z ) = ξ j ( z 1 , … , z m ) are holomorphic functions. \Xi=\sum_{j=1}^{m}\xi_j(z)\frac{\partial}{\partial z_j},\qquad\text{where }\xi_j(z)=\xi_j(z_1,\ldots,z_m)\text{ are holomorphic functions.} Ξ = j = 1 ∑ m ξ j ( z ) ∂ z j ∂ , where ξ j ( z ) = ξ j ( z 1 , … , z m ) are holomorphic functions.
In another system of holomorphic coordinates w = ( w 1 , … , w m ) w=(w_1,\ldots,w_m) w = ( w 1 , … , w m ) , the new coefficients are obtained via the chain rule by substituting into the expression for Ξ \Xi Ξ : ξ ~ k ( w ) = ∑ j = 1 m ξ j ( z ) ∂ w k ∂ z j \tilde{\xi}_k(w)=\sum_{j=1}^{m}\xi_j(z)\frac{\partial w_k}{\partial z_j} ξ ~ k ( w ) = ∑ j = 1 m ξ j ( z ) ∂ z j ∂ w k . Let M p ⊂ M M_p \subset M M p ⊂ M be an open neighborhood of the point p p p , and let Vec h o l ( M p ) \operatorname{Vec}_{\mathrm{hol}}(M_p) Vec hol ( M p ) denote the space of holomorphic vector fields on M p M_p M p . The Lie bracket of two holomorphic vector fields is again a holomorphic vector field; hence Vec h o l ( M p ) \operatorname{Vec}_{\mathrm{hol}}(M_p) Vec hol ( M p ) is a Lie algebra over C \mathbb{C} C . Consider homomorphisms from the algebra g \mathfrak{g} g into Vec h o l ( M p ) \operatorname{Vec}_{\mathrm{hol}}(M_p) Vec hol ( M p ) , i.e., assign to a basis of g \mathfrak{g} g a collection of local vector fields Ξ i = ξ i a ( z ) ∂ z a \Xi_i = \xi_i^a(z)\partial_{z_a} Ξ i = ξ i a ( z ) ∂ z a whose commutators [ Ξ i , Ξ j ] [\Xi_i,\Xi_j] [ Ξ i , Ξ j ] reproduce the given structure constants of the algebra.
Definition 5. A realization of a complex Lie algebra g \mathfrak{g} g in a neighborhood of a point p ∈ M p \in M p ∈ M is a homomorphism R : g → Vec h o l ( M p ) R:\mathfrak{g} \to\operatorname{Vec}_{\mathrm{hol}}(M_p) R : g → Vec hol ( M p ) .
A realization is called faithful if R R R is injective (i.e., ker R = { 0 } \ker R = \{0\} ker R = { 0 } ), and degenerate if the vector fields R ( e 1 ) , … , R ( e n ) R(e_1),\ldots,R(e_n) R ( e 1 ) , … , R ( e n ) are linearly dependent over C \mathbb{C} C with constant coefficients.
Given two local realizations R 1 R_1 R 1 and R 2 R_2 R 2 in neighborhoods of points p 1 p_1 p 1 and p 2 p_2 p 2 (neighborhoods M 1 M_1 M 1 and M 2 M_2 M 2 , respectively), R 1 : g ⟶ Vec h o l ( M 1 ) R_1:\mathfrak{g}\longrightarrow\operatorname{Vec}_{\mathrm{hol}}(M_1) R 1 : g ⟶ Vec hol ( M 1 ) and R 2 : g ⟶ Vec h o l ( M 2 ) R_2:\mathfrak{g}\longrightarrow\operatorname{Vec}_{\mathrm{hol}}(M_2) R 2 : g ⟶ Vec hol ( M 2 ) , establishing their equivalence requires accounting for both geometric holomorphic transformations of the realization domains and complex-linear automorphisms of the algebra g \mathfrak{g} g . In the real case, realizations are compared via diffeomorphisms; in the complex case, one must impose the stricter requirement of a biholomorphic transformation between coordinates. Denote the automorphism group of the algebra g \mathfrak{g} g by Aut ( g ) = { ψ ∈ GL ( g ) ∣ ψ ( [ x , y ] ) = [ ψ ( x ) , ψ ( y ) ] } \operatorname{Aut}(\mathfrak{g})=\{\psi\in\operatorname{GL}(\mathfrak{g})\mid\psi([x,y])=[\psi(x),\psi(y)]\} Aut ( g ) = { ψ ∈ GL ( g ) ∣ ψ ([ x , y ]) = [ ψ ( x ) , ψ ( y )]} .
Definition 7. Realizations R 1 : g → Vec h o l ( M 1 ) R_1:\mathfrak{g}\to\operatorname{Vec}_{\mathrm{hol}}(M_1) R 1 : g → Vec hol ( M 1 ) and R 2 : g → Vec h o l ( M 2 ) R_2:\mathfrak{g}\to\operatorname{Vec}_{\mathrm{hol}}(M_2) R 2 : g → Vec hol ( M 2 ) are called weakly equivalent if there exist an automorphism ψ ∈ Aut ( g ) \psi\in\operatorname{Aut}(\mathfrak{g}) ψ ∈ Aut ( g ) and a biholomorphism Φ : M 1 → M 2 \Phi:M_1\to M_2 Φ : M 1 → M 2 such that R 2 ( ψ ( v ) ) = Φ ∗ R 1 ( v ) R_2(\psi(v))=\Phi_*R_1(v) R 2 ( ψ ( v )) = Φ ∗ R 1 ( v ) for all v ∈ g v\in\mathfrak{g} v ∈ g . If ψ \psi ψ is the identity map, the realizations are mapped into one another solely by a local biholomorphism. In this special case, the realizations are called strongly equivalent .
Other types of equivalence can also be defined.
An important difference between the real [31 ] and complex cases is the local nature of a complex realization. However, since M M M is a connected complex manifold, a holomorphic vector field that vanishes on a nonempty open subset vanishes on the entire connected component of its domain. This yields the following consequences:
if two holomorphic realizations coincide on an open subset of the intersection of their domains, they coincide on the entire corresponding connected component;
if a local holomorphic continuation of a realization exists, it is unique.
At the same time, a local realization does not always extend automatically to the entire manifold: global continuation requires taking into account the topology of the manifold, monodromy, and possible analytic obstructions.
Definition 8. The rank of a realization at a point p p p is defined as rank R p = dim C span { R ( v ) ∣ p } v ∈ g \operatorname{rank}R_p=\dim_{\mathbb{C}}\operatorname{span}\{R(v)|_p\}_{v\in\mathfrak{g}} rank R p = dim C span { R ( v ) ∣ p } v ∈ g , and rank R = max p rank R p \operatorname{rank}R=\max_p\operatorname{rank}R_p rank R = max p rank R p .
A realization R R R is called transitive if rank R p = m \operatorname{rank}R_p=m rank R p = m for every point of its domain, and regular if rank R p \operatorname{rank}R_p rank R p is locally constant; the regular rank may be smaller than m m m .
Note that the condition max p rank R p = m \max_p\operatorname{rank}R_p=m max p rank R p = m yields transitivity only on the open subset { p : rank R p = m } \{p:\operatorname{rank}R_p=m\} { p : rank R p = m } , which may be proper. For example, the field x ∂ x x\partial_x x ∂ x on C \mathbb{C} C has maximal rank 1 1 1 , while the point 0 0 0 is fixed. The rank of a realization is conveniently computed by forming the matrix of coefficients of the vector fields.
For a subset h ⊆ g \mathfrak{h} \subseteq\mathfrak{g} h ⊆ g , define
rank R ( h ) p = dim C span C { R ( v ) ∣ p } v ∈ h . \operatorname{rank} R(\mathfrak{h})_p = \dim_{\mathbb{C}} \operatorname{span}_{\mathbb{C}}\{R(v)|_p\}_{v\in\mathfrak{h}}. rank R ( h ) p = dim C span C { R ( v ) ∣ p } v ∈ h .
If h \mathfrak{h} h is invariant under all automorphisms of the algebra g \mathfrak{g} g , then its rank at corresponding points is preserved under weak equivalence. Consequently, a discrepancy between these ranks at corresponding points provides a criterion for inequivalence. For an arbitrary subset, one must also take into account its image under the action of an automorphism.
The next task toward constructing systems of Newtonian type is the explicit construction of representatives of equivalence classes. The problem of classifying realizations splits into two interrelated parts: constructing a system of vector fields realizing a given abstract Lie algebra, and isolating all inequivalent cases. This topic has been widely addressed in the literature, yet no complete algorithm for constructing all realizations currently exists. We consider two approaches to constructing realizations, the choice between which is dictated by the specifics of a given problem.
The direct method [31 ] starts from the classification of abstract Lie algebras and uses the technique of megaideals to reduce the problem. The direct construction of realizations reduces to solving systems of partial differential equations for the coefficient functions of the vector fields, followed by classifying the resulting realizations up to weak equivalence. This stage requires computing automorphism groups of the algebra, finding its ideals, and analyzing admissible nonsingular changes of variables.
The algebraic method [21 ] , proposed by Magazev, Mikheyev, and Shirokov, proceeds from a fundamentally different idea: a realization is constructed as a set of left-invariant vector fields on the Lie group itself using solely the structure constants of the algebra. In canonical coordinates of the second kind, the components of left-invariant 1 1 1 -forms ω \omega ω are computed via the matrices of the adjoint representation, after which the corresponding vector fields are found by inverting the matrix ω \omega ω . Thus, the problem reduces to purely linear-algebraic operations—computing matrix exponentials and matrix inverses. The method produces transitive realizations: simply transitive on G G G (the case h = 0 \mathfrak{h}=0 h = 0 ) and transitive on homogeneous spaces G / H G/H G / H .
Note that all subsequent constructions are carried out in the complex holomorphic category, where M M M is a complex manifold with dim C M = m \dim_{\mathbb{C}} M=m dim C M = m , and the fields of a realization are holomorphic sections of T 1 , 0 M T^{1,0}M T 1 , 0 M . The complexification of the real tangent space T p M ⊗ C T_pM\otimes\mathbb{C} T p M ⊗ C (of dimension 2 m 2m 2 m ) is not used here. All constructions are local, so a coordinate neighborhood suffices; what is essential is not locality itself, but the identity theorem for holomorphic coefficients.
By the classical theory of Sophus Lie, vector fields realizing a Lie algebra g \mathfrak{g} g can be identified with the infinitesimal generators of a local Lie group G G G of transformations of the manifold M M M . In [21 ] , it is shown that any transitive local realization of a Lie algebra g \mathfrak{g} g by vector fields on an m m m -dimensional manifold M M M is determined by a subalgebra h \mathfrak{h} h of codimension m m m , h = { v ∈ g ∣ R ( v ) ∣ p = 0 } \mathfrak{h}=\{v\in\mathfrak{g}\mid R(v)|_p=0\} h = { v ∈ g ∣ R ( v ) ∣ p = 0 } , which is the stabilizer of the point p ∈ M p\in M p ∈ M . That is, there is a one-to-one correspondence between classes of locally equivalent transitive realizations of g \mathfrak{g} g and conjugacy classes of subalgebras h \mathfrak{h} h .
From a geometric standpoint, this follows from the fact that any transitive action of a Lie group G G G is locally equivalent to the action on the space of left cosets G / H G/H G / H by the subgroup H H H corresponding to the subalgebra h \mathfrak{h} h .
Let R R R be a transitive realization on M M M and let h = { v ∈ g ∣ R ( v ) ∣ p = 0 } \mathfrak{h}=\{v\in\mathfrak{g}\mid R(v)|_p=0\} h = { v ∈ g ∣ R ( v ) ∣ p = 0 } be the stabilizer of a point p p p ; then codim h = m \operatorname{codim}\mathfrak{h}=m codim h = m . According to [21 ] , ker R = core g ( h ) \ker R=\operatorname{core}_{\mathfrak{g}}(\mathfrak{h}) ker R = core g ( h ) , where core g ( h ) \operatorname{core}_{\mathfrak{g}}(\mathfrak{h}) core g ( h ) is the largest ideal of g \mathfrak{g} g contained in h \mathfrak{h} h . In particular, the realization R R R is faithful if and only if h \mathfrak{h} h contains no nonzero ideals of g \mathfrak{g} g . The classification of transitive realizations reduces to the classification of subalgebras of codimension m m m , where h ′ = h ⟺ R ′ \mathfrak{h}'=\mathfrak{h}\Longleftrightarrow R' h ′ = h ⟺ R ′ is strongly equivalent to R R R , h ′ ∈ Ad G h ⟺ R ′ \mathfrak{h}'\in\operatorname{Ad}_{G}\mathfrak{h}\Longleftrightarrow R' h ′ ∈ Ad G h ⟺ R ′ is equivalent to R R R up to Inn ( g ) \operatorname{Inn}(\mathfrak{g}) Inn ( g ) , and h ′ ∈ Aut ( g ) h ⟺ R ′ \mathfrak{h}'\in\operatorname{Aut}(\mathfrak{g})\mathfrak{h}\Longleftrightarrow R' h ′ ∈ Aut ( g ) h ⟺ R ′ is weakly equivalent to R R R .
To construct a realization explicitly via invariant vector fields on a Lie group, canonical coordinates of the second kind ( x 1 , … , x n ) (x_1,\ldots,x_n) ( x 1 , … , x n ) are used for group elements G ∋ g = exp ( x 1 e 1 ) ⋯ exp ( x n e n ) G \ni g=\exp(x_1e_1)\cdots\exp(x_ne_n) G ∋ g = exp ( x 1 e 1 ) ⋯ exp ( x n e n ) , followed by projection onto the quotient space.
Let h ⊂ g \mathfrak{h}\subset\mathfrak{g} h ⊂ g be a subalgebra of codimension m m m . Choose a basis ( e 1 , … , e n ) (e_1,\ldots,e_n) ( e 1 , … , e n ) of the algebra g \mathfrak{g} g such that the vectors ( e m + 1 , … , e n ) (e_{m+1},\ldots,e_n) ( e m + 1 , … , e n ) form a basis of the subalgebra h \mathfrak{h} h . The component matrix of the Maurer–Cartan 1-form ω ( x ) \omega(x) ω ( x ) is computed via the adjoint representation:
ω ( x ) i j = ( exp ( − x n ad e n ) exp ( − x n − 1 ad e n − 1 ) ⋯ exp ( − x j + 1 ad e j + 1 ) ) i j , (4) \omega(x)_{ij}=\left(\exp(-x_n\operatorname{ad}_{e_n})\exp(-x_{n-1}\operatorname{ad}_{e_{n-1}})\cdots\exp(-x_{j+1}\operatorname{ad}_{e_{j+1}})\right)_{ij}, \tag*{(4)} ω ( x ) ij = ( exp ( − x n ad e n ) exp ( − x n − 1 ad e n − 1 ) ⋯ exp ( − x j + 1 ad e j + 1 ) ) ij , (4)
where the indices i i i and j j j correspond to the component number and the form number, respectively. Then the vector fields of the realization in local coordinates ( x 1 , … , x m ) (x_1,\ldots,x_m) ( x 1 , … , x m ) on G / H G/H G / H are obtained as projections of invariant vector fields: R ( e j ) = ∑ i = 1 m ( ω − 1 ( x 1 , … , x m , 0 , … , 0 ) ) i j ∂ ∂ x i R(e_j)=\sum_{i=1}^{m}\left(\omega^{-1}(x_1,\ldots,x_m,0,\ldots,0)\right)_{ij}\frac{\partial}{\partial x_i} R ( e j ) = ∑ i = 1 m ( ω − 1 ( x 1 , … , x m , 0 , … , 0 ) ) ij ∂ x i ∂ .
This method can be automated in computer algebra systems (Maple was used in the present work) and also generalized to intransitive cases. To verify the constructed intransitive realizations, inequivalent representatives obtained by the direct method [31 ] were used. In § 3 , this algebraic method will be modified and applied to construct parameterized realizations that yield faithful realizations after taking the limit in the contraction parameter.
Contractions of the algebra s l ( 2 , C ) \mathfrak{sl}(2,\mathbb{C}) sl ( 2 , C )
Contractions of three-dimensional Lie algebras over the real and complex fields were exhaustively investigated in [24 ] ; however, to construct limit transitions between realizations, it is necessary to align the bases of the algebras with chains of nested subalgebras and construct contraction matrices compatible with the chosen subalgebras. Therefore, we revisit this problem in the present section.
Using existence criteria for contractions, we determine which complex three-dimensional Lie algebras can be obtained by contracting the simple algebra s l ( 2 , C ) \mathfrak{sl}(2,\mathbb{C}) sl ( 2 , C ) . The most effective criterion for this algebra is the unimodularity criterion : if there exists a contraction g → g 0 \mathfrak{g}\to\mathfrak{g}_0 g → g 0 and the algebra g \mathfrak{g} g is unimodular, then the algebra g 0 \mathfrak{g}_0 g 0 is also unimodular. A Lie algebra g \mathfrak{g} g is called unimodular if tr ( ad v ) = 0 \operatorname{tr}(\operatorname{ad}_v)=0 tr ( ad v ) = 0 for all v ∈ g v\in\mathfrak{g} v ∈ g , where ad v ( w ) = [ v , w ] \operatorname{ad}_v(w)=[v,w] ad v ( w ) = [ v , w ] .
The algebra s l ( 2 , C ) \mathfrak{sl}(2,\mathbb{C}) sl ( 2 , C ) is unimodular. Checking the unimodularity condition for the remaining algebras in the classification list (2) shows that among them, only the algebras g 3.4 − 1 \mathfrak{g}_{3.4}^{-1} g 3.4 − 1 , g 3.1 \mathfrak{g}_{3.1} g 3.1 , and 3 g 1 3\mathfrak{g}_1 3 g 1 are unimodular.
Indeed, for g 0 = g 2.1 ⊕ g 1 \mathfrak{g}_0=\mathfrak{g}_{2.1}\oplus\mathfrak{g}_1 g 0 = g 2.1 ⊕ g 1 we have tr ( ad e 2 ) = − 1 ≠ 0 \operatorname{tr}(\operatorname{ad}_{e_2})=-1\ne0 tr ( ad e 2 ) = − 1 = 0 ; for g 3.2 \mathfrak{g}_{3.2} g 3.2 , tr ( ad e 3 ) = − 2 ≠ 0 \operatorname{tr}(\operatorname{ad}_{e_3})=-2\ne0 tr ( ad e 3 ) = − 2 = 0 ; for g 3.3 \mathfrak{g}_{3.3} g 3.3 , tr ( ad e 3 ) = − 2 ≠ 0 \operatorname{tr}(\operatorname{ad}_{e_3})=-2\ne0 tr ( ad e 3 ) = − 2 = 0 ; and for the family g 3.4 a \mathfrak{g}_{3.4}^{a} g 3.4 a we obtain tr ( ad e 3 ) = − ( 1 + a ) \operatorname{tr}(\operatorname{ad}_{e_3})=-(1+a) tr ( ad e 3 ) = − ( 1 + a ) , so only the algebra g 3.4 − 1 \mathfrak{g}_{3.4}^{-1} g 3.4 − 1 , corresponding to the parameter value a = − 1 a=-1 a = − 1 , is unimodular.
Thus, among algebras non-isomorphic to s l ( 2 , C ) \mathfrak{sl}(2,\mathbb{C}) sl ( 2 , C ) , the necessary unimodularity condition is satisfied only by g 3.4 − 1 \mathfrak{g}_{3.4}^{-1} g 3.4 − 1 , g 3.1 \mathfrak{g}_{3.1} g 3.1 , and 3 g 1 3\mathfrak{g}_1 3 g 1 . Let us compare the dimensions of their derivation algebras:
Table 4.
Under a proper contraction, the dimension of the derivation algebra increases strictly. Hence, for these four algebras, the possible direction of proper contractions is consistent with the ordering
s l ( 2 , C ) → g 3.4 − 1 → g 3.1 → 3 g 1 . (5) \mathfrak{sl}(2,\mathbb{C})\to\mathfrak{g}_{3.4}^{-1}\to\mathfrak{g}_{3.1}\to3\mathfrak{g}_1. \tag*{(5)} sl ( 2 , C ) → g 3.4 − 1 → g 3.1 → 3 g 1 . (5)
This criterion rules out proper contractions in the reverse direction, but by itself does not prove the existence of transitions in the direction of the arrows. The existence of nontrivial transitions between adjacent algebras, as well as the direct contraction s l ( 2 , C ) → g 3.1 \mathfrak{sl}(2,\mathbb{C}) \to\mathfrak{g}_{3.1} sl ( 2 , C ) → g 3.1 , is established by the matrices presented below.
In addition to transitions between adjacent algebras, the ordering above admits the pairs s l ( 2 , C ) → g 3.1 \mathfrak{sl}(2,\mathbb{C}) \to\mathfrak{g}_{3.1} sl ( 2 , C ) → g 3.1 , s l ( 2 , C ) → 3 g 1 \mathfrak{sl}(2,\mathbb{C}) \to3\mathfrak{g}_{1} sl ( 2 , C ) → 3 g 1 , and g 3.4 − 1 → 3 g 1 \mathfrak{g}_{3.4}^{-1} \to3\mathfrak{g}_{1} g 3.4 − 1 → 3 g 1 . Their existence also follows from the matrix U 3 ( ε ) U_{3}(\varepsilon) U 3 ( ε ) given below and from the diagonal contraction to the Abelian algebra.
In this paper, we omit the commutative algebra and consider only three of the six possible contraction pairs, namely s l ( 2 , C ) → g 3.4 − 1 \mathfrak{sl}(2,\mathbb{C}) \to\mathfrak{g}_{3.4}^{-1} sl ( 2 , C ) → g 3.4 − 1 , s l ( 2 , C ) → g 3.1 \mathfrak{sl}(2,\mathbb{C}) \to\mathfrak{g}_{3.1} sl ( 2 , C ) → g 3.1 , and g 3.4 − 1 → g 3.1 \mathfrak{g}_{3.4}^{-1} \to\mathfrak{g}_{3.1} g 3.4 − 1 → g 3.1 . It is for these contractions that we address the construction of limit realizations and, in one case, of invariants.
Contraction to the Abelian algebra is well known and can be realized for any algebra using a diagonal matrix with the contraction parameter on the diagonal. Indeed, for U ( ε ) = ε I U(\varepsilon)=\varepsilon I U ( ε ) = ε I , we have [ v , w ] ε = ε [ v , w ] → 0 [v,w]_{\varepsilon}=\varepsilon[v,w]\to0 [ v , w ] ε = ε [ v , w ] → 0 . In the present work, primary attention is given to nontrivial limit realizations and the corresponding invariants, so the Abelian algebra is excluded from consideration.
Consequently, up to isomorphism of the resulting algebra, the two nontrivial proper contractions of s l ( 2 , C ) \mathfrak{sl}(2,\mathbb{C}) sl ( 2 , C ) listed above exhaust all possible cases. This statement applies to classes of limit algebras; it is not a classification of all contraction matrices or all limit transitions between their realizations.
For each of the pairs under consideration, we construct a corresponding matrix as a product U ( ε ) = I ⋅ W ( ε ) U(\varepsilon)=I\cdot W(\varepsilon) U ( ε ) = I ⋅ W ( ε ) , where I I I is a constant nonsingular matrix and W ( ε ) W(\varepsilon) W ( ε ) is a diagonal matrix with exponents 0 and 1. The matrix I I I defines a preliminary change of basis in the initial algebra, after which W ( ε ) W(\varepsilon) W ( ε ) specifies an Inönü–Wigner contraction. Without such a change of basis, the matrix W ( ε ) W(\varepsilon) W ( ε ) alone does not necessarily define the same contraction.
To perform contractions exclusively via diagonal matrices (when a GIW contraction exists), one must first align the bases of the initial and resulting algebras. The problem of aligning bases for all representatives of non-isomorphic Lie algebras of a given dimension is extraordinarily difficult. A similar issue arises when constructing optimal subalgebra trees in order to minimize computations needed to obtain all inequivalent realizations, as well as when choosing optimal bases to construct the Hasse diagram of contractions. For certain small sets of algebras, one can select aligned bases starting from the algebra at the end of the chain, though this is not always possible. Moreover, there exist pairs of algebras for which constructing a GIW contraction requires simultaneous changes of basis in both the initial and target algebras.
The contraction s l ( 2 , C ) → g 3.4 − 1 \mathfrak{sl}(2,\mathbb{C}) \to\mathfrak{g}_{3.4}^{-1} sl ( 2 , C ) → g 3.4 − 1 is realized by the matrix U 1 ( ε ) = I 1 ⋅ W 1 ( ε ) U_{1}(\varepsilon)=I_{1}\cdot W_{1}(\varepsilon) U 1 ( ε ) = I 1 ⋅ W 1 ( ε ) , where
I 1 = ( 1 0 0 0 0 1 0 1 0 ) , W 1 ( ε ) = ( ε 0 0 0 1 0 0 0 1 ) , U 1 ( ε ) = ( ε 0 0 0 0 1 0 1 0 ) . I_{1}=\begin{pmatrix}1&0&0\\0&0&1\\0&1&0\end{pmatrix},\qquad W_{1}(\varepsilon)=\begin{pmatrix}\varepsilon&0&0\\0&1&0\\0&0&1\end{pmatrix},\qquad U_{1}(\varepsilon)=\begin{pmatrix}\varepsilon&0&0\\0&0&1\\0&1&0\end{pmatrix}. I 1 = 1 0 0 0 0 1 0 1 0 , W 1 ( ε ) = ε 0 0 0 1 0 0 0 1 , U 1 ( ε ) = ε 0 0 0 0 1 0 1 0 .
This matrix corresponds to the subalgebra ⟨ e 2 , e 3 ⟩ \langle e_{2},e_{3}\rangle ⟨ e 2 , e 3 ⟩ . Applying (3) , we obtain the parameterized structure constants that will be used in § 3 :
c 12 3 ( ε ) = − c ~ 21 3 ( ε ) = 2 ε → 0 , ε → 0 ; c 13 1 ( ε ) = − c ~ 31 1 ( ε ) = 1 → 1 , ε → 0 ; c 23 2 ( ε ) = − c ~ 32 2 ( ε ) = − 1 → − 1 , ε → 0. (6) \begin{aligned}
c_{12}^{3}(\varepsilon)&=-\widetilde{c}_{21}^{3}(\varepsilon)=2\varepsilon\to0,\qquad\varepsilon\to0;\\
c_{13}^{1}(\varepsilon)&=-\widetilde{c}_{31}^{1}(\varepsilon)=1\to1,\qquad\varepsilon\to0;\\
c_{23}^{2}(\varepsilon)&=-\widetilde{c}_{32}^{2}(\varepsilon)=-1\to-1,\qquad\varepsilon\to0. \tag*{(6)}
\end{aligned} c 12 3 ( ε ) c 13 1 ( ε ) c 23 2 ( ε ) = − c 21 3 ( ε ) = 2 ε → 0 , ε → 0 ; = − c 31 1 ( ε ) = 1 → 1 , ε → 0 ; = − c 32 2 ( ε ) = − 1 → − 1 , ε → 0. (6)
The limit structure constants define the algebra g 3.4 − 1 \mathfrak{g}_{3.4}^{-1} g 3.4 − 1 .
Similarly, the contraction g 3.4 − 1 → g 3.1 \mathfrak{g}_{3.4}^{-1}\to\mathfrak{g}_{3.1} g 3.4 − 1 → g 3.1 is realized by the matrix U 2 ( ε ) = I 2 ⋅ W 2 ( ε ) U_{2}(\varepsilon)=I_{2}\cdot W_{2}(\varepsilon) U 2 ( ε ) = I 2 ⋅ W 2 ( ε ) , which corresponds to the subalgebra ⟨ e 1 + e 2 ⟩ \langle e_{1}+e_{2}\rangle ⟨ e 1 + e 2 ⟩ :
I 2 = ( 2 1 0 0 1 0 0 0 1 ) , W 2 ( ε ) = ( ε 0 0 0 1 0 0 0 ε ) , U 2 ( ε ) = ( 2 ε 1 0 0 1 0 0 0 ε ) . I_{2}=\begin{pmatrix}2&1&0\\0&1&0\\0&0&1\end{pmatrix},\qquad W_{2}(\varepsilon)=\begin{pmatrix}\varepsilon&0&0\\0&1&0\\0&0&\varepsilon\end{pmatrix},\qquad U_{2}(\varepsilon)=\begin{pmatrix}2\varepsilon&1&0\\0&1&0\\0&0&\varepsilon\end{pmatrix}. I 2 = 2 0 0 1 1 0 0 0 1 , W 2 ( ε ) = ε 0 0 0 1 0 0 0 ε , U 2 ( ε ) = 2 ε 0 0 1 1 0 0 0 ε .
c 13 1 ( ε ) = − c ~ 31 1 ( ε ) = ε → 0 , ε → 0 ; c^{1}_{13}(\varepsilon)=-\widetilde{c}^{1}_{31}(\varepsilon)=\varepsilon\to0,\quad\varepsilon\to0; c 13 1 ( ε ) = − c 31 1 ( ε ) = ε → 0 , ε → 0 ;
c 23 1 ( ε ) = − c ~ 32 1 ( ε ) = 1 → 1 , ε → 0 ; c^{1}_{23}(\varepsilon)=-\widetilde{c}^{1}_{32}(\varepsilon)=1\to1,\quad\varepsilon\to0; c 23 1 ( ε ) = − c 32 1 ( ε ) = 1 → 1 , ε → 0 ;
c 23 2 ( ε ) = − c ~ 32 2 ( ε ) = − ε → 0 , ε → 0. (7) c^{2}_{23}(\varepsilon)=-\widetilde{c}^{2}_{32}(\varepsilon)=-\varepsilon\to0,\quad\varepsilon\to0. \tag*{(7)} c 23 2 ( ε ) = − c 32 2 ( ε ) = − ε → 0 , ε → 0. (7)
There exists an alternative matrix U ~ 2 ( ε ) \widetilde{U}_{2}(\varepsilon) U 2 ( ε ) realizing the same contraction g 3.4 − 1 → g 3.1 \mathfrak{g}_{3.4}^{-1}\to\mathfrak{g}_{3.1} g 3.4 − 1 → g 3.1 and preserving the same subalgebra ⟨ e 1 + e 2 ⟩ \langle e_{1}+e_{2}\rangle ⟨ e 1 + e 2 ⟩ , which has the form
I ~ 2 = ( 1 1 0 − 1 1 0 0 0 1 ) , W ~ 2 ( ε ) = ( ε 0 0 0 1 0 0 0 ε ) , U ~ 2 ( ε ) = ( ε 1 0 − ε 1 0 0 0 ε ) . \widetilde{I}_{2}=\begin{pmatrix}1&1&0\\-1&1&0\\0&0&1\end{pmatrix},\quad
\widetilde{W}_{2}(\varepsilon)=\begin{pmatrix}\varepsilon&0&0\\0&1&0\\0&0&\varepsilon\end{pmatrix},\quad
\widetilde{U}_{2}(\varepsilon)=\begin{pmatrix}\varepsilon&1&0\\-\varepsilon&1&0\\0&0&\varepsilon\end{pmatrix}. I 2 = 1 − 1 0 1 1 0 0 0 1 , W 2 ( ε ) = ε 0 0 0 1 0 0 0 ε , U 2 ( ε ) = ε − ε 0 1 1 0 0 0 ε .
c 13 2 ( ε ) = − c ~ 31 2 ( ε ) = ε 2 → 0 , ε → 0 ; c^{2}_{13}(\varepsilon)=-\widetilde{c}^{2}_{31}(\varepsilon)=\varepsilon^{2}\to0,\quad\varepsilon\to0; c 13 2 ( ε ) = − c 31 2 ( ε ) = ε 2 → 0 , ε → 0 ;
c 23 1 ( ε ) = − c ~ 32 1 ( ε ) = 1 → 1 , ε → 0. (8) c^{1}_{23}(\varepsilon)=-\widetilde{c}^{1}_{32}(\varepsilon)=1\to1,\quad\varepsilon\to0. \tag*{(8)} c 23 1 ( ε ) = − c 32 1 ( ε ) = 1 → 1 , ε → 0. (8)
The contraction s l ( 2 , C ) → g 3.1 \mathfrak{sl}(2,\mathbb{C})\to\mathfrak{g}_{3.1} sl ( 2 , C ) → g 3.1 can be realized by the matrix U 3 ( ε ) = I 3 ⋅ W 3 ( ε ) U_{3}(\varepsilon)=I_{3}\cdot W_{3}(\varepsilon) U 3 ( ε ) = I 3 ⋅ W 3 ( ε ) , which preserves the subalgebra ⟨ e 3 ⟩ \langle e_{3}\rangle ⟨ e 3 ⟩ :
I 3 = ( 0 1 0 1 0 0 0 0 1 2 ) , W 3 ( ε ) = ( ε 0 0 0 ε 0 0 0 1 ) , U 3 ( ε ) = ( 0 ε 0 ε 0 0 0 0 1 2 ) . I_{3}=\begin{pmatrix}0&1&0\\1&0&0\\0&0&\frac{1}{2}\end{pmatrix},\quad
W_{3}(\varepsilon)=\begin{pmatrix}\varepsilon&0&0\\0&\varepsilon&0\\0&0&1\end{pmatrix},\quad
U_{3}(\varepsilon)=\begin{pmatrix}0&\varepsilon&0\\\varepsilon&0&0\\0&0&\frac{1}{2}\end{pmatrix}. I 3 = 0 1 0 1 0 0 0 0 2 1 , W 3 ( ε ) = ε 0 0 0 ε 0 0 0 1 , U 3 ( ε ) = 0 ε 0 ε 0 0 0 0 2 1 .
c 12 2 ( ε ) = − c ~ 21 2 ( ε ) = − ε → 0 , ε → 0 ; c^{2}_{12}(\varepsilon)=-\widetilde{c}^{2}_{21}(\varepsilon)=-\varepsilon\to0,\quad\varepsilon\to0; c 12 2 ( ε ) = − c 21 2 ( ε ) = − ε → 0 , ε → 0 ;
c 13 3 ( ε ) = − c ~ 31 3 ( ε ) = ε → 0 , ε → 0 ; c^{3}_{13}(\varepsilon)=-\widetilde{c}^{3}_{31}(\varepsilon)=\varepsilon\to0,\quad\varepsilon\to0; c 13 3 ( ε ) = − c 31 3 ( ε ) = ε → 0 , ε → 0 ;
c 23 1 ( ε ) = − c ~ 32 1 ( ε ) = 1 → 1 , ε → 0. (9) c^{1}_{23}(\varepsilon)=-\widetilde{c}^{1}_{32}(\varepsilon)=1\to1,\quad\varepsilon\to0. \tag*{(9)} c 23 1 ( ε ) = − c 32 1 ( ε ) = 1 → 1 , ε → 0. (9)
Parameterized realizations and their contractions
Section 1 described the algebraic method for constructing realizations, whereby each transitive local realization of an algebra g \mathfrak{g} g by vector fields on an m m m -dimensional manifold M M M is determined by a pair ( g , h ) (\mathfrak{g},\mathfrak{h}) ( g , h ) , where h ⊂ g \mathfrak{h}\subset\mathfrak{g} h ⊂ g is a subalgebra of codimension m m m playing the role of the stationary subalgebra.
The practical algorithm for constructing a realization consists of the following steps:
Choose a basis { e 1 , … , e n } \{e_{1},\ldots,e_{n}\} { e 1 , … , e n } of the algebra g \mathfrak{g} g such that the vectors e m + 1 , … , e n e_{m+1},\ldots,e_{n} e m + 1 , … , e n form a basis of the subalgebra h \mathfrak{h} h .
Construct the matrices of the adjoint representation ( ad i ) j k = c i j k (\operatorname{ad}_{i})^{k}_{j}=c^{k}_{ij} ( ad i ) j k = c ij k .
In canonical coordinates of the second kind, compute the component matrix of the Maurer–Cartan form. To be specific, set
g ( x ) = exp ( − x 1 e 1 ) ⋯ exp ( − x n e n ) , d g g − 1 = ∑ i , j = 1 n ω j i ( x ) e i d x j . g(x)=\exp(-x_{1}e_{1})\cdots\exp(-x_{n}e_{n}),\qquad
dg\,g^{-1}=\sum_{i,j=1}^{n}\omega^{i}_{j}(x)e_{i}\,dx_{j}. g ( x ) = exp ( − x 1 e 1 ) ⋯ exp ( − x n e n ) , d g g − 1 = i , j = 1 ∑ n ω j i ( x ) e i d x j .
Then the j j j -th column of the matrix ω \omega ω is given by
ω ⋅ j ( x ) = − exp ( − x 1 ad e 1 ) ⋯ exp ( − x j − 1 ad e j − 1 ) e j , \omega_{\cdot j}(x)=-\exp(-x_{1}\operatorname{ad}_{e_{1}})\cdots\exp(-x_{j-1}\operatorname{ad}_{e_{j-1}})e_{j}, ω ⋅ j ( x ) = − exp ( − x 1 ad e 1 ) ⋯ exp ( − x j − 1 ad e j − 1 ) e j ,
where e j e_{j} e j on the right-hand side is viewed as the coordinate column vector of the basis element. For j = 1 j=1 j = 1 , the product of exponentials equals the identity matrix.
Find the vector fields
Ξ j = − ∑ i = 1 n ( ω − 1 ) i j ∂ ∂ x i , j = 1 , … , n . \Xi_j=-\sum_{i=1}^{n}(\omega^{-1})^i{}_j\frac{\partial}{\partial x_i},\qquad j=1,\ldots,n. Ξ j = − i = 1 ∑ n ( ω − 1 ) i j ∂ x i ∂ , j = 1 , … , n .
The chosen sign ensures the commutation relations [ Ξ i , Ξ j ] = c i j k Ξ k [\Xi_i,\Xi_j]=c^k_{ij}\Xi_k [ Ξ i , Ξ j ] = c ij k Ξ k .
Project the vector fields onto the quotient space G / H G/H G / H for the right action of the subgroup H H H on $G. In the chosen local coordinates, x m + 1 , … , x n x_{m+1},\ldots,x_n x m + 1 , … , x n are coordinates along the fibers. Once projectability of the fields is verified, their components along ∂ / ∂ x 1 , … , ∂ / ∂ x m \partial/\partial x_1,\ldots,\partial/\partial x_m ∂ / ∂ x 1 , … , ∂ / ∂ x m are evaluated on the section x m + 1 = ⋯ = x n = 0 x_{m+1}=\cdots=x_n=0 x m + 1 = ⋯ = x n = 0 . It is projectability, rather than merely setting these coordinates to zero, that guarantees the validity of this procedure.
In formula (5), for the coordinates g = exp ( x 1 e 1 ) ⋯ exp ( x n e n ) g=\exp(x_1e_1)\cdots\exp(x_ne_n) g = exp ( x 1 e 1 ) ⋯ exp ( x n e n ) , the left-invariant form g − 1 d g g^{-1}dg g − 1 d g is used. Its dual system consists of left-invariant vector fields, which are not in general projectable onto G / H G/H G / H under the right action of H H H . Therefore, the steps above employ the compatible right-invariant form. The matrix construction of invariant forms and vector fields is presented in [21 ] .
As an example, consider the Weyl–Heisenberg algebra with the sole nonzero commutator [ e 2 , e 3 ] = e 1 [e_2,e_3]=e_1 [ e 2 , e 3 ] = e 1 and the subalgebra h = ⟨ e 3 ⟩ \mathfrak{h}=\langle e_3\rangle h = ⟨ e 3 ⟩ . For g = exp ( − x 1 e 1 ) exp ( − x 2 e 2 ) exp ( − x 3 e 3 ) g=\exp(-x_1e_1)\exp(-x_2e_2)\exp(-x_3e_3) g = exp ( − x 1 e 1 ) exp ( − x 2 e 2 ) exp ( − x 3 e 3 ) , we have
ω = ( − 1 0 x 2 0 − 1 0 0 0 − 1 ) , − ω − 1 = ( 1 0 x 2 0 1 0 0 0 1 ) . \omega=\begin{pmatrix}-1&0&x_2\\0&-1&0\\0&0&-1\end{pmatrix},\qquad
-\omega^{-1}=\begin{pmatrix}1&0&x_2\\0&1&0\\0&0&1\end{pmatrix}. ω = − 1 0 0 0 − 1 0 x 2 0 − 1 , − ω − 1 = 1 0 0 0 1 0 x 2 0 1 .
Thus, on the group we obtain the fields Ξ 1 = ∂ 1 \Xi_1=\partial_1 Ξ 1 = ∂ 1 , Ξ 2 = ∂ 2 \Xi_2=\partial_2 Ξ 2 = ∂ 2 , Ξ 3 = x 2 ∂ 1 + ∂ 3 \Xi_3=x_2\partial_1+\partial_3 Ξ 3 = x 2 ∂ 1 + ∂ 3 . Their first two components do not depend on the fiber coordinate x 3 x_3 x 3 , so after projecting onto G / H G/H G / H we get
R ( e 1 ) = ∂ 1 , R ( e 2 ) = ∂ 2 , R ( e 3 ) = x 2 ∂ 1 . R(e_1)=\partial_1,\qquad R(e_2)=\partial_2,\qquad R(e_3)=x_2\partial_1. R ( e 1 ) = ∂ 1 , R ( e 2 ) = ∂ 2 , R ( e 3 ) = x 2 ∂ 1 .
Direct verification yields [ R ( e 2 ) , R ( e 3 ) ] = R ( e 1 ) [R(e_2),R(e_3)]=R(e_1) [ R ( e 2 ) , R ( e 3 )] = R ( e 1 ) , with all remaining commutators equal to zero.
The key idea in modifying the algorithm to find contractions of realizations is to substitute the parameterized structure constants c ~ i j k ( ε ) \widetilde{c}^k_{ij}(\varepsilon) c ij k ( ε ) from (4) into the algorithm instead of the fixed structure constants c i j k c^k_{ij} c ij k . This leads to ε \varepsilon ε -dependent matrices of the adjoint representation
( ad j ) k i ( ε ) = c ~ i j k ( ε ) = ( ( U ( ε ) − 1 ) k k ′ ( U ( ε ) ) i ′ i ( U ( ε ) ) j ′ j c k ′ i ′ j ′ ) , (10) (\operatorname{ad}_j)^k{}_i(\varepsilon)=\widetilde{c}^k_{ij}(\varepsilon)=\left((U(\varepsilon)^{-1})^k{}_{k'}(U(\varepsilon))^{i'}{}_i(U(\varepsilon))^{j'}{}_j c^{k'}{}_{i'j'}\right), \tag*{(10)} ( ad j ) k i ( ε ) = c ij k ( ε ) = ( ( U ( ε ) − 1 ) k k ′ ( U ( ε ) ) i ′ i ( U ( ε ) ) j ′ j c k ′ i ′ j ′ ) , (10)
and hence to an ε \varepsilon ε -dependent Maurer–Cartan form matrix ω ε ( x ) \omega_\varepsilon(x) ω ε ( x ) and vector fields
Ξ j ( ε ) = − ∑ i = 1 n ( ω ε − 1 ) i j ∂ ∂ x i . \Xi_j(\varepsilon)=-\sum_{i=1}^{n}(\omega_\varepsilon^{-1})^i{}_j\frac{\partial}{\partial x_i}. Ξ j ( ε ) = − i = 1 ∑ n ( ω ε − 1 ) i j ∂ x i ∂ .
Upon projection, we obtain a parameterized realization R ε R^\varepsilon R ε of the algebra g ε = ( V , [ ⋅ , ⋅ ] ε ) \mathfrak{g}_\varepsilon=(V,[\cdot,\cdot]_\varepsilon) g ε = ( V , [ ⋅ , ⋅ ] ε ) :
[ R ε ( e i ) , R ε ( e j ) ] = ∑ k c ~ i j k ( ε ) R ε ( e k ) . [R^\varepsilon(e_i),R^\varepsilon(e_j)]=\sum_k\widetilde{c}^k_{ij}(\varepsilon)R^\varepsilon(e_k). [ R ε ( e i ) , R ε ( e j )] = k ∑ c ij k ( ε ) R ε ( e k ) .
For ε ≠ 0 \varepsilon\ne0 ε = 0 , the map U ( ε ) : g ε → g U(\varepsilon):\mathfrak{g}_\varepsilon\to\mathfrak{g} U ( ε ) : g ε → g is an isomorphism. Therefore, a realization of the original fixed algebra g \mathfrak{g} g is obtained as R ε ∘ U ( ε ) − 1 R^\varepsilon\circ U(\varepsilon)^{-1} R ε ∘ U ( ε ) − 1 . It is after this identification that the definition of weak equivalence of realizations of a fixed algebra is applied.
By a limit transition between realizations, we mean the convergence of the coefficients of the fields R ε ( e i ) R^\varepsilon(e_i) R ε ( e i ) to the coefficients of the fields R 0 ( e i ) R^0(e_i) R 0 ( e i ) in common local coordinates under the conditions specified below. A contraction of an abstract algebra alone does not uniquely determine a limit transition between realizations: one must also specify a family of realizations, coordinates, stationary subalgebras, and the mode of convergence.
Formula (11) admits a simple matrix interpretation:
ad j ( ε ) = ( U ( ε ) ) − 1 ( ( U ( ε ) ) j j ′ ad j ′ ) U ( ε ) . \operatorname{ad}_{j}(\varepsilon) = (U(\varepsilon))^{-1}\left((U(\varepsilon))^{j'}_{j}\operatorname{ad}_{j'}\right)U(\varepsilon). ad j ( ε ) = ( U ( ε ) ) − 1 ( ( U ( ε ) ) j j ′ ad j ′ ) U ( ε ) .
where the expression in parentheses is a linear combination of the initial adjoint representation matrices with ε \varepsilon ε -dependent coefficients.
For commutators to pass to the limit, local C 1 C^{1} C 1 -convergence of the vector field coefficients suffices. Pointwise convergence of the matrices ω ε \omega_{\varepsilon} ω ε alone is not sufficient for this purpose. For prolongations to pass to the limit, convergence of the corresponding number of derivatives is required.
In the matrix construction above, the convergence c ~ i j k ( ε ) → c ^ i j k \widetilde{c}^{k}_{ij}(\varepsilon)\to\widehat{c}^{k}_{ij} c ij k ( ε ) → c ij k implies local uniform convergence of the matrix exponentials and their coordinate derivatives. Since ω 0 ( 0 ) = − I \omega_{0}(0)=-I ω 0 ( 0 ) = − I , one can fix a common neighborhood of the origin in which ω ε \omega_{\varepsilon} ω ε are nonsingular for all sufficiently small ε \varepsilon ε . In this neighborhood, the inverse matrices and their coordinate derivatives also converge locally uniformly. With a compatible choice of projectable fields and local coordinates on the quotient space, we obtain in the limit a realization of the algebra g 0 \mathfrak{g}_{0} g 0 .
For a limit transition between invariant systems, one must additionally verify the convergence of the systems themselves and of the prolonged generators; the corresponding conditions are formulated in [16 ] .
Let h ⊂ g \mathfrak{h}\subset\mathfrak{g} h ⊂ g be a stationary subalgebra defining a transitive realization R R R of the algebra g \mathfrak{g} g . It corresponds to the family of subalgebras h ε = U ( ε ) − 1 h \mathfrak{h}_{\varepsilon}=U(\varepsilon)^{-1}\mathfrak{h} h ε = U ( ε ) − 1 h of the algebras g ε \mathfrak{g}_{\varepsilon} g ε . If h ε → h 0 \mathfrak{h}_{\varepsilon}\to\mathfrak{h}_{0} h ε → h 0 in the Grassmannian, then h 0 \mathfrak{h}_{0} h 0 is a subalgebra of g 0 \mathfrak{g}_{0} g 0 of the same dimension. Indeed, for convergent bases u i ( ε ) → u i ( 0 ) u_{i}(\varepsilon)\to u_{i}(0) u i ( ε ) → u i ( 0 ) , we have
[ u i ( ε ) , u j ( ε ) ] ε ⟶ [ u i ( 0 ) , u j ( 0 ) ] 0 ∈ h 0 . [u_{i}(\varepsilon),u_{j}(\varepsilon)]_{\varepsilon}\longrightarrow[u_{i}(0),u_{j}(0)]_{0}\in\mathfrak{h}_{0}. [ u i ( ε ) , u j ( ε ) ] ε ⟶ [ u i ( 0 ) , u j ( 0 ) ] 0 ∈ h 0 .
The full limit of a family of subspaces need not exist, but every Grassmannian limit point of it is a subalgebra. Moreover, diagonality of U ( ε ) U(\varepsilon) U ( ε ) by itself does not imply that an arbitrary subspace remains invariant. For instance, diag ( ε , 1 ) ⟨ e 1 + e 2 ⟩ = ⟨ ε e 1 + e 2 ⟩ \operatorname{diag}(\varepsilon,1)\langle e_{1}+e_{2}\rangle=\langle\varepsilon e_{1}+e_{2}\rangle diag ( ε , 1 ) ⟨ e 1 + e 2 ⟩ = ⟨ ε e 1 + e 2 ⟩ . Preservation of a subspace requires compatibility with the weight decomposition. Furthermore, the statement that a fixed subspace is a subalgebra for all g ε \mathfrak{g}_{\varepsilon} g ε and the statement U ( ε ) h = h U(\varepsilon)\mathfrak{h}=\mathfrak{h} U ( ε ) h = h are distinct.
In all cases considered in this paper, the subalgebras from which the realizations are constructed are listed in the corresponding tables, and their closure under the parameterized brackets is verified separately.
For a transitive local realization R R R with stationary subalgebra h \mathfrak{h} h , the standard relation ker R = core g ( h ) \ker R=\operatorname{core}_{\mathfrak{g}}(\mathfrak{h}) ker R = core g ( h ) holds, where core g ( h ) \operatorname{core}_{\mathfrak{g}}(\mathfrak{h}) core g ( h ) is the largest ideal of g \mathfrak{g} g contained in h \mathfrak{h} h . Consequently, R R R is faithful if and only if h \mathfrak{h} h contains no nonzero ideals of g \mathfrak{g} g .
For a regular family of orbit realizations with stationary subalgebras h α \mathfrak{h}_{\alpha} h α , the kernel of the entire realization equals ⋂ α core g ( h α ) \bigcap_{\alpha}\operatorname{core}_{\mathfrak{g}}(\mathfrak{h}_{\alpha}) ⋂ α core g ( h α ) . Thus, faithfulness of each orbit restriction is sufficient, though not necessary, for faithfulness of the entire family. Applying this well-known criterion to the limit realization does not constitute a separate criterion for preservation of faithfulness under contraction.
Under contraction, the situation may change: the limit subalgebra h 0 ⊂ g 0 \mathfrak{h}_{0}\subset\mathfrak{g}_{0} h 0 ⊂ g 0 may contain a nontrivial ideal of g 0 \mathfrak{g}_{0} g 0 , even if h \mathfrak{h} h contained no ideals of g \mathfrak{g} g . This is because upon contraction, the algebra g 0 \mathfrak{g}_{0} g 0 may possess a larger set of ideals than g \mathfrak{g} g (in particular, contraction can increase the dimension of the center and decrease the dimension of the derived subalgebra). The dimension of the derived subalgebra does not increase, since it is the rank of the map defined by the Lie bracket. For the chain extended to the Abelian algebra, these dimensions are 3 → 2 → 1 → 0 3\to2\to1\to0 3 → 2 → 1 → 0 . Nevertheless, in all cases considered below, we succeeded in constructing contractions in which the limit stationary subalgebras contain no ideals, so the corresponding limit realizations are faithful.
Contractions of realizations of the algebra s l ( 2 , C ) \mathfrak{sl}(2,\mathbb{C}) sl ( 2 , C )
Before constructing ε \varepsilon ε -realizations and examining their limit relations, we list the realizations of the three algebras under study obtained by the direct method.
List of inequivalent realizations of the algebra s l ( 2 , C ) \mathfrak{sl}(2,\mathbb{C}) sl ( 2 , C ) :
R 1 : ∂ 1 , x 1 ∂ 1 + x 2 ∂ 2 , x 1 2 ∂ 1 + 2 x 1 x 2 ∂ 2 + x 2 ∂ 3 ; R 2 : ∂ 1 , x 1 ∂ 1 + x 2 ∂ 2 , ( x 1 2 + x 2 2 ) ∂ 1 + 2 x 1 x 2 ∂ 2 ; R 3 : ∂ 1 , x 1 ∂ 1 + x 2 ∂ 2 , x 1 2 ∂ 1 + 2 x 1 x 2 ∂ 2 ; R 4 : ∂ 1 , x 1 ∂ 1 , x 1 2 ∂ 1 . (11) \begin{aligned}
\mathcal{R}_1 &: \partial_1,\ x_1\partial_1+x_2\partial_2,\ x_1^2\partial_1+2x_1x_2\partial_2+x_2\partial_3;\\
\mathcal{R}_2 &: \partial_1,\ x_1\partial_1+x_2\partial_2,\ (x_1^2+x_2^2)\partial_1+2x_1x_2\partial_2;\\
\mathcal{R}_3 &: \partial_1,\ x_1\partial_1+x_2\partial_2,\ x_1^2\partial_1+2x_1x_2\partial_2;\\
\mathcal{R}_4 &: \partial_1,\ x_1\partial_1,\ x_1^2\partial_1. \tag*{(11)}
\end{aligned} R 1 R 2 R 3 R 4 : ∂ 1 , x 1 ∂ 1 + x 2 ∂ 2 , x 1 2 ∂ 1 + 2 x 1 x 2 ∂ 2 + x 2 ∂ 3 ; : ∂ 1 , x 1 ∂ 1 + x 2 ∂ 2 , ( x 1 2 + x 2 2 ) ∂ 1 + 2 x 1 x 2 ∂ 2 ; : ∂ 1 , x 1 ∂ 1 + x 2 ∂ 2 , x 1 2 ∂ 1 + 2 x 1 x 2 ∂ 2 ; : ∂ 1 , x 1 ∂ 1 , x 1 2 ∂ 1 . (11)
List of inequivalent realizations of the algebra g 3.4 − 1 \mathfrak{g}_{3.4}^{-1} g 3.4 − 1 :
R 1 : ∂ 1 , ∂ 2 , x 1 ∂ 1 − x 2 ∂ 2 + ∂ 3 ; R 2 : ∂ 1 , ∂ 2 , x 1 ∂ 1 − x 2 ∂ 2 ; R 3 : ∂ 1 , x 2 ∂ 1 , x 1 ∂ 1 + 2 x 2 ∂ 2 . (12) \begin{aligned}
\mathcal{R}_1 &: \partial_1,\ \partial_2,\ x_1\partial_1-x_2\partial_2+\partial_3;\\
\mathcal{R}_2 &: \partial_1,\ \partial_2,\ x_1\partial_1-x_2\partial_2;\\
\mathcal{R}_3 &: \partial_1,\ x_2\partial_1,\ x_1\partial_1+2x_2\partial_2. \tag*{(12)}
\end{aligned} R 1 R 2 R 3 : ∂ 1 , ∂ 2 , x 1 ∂ 1 − x 2 ∂ 2 + ∂ 3 ; : ∂ 1 , ∂ 2 , x 1 ∂ 1 − x 2 ∂ 2 ; : ∂ 1 , x 2 ∂ 1 , x 1 ∂ 1 + 2 x 2 ∂ 2 . (12)
List of inequivalent realizations of the algebra g 3.1 \mathfrak{g}_{3.1} g 3.1 :
R 1 : ∂ 1 , ∂ 2 , x 2 ∂ 1 + ∂ 3 ; R 2 : ∂ 1 , ∂ 2 , x 2 ∂ 1 + x 3 ∂ 2 ; R 3 : ∂ 1 , ∂ 2 , x 2 ∂ 1 . (13) \begin{aligned}
\mathcal{R}_1 &: \partial_1,\ \partial_2,\ x_2\partial_1+\partial_3;\\
\mathcal{R}_2 &: \partial_1,\ \partial_2,\ x_2\partial_1+x_3\partial_2;\\
\mathcal{R}_3 &: \partial_1,\ \partial_2,\ x_2\partial_1. \tag*{(13)}
\end{aligned} R 1 R 2 R 3 : ∂ 1 , ∂ 2 , x 2 ∂ 1 + ∂ 3 ; : ∂ 1 , ∂ 2 , x 2 ∂ 1 + x 3 ∂ 2 ; : ∂ 1 , ∂ 2 , x 2 ∂ 1 . (13)
If we directly apply a contraction matrix as a linear transformation of a Lie algebra of vector fields, we may obtain an unfaithful realization. The following example shows that this can occur for a particular choice of the initial realization. Indeed, consider the realization R 1 \mathcal{R}_1 R 1 of the algebra g 3.4 − 1 \mathfrak{g}_{3.4}^{-1} g 3.4 − 1 and perform the linear transformation corresponding to the contraction matrix U 2 ( ε ) U_2(\varepsilon) U 2 ( ε ) . Taking the limit as ε → 0 \varepsilon\to0 ε → 0 , we obtain an unfaithful realization of the algebra g 3.1 \mathfrak{g}_{3.1} g 3.1 :
R 1 ε ( e 1 ) = 2 ε ∂ 1 → 0 , R 1 ε ( e 2 ) = ∂ 1 + ∂ 2 → ∂ 1 + ∂ 2 , R 1 ε ( e 3 ) = ε ( x 1 ∂ 1 − x 2 ∂ 2 + ∂ 3 ) → 0. \begin{aligned}
\mathcal{R}_1^\varepsilon(e_1)&=2\varepsilon\partial_1\to0,\\
\mathcal{R}_1^\varepsilon(e_2)&=\partial_1+\partial_2\to\partial_1+\partial_2,\\
\mathcal{R}_1^\varepsilon(e_3)&=\varepsilon(x_1\partial_1-x_2\partial_2+\partial_3)\to0.
\end{aligned} R 1 ε ( e 1 ) R 1 ε ( e 2 ) R 1 ε ( e 3 ) = 2 ε ∂ 1 → 0 , = ∂ 1 + ∂ 2 → ∂ 1 + ∂ 2 , = ε ( x 1 ∂ 1 − x 2 ∂ 2 + ∂ 3 ) → 0.
Therefore, in order to use the contraction matrices already obtained to find limit transitions between realizations, it is necessary to construct a parameterized family of realizations for the parameterized family of Lie algebras g ε \mathfrak{g}_\varepsilon g ε with structure constants c ~ ( ε ) \widetilde{c}(\varepsilon) c ( ε ) . To construct ε \varepsilon ε -realizations, we apply the modified algebraic method described above.
Let us construct a parameterized transitive realization of g ε \mathfrak{g}_\varepsilon g ε with respect to the zero subalgebra, using the parameterized structure constants tensor from (7) . After identifying g ε ≃ g 3.4 − 1 \mathfrak{g}_\varepsilon\simeq\mathfrak{g}_{3.4}^{-1} g ε ≃ g 3.4 − 1 for ε ≠ 0 \varepsilon\ne0 ε = 0 , we obtain a realization weakly equivalent to R 1 [ g 3.4 − 1 ] \mathcal{R}_1[\mathfrak{g}_{3.4}^{-1}] R 1 [ g 3.4 − 1 ] . This family converges to the realization R 1 [ g 3.1 ] \mathcal{R}_1[\mathfrak{g}_{3.1}] R 1 [ g 3.1 ] as ε → 0 \varepsilon\to0 ε → 0 :
R ε ( e 1 ) = ∂ 1 → ∂ 1 , ε → 0 ; R ε ( e 2 ) = ∂ 2 → ∂ 2 , ε → 0 ; R ε ( e 3 ) = ( ε x 1 + x 2 ) ∂ 1 − ε x 2 ∂ 2 + ∂ 3 → x 2 ∂ 1 + ∂ 3 , ε → 0. \begin{aligned}
\mathcal{R}^\varepsilon(e_1)&=\partial_1\to\partial_1,\quad\varepsilon\to0;\\
\mathcal{R}^\varepsilon(e_2)&=\partial_2\to\partial_2,\quad\varepsilon\to0;\\
\mathcal{R}^\varepsilon(e_3)&=(\varepsilon x_1+x_2)\partial_1-\varepsilon x_2\partial_2+\partial_3\to x_2\partial_1+\partial_3,\quad\varepsilon\to0.
\end{aligned} R ε ( e 1 ) R ε ( e 2 ) R ε ( e 3 ) = ∂ 1 → ∂ 1 , ε → 0 ; = ∂ 2 → ∂ 2 , ε → 0 ; = ( ε x 1 + x 2 ) ∂ 1 − ε x 2 ∂ 2 + ∂ 3 → x 2 ∂ 1 + ∂ 3 , ε → 0.
Consider one more example of constructing a parameterized realization for the algebra s l ( 2 , C ) \mathfrak{sl}(2,\mathbb{C}) sl ( 2 , C ) contracting to g 3.1 \mathfrak{g}_{3.1} g 3.1 . For the zero subalgebra h 0 = { 0 } \mathfrak{h}_0=\{0\} h 0 = { 0 } and the contraction matrix U 3 ( ε ) U_3(\varepsilon) U 3 ( ε ) , we obtain a family of realizations converging to R 1 [ g 3.1 ] \mathcal{R}_1[\mathfrak{g}_{3.1}] R 1 [ g 3.1 ] :
R 1 ε ( e 1 ) = ∂ 1 → ∂ 1 , ε → 0 ; R 1 ε ( e 2 ) = e − ε x 1 ∂ 2 → ∂ 2 , ε → 0 ; R 1 ε ( e 3 ) = x 2 e ε x 1 ∂ 1 − 1 2 e ε x 1 ε x 2 2 ∂ 2 + e ε x 1 ∂ 3 → x 2 ∂ 1 + ∂ 3 , ε → 0. \begin{aligned}
\mathcal{R}_1^\varepsilon(e_1)&=\partial_1\to\partial_1,\quad\varepsilon\to0;\\
\mathcal{R}_1^\varepsilon(e_2)&=e^{-\varepsilon x_1}\partial_2\to\partial_2,\quad\varepsilon\to0;\\
\mathcal{R}_1^\varepsilon(e_3)&=x_2e^{\varepsilon x_1}\partial_1-\frac{1}{2}e^{\varepsilon x_1}\varepsilon x_2^2\partial_2+e^{\varepsilon x_1}\partial_3\to x_2\partial_1+\partial_3,\quad\varepsilon\to0.
\end{aligned} R 1 ε ( e 1 ) R 1 ε ( e 2 ) R 1 ε ( e 3 ) = ∂ 1 → ∂ 1 , ε → 0 ; = e − ε x 1 ∂ 2 → ∂ 2 , ε → 0 ; = x 2 e ε x 1 ∂ 1 − 2 1 e ε x 1 ε x 2 2 ∂ 2 + e ε x 1 ∂ 3 → x 2 ∂ 1 + ∂ 3 , ε → 0.
Similarly, applying the corresponding contraction matrices and the modified algebraic method to the other pairs of algebras yields the results compiled in the tables and analyzed below.
Each table below specifies the subalgebra h \mathfrak{h} h and complementary set of elements h ˉ \bar{\mathfrak{h}} h ˉ used to construct each parameterized realization, as well as the corresponding canonical realizations from the lists (12), (13), and (14) to which they are equivalent. In non-obvious cases, the corresponding biholomorphisms and changes of basis are written out explicitly after the tables. Equivalence for ε ≠ 0 \varepsilon\ne0 ε = 0 is understood after identifying g ε \mathfrak{g}_{\varepsilon} g ε with the corresponding initial algebra.
The equivalence of the parameterized realization R 1 ε R_{1}^{\varepsilon} R 1 ε for ε ≠ 0 \varepsilon\ne0 ε = 0 to the canonical case R 1 R_{1} R 1 from (12) is established by the biholomorphism x ~ 1 = x 1 \tilde{x}_{1}=x_{1} x ~ 1 = x 1 , x ~ 2 = e ε x 3 \tilde{x}_{2}=e^{\varepsilon x_{3}} x ~ 2 = e ε x 3 , x ~ 3 = ε x 2 e ε x 3 \tilde{x}_{3}=\varepsilon x_{2}e^{\varepsilon x_{3}} x ~ 3 = ε x 2 e ε x 3 and the change of basis e ~ 1 = e 1 \tilde{e}_{1}=e_{1} e ~ 1 = e 1 , e ~ 2 = e 3 \tilde{e}_{2}=e_{3} e ~ 2 = e 3 , e ~ 3 = 1 ε e 2 \tilde{e}_{3}=\frac{1}{\varepsilon}e_{2} e ~ 3 = ε 1 e 2 .
The equivalence of the realization R 2 ε R_{2}^{\varepsilon} R 2 ε constructed from the subalgebra ⟨ e 3 ⟩ \langle e_{3}\rangle ⟨ e 3 ⟩ for ε ≠ 0 \varepsilon\ne0 ε = 0 to the canonical case R 2 R_{2} R 2 from (12) is established by the biholomorphism x ~ 1 = ε x 1 − 1 2 x 2 \tilde{x}_{1}=\varepsilon x_{1}-\frac{1}{2x_{2}} x ~ 1 = ε x 1 − 2 x 2 1 , x ~ 2 = i 2 x 2 \tilde{x}_{2}=\frac{i}{2x_{2}} x ~ 2 = 2 x 2 i and the change of basis e ~ 1 = e 1 \tilde{e}_{1}=e_{1} e ~ 1 = e 1 , e ~ 2 = e 3 \tilde{e}_{2}=e_{3} e ~ 2 = e 3 , e ~ 3 = e 2 \tilde{e}_{3}=e_{2} e ~ 3 = e 2 .
The equivalence of the third parameterized realization R 2 ε R_{2}^{\varepsilon} R 2 ε for ε ≠ 0 \varepsilon\ne0 ε = 0 to the canonical case R 2 R_{2} R 2 from (12) is established by the biholomorphism x ~ 1 = x 1 \tilde{x}_{1}=x_{1} x ~ 1 = x 1 , x ~ 2 = i ε e x 2 \tilde{x}_{2}=\frac{i}{\sqrt{\varepsilon}}e^{x_{2}} x ~ 2 = ε i e x 2 and the change of basis e ~ 1 = e 1 \tilde{e}_{1}=e_{1} e ~ 1 = e 1 , e ~ 2 = 1 ε e 3 \tilde{e}_{2}=\frac{1}{\varepsilon}e_{3} e ~ 2 = ε 1 e 3 , e ~ 3 = e 2 \tilde{e}_{3}=e_{2} e ~ 3 = e 2 .
To reduce the limit realization obtained from the subalgebra ⟨ e 2 − e 1 ⟩ \langle e_{2}-e_{1}\rangle ⟨ e 2 − e 1 ⟩ to the canonical form R 3 \mathfrak{R}_{3} R 3 of the algebra g 3.4 − 1 \mathfrak{g}^{-1}_{3.4} g 3.4 − 1 from (13), the biholomorphism x ~ 1 = x 1 \tilde{x}_{1}=x_{1} x ~ 1 = x 1 , x ~ 2 = e 2 x 2 \tilde{x}_{2}=e^{2x_{2}} x ~ 2 = e 2 x 2 suffices.
Remark 9. The last two ε \varepsilon ε -realizations of the algebra s l ( 2 , C ) \mathfrak{sl}(2,\mathbb{C}) sl ( 2 , C ) in Table 1 are equivalent for every ε ≠ 0 \varepsilon\ne0 ε = 0 , yet in the limit they yield two distinct faithful and inequivalent realizations of the algebra g 3.4 − 1 \mathfrak{g}^{-1}_{3.4} g 3.4 − 1 . They correspond to the same contraction matrix, but are constructed from different one-dimensional subalgebras.
Table 1. Realizations corresponding to the contraction U 1 ( ε ) : s l ( 2 , C ) → g 3.4 − 1 U_{1}(\varepsilon):\mathfrak{sl}(2,\mathbb{C})\to\mathfrak{g}^{-1}_{3.4} U 1 ( ε ) : sl ( 2 , C ) → g 3.4 − 1 .
Table 2. Realizations corresponding to the contraction U ~ 2 ( ε ) : g 3.4 − 1 → g 3.1 \widetilde{U}_2(\varepsilon):\mathfrak{g}_{3.4}^{-1}\to\mathfrak{g}_{3.1} U 2 ( ε ) : g 3.4 − 1 → g 3.1 .
The equivalence between the parameterized realization R 1 ε \mathcal{R}_1^\varepsilon R 1 ε (ε ≠ 0 \varepsilon\ne0 ε = 0 ) and the canonical realization R 1 \mathcal{R}_1 R 1 is established by the biholomorphism x ~ 1 = ε x 1 + x 2 \tilde{x}_1=\varepsilon x_1+x_2 x ~ 1 = ε x 1 + x 2 , x ~ 2 = x 2 − ε x 1 \tilde{x}_2=x_2-\varepsilon x_1 x ~ 2 = x 2 − ε x 1 , x ~ 3 = ε x 3 \tilde{x}_3=\varepsilon x_3 x ~ 3 = ε x 3 and the change of basis e ~ 1 = 1 2 ( e 2 + 1 ε e 1 ) \tilde{e}_1=\frac{1}{2}(e_2+\frac{1}{\varepsilon}e_1) e ~ 1 = 2 1 ( e 2 + ε 1 e 1 ) , e ~ 2 = 1 2 ( e 2 − 1 ε e 1 ) \tilde{e}_2=\frac{1}{2}(e_2-\frac{1}{\varepsilon}e_1) e ~ 2 = 2 1 ( e 2 − ε 1 e 1 ) , e ~ 3 = 1 ε e 3 \tilde{e}_3=\frac{1}{\varepsilon}e_3 e ~ 3 = ε 1 e 3 .
The equivalence between both realizations R 2 ε \mathcal{R}_2^\varepsilon R 2 ε and the canonical realization R 2 \mathcal{R}_2 R 2 is established by the biholomorphism x ~ 1 = ε x 1 + x 2 + α ε \tilde{x}_1=\varepsilon x_1+x_2+\frac{\alpha}{\varepsilon} x ~ 1 = ε x 1 + x 2 + ε α , x ~ 2 = x 2 − ε x 1 − α ε \tilde{x}_2=x_2-\varepsilon x_1-\frac{\alpha}{\varepsilon} x ~ 2 = x 2 − ε x 1 − ε α and the same change of basis as in the previous case.
Remark 10. To obtain the intransitive realization R 2 = ⟨ ∂ 1 , ∂ 2 , x 2 ∂ 1 + x 3 ∂ 2 ⟩ \mathcal{R}_2=\langle\partial_1,\partial_2,x_2\partial_1+x_3\partial_2\rangle R 2 = ⟨ ∂ 1 , ∂ 2 , x 2 ∂ 1 + x 3 ∂ 2 ⟩ from the list (14), we interpret the continuous parameter α \alpha α as the third coordinate variable x 3 x_3 x 3 , which is an invariant of this realization constructed on a two-dimensional manifold.
As in Remark 9, the last two ε \varepsilon ε -realizations are equivalent to each other, yet in the limit they yield inequivalent realizations.
In all three rows of Table 3 , establishing the equivalences for the realizations in the third column uses the change of basis e ~ 1 = e 2 \tilde{e}_1=e_2 e ~ 1 = e 2 , e ~ 2 = 1 2 ε e 1 \tilde{e}_2=\frac{1}{2\varepsilon}e_1 e ~ 2 = 2 ε 1 e 1 , e ~ 3 = 1 ε e 3 \tilde{e}_3=\frac{1}{\varepsilon}e_3 e ~ 3 = ε 1 e 3 together with nonsingular coordinate transformations. The equivalence of the parameterized realization in the first row to the canonical realization R 4 \mathcal{R}_4 R 4 is established by the biholomorphism x ~ 1 = x 2 e 2 ε x 1 \tilde{x}_1=x_2e^{2\varepsilon x_1} x ~ 1 = x 2 e 2 ε x 1 , x ~ 2 = e 2 ε x 1 \tilde{x}_2=e^{2\varepsilon x_1} x ~ 2 = e 2 ε x 1 , x ~ 3 = ε x 3 \tilde{x}_3=\varepsilon x_3 x ~ 3 = ε x 3 and the basis change e ~ 1 = e 2 \tilde{e}_1=e_2 e ~ 1 = e 2 , e ~ 2 = 1 2 ε e 1 \tilde{e}_2=\frac{1}{2\varepsilon}e_1 e ~ 2 = 2 ε 1 e 1 , e ~ 3 = 1 ε e 3 \tilde{e}_3=\frac{1}{\varepsilon}e_3 e ~ 3 = ε 1 e 3 .
Table 3. Realizations corresponding to the contraction U 3 ( ε ) : s l ( 2 , C ) → g 3.1 U_3(\varepsilon):\mathfrak{sl}(2,\mathbb{C})\to\mathfrak{g}_{3.1} U 3 ( ε ) : sl ( 2 , C ) → g 3.1 .
The parameterized realization R 2 ε R^{\varepsilon}_{2} R 2 ε is equivalent to the canonical realization R 2 R_{2} R 2 after the biholomorphic change of variables x ~ 1 = x 2 e 2 ε x 1 \tilde{x}_{1}=x_{2}e^{2\varepsilon x_{1}} x ~ 1 = x 2 e 2 ε x 1 , x ~ 2 = − ( α ε ) − 1 / 2 e − 2 ε x 1 \tilde{x}_{2}=-\left(\frac{\alpha}{\varepsilon}\right)^{-1/2}e^{-2\varepsilon x_{1}} x ~ 2 = − ( ε α ) − 1/2 e − 2 ε x 1 .
The equivalence R 3 ε ∼ R 3 R^{\varepsilon}_{3}\sim R_{3} R 3 ε ∼ R 3 is established by the biholomorphism x ~ 1 = x 2 e 2 ε x 1 \tilde{x}_{1}=x_{2}e^{2\varepsilon x_{1}} x ~ 1 = x 2 e 2 ε x 1 , x ~ 2 = e 2 ε x 1 \tilde{x}_{2}=e^{2\varepsilon x_{1}} x ~ 2 = e 2 ε x 1 .
Remark 11. In the case of the contraction s l ( 2 , C ) → g 3.1 \mathfrak{sl}(2,\mathbb{C})\to\mathfrak{g}_{3.1} sl ( 2 , C ) → g 3.1 , no splitting of families of equivalent realizations into inequivalent ones occurs in the limit; namely, the parameterized realizations of s l ( 2 , C ) \mathfrak{sl}(2,\mathbb{C}) sl ( 2 , C ) in Table 3 constructed from the subalgebras ⟨ e 2 + α e 3 ⟩ \langle e_{2}+\alpha e_{3}\rangle ⟨ e 2 + α e 3 ⟩ and ⟨ e 2 ⟩ \langle e_{2}\rangle ⟨ e 2 ⟩ are inequivalent, and in the limit they likewise pass into inequivalent canonical realizations of the algebra g 3.1 \mathfrak{g}_{3.1} g 3.1 .
Differential invariants and invariant systems
The problem of describing differential equations invariant under a given Lie algebra reduces to finding the differential invariants of its realizations. On regular domains, invariant systems can be specified by relations among differential invariants of the appropriate order.
The classical theory of differential invariants originates in the works of Lie [18 , 19 ] and Tresse [37 , 38 ] ; in this section, we rely on the systematic exposition of equivalence and invariants in Olver’s monograph [28 ] and on the method of moving coframes proposed by Fels and Olver [5 , 6 ] . For transitions between different realizations, multiple dependent functions, and multiple independent variables, see [8 , 40 ] . A generalization of Lie’s classical theorem on the basis of differential invariants of one-parameter local transformation groups to an arbitrary number of independent and dependent variables, as well as a method for constructing a complete set of functionally independent invariants via a single quadrature and differentiation using a known universal invariant, is given in [30 ] , which also considers applications of first-order differential invariants to Riccati-type systems.
Consider the case of one independent variable t t t and two dependent variables x , y x,y x , y . Suppose a realization of a Lie algebra g \mathfrak{g} g is given by holomorphic basis vector fields in a neighborhood of a three-dimensional complex manifold M M M :
R ( e i ) = ξ i ( t , x , y ) ∂ t + η i ( t , x , y ) ∂ x + ζ i ( t , x , y ) ∂ y , i = 1 , … , n . (14) R(e_i)=\xi_i(t,x,y)\partial_t+\eta_i(t,x,y)\partial_x+\zeta_i(t,x,y)\partial_y,\qquad i=1,\ldots,n. \tag*{(14)} R ( e i ) = ξ i ( t , x , y ) ∂ t + η i ( t , x , y ) ∂ x + ζ i ( t , x , y ) ∂ y , i = 1 , … , n . (14)
For brevity in this section, we denote the realizations of the basis elements and their prolongations by e i e_i e i and e i ( k ) e_i^{(k)} e i ( k ) instead of R ( e i ) R(e_i) R ( e i ) and R ( k ) ( e i ) R^{(k)}(e_i) R ( k ) ( e i ) . The local action of the algebra g \mathfrak{g} g on the jet space M ( k ) M^{(k)} M ( k ) is defined by the k k k -th prolonged operators
e i ( k ) = e i + ∑ l = 1 k ( η i l ( t , x ( l ) , y ( l ) ) ∂ x ( l ) + ζ i l ( t , x ( l ) , y ( l ) ) ∂ y ( l ) ) , (15) e_i^{(k)}=e_i+\sum_{l=1}^{k}\left(\eta_i^l(t,x_{(l)},y_{(l)})\partial_{x_{(l)}}+\zeta_i^l(t,x_{(l)},y_{(l)})\partial_{y_{(l)}}\right), \tag*{(15)} e i ( k ) = e i + l = 1 ∑ k ( η i l ( t , x ( l ) , y ( l ) ) ∂ x ( l ) + ζ i l ( t , x ( l ) , y ( l ) ) ∂ y ( l ) ) , (15)
where x ( l ) = ( x , x ˙ , … , x ( l ) ) x_{(l)}=(x,\dot{x},\ldots,x^{(l)}) x ( l ) = ( x , x ˙ , … , x ( l ) ) , y ( l ) = ( y , y ˙ , … , y ( l ) ) y_{(l)}=(y,\dot{y},\ldots,y^{(l)}) y ( l ) = ( y , y ˙ , … , y ( l ) ) , and the coefficients η i l , ζ i l \eta_i^l,\zeta_i^l η i l , ζ i l are computed via the standard Lie–Ovsiannikov formulas [29 ] using the total derivative operator D t D_t D t .
Definition 12. A differential invariant of order k k k of the algebra g \mathfrak{g} g is a holomorphic function I ( t , x ( k ) , y ( k ) ) I(t,x_{(k)},y_{(k)}) I ( t , x ( k ) , y ( k ) ) satisfying the determining system e i ( k ) I = 0 e_i^{(k)}I=0 e i ( k ) I = 0 , i = 1 , … , n i=1,\ldots,n i = 1 , … , n .
Analogously to the rank of a realization, one can consider the ranks of prolonged realizations. Let r k r_k r k denote the maximal rank of the matrix formed by the coefficients of the prolonged operators e i ( k ) e_i^{(k)} e i ( k ) ; henceforth, we consider domains on which the corresponding rank is constant.
For the faithful local holomorphic realizations considered here, the nondecreasing sequence of maximal ranks { r k } \{r_k\} { r k } stabilizes at the value n n n . Hence, there exists ν = min { k ∈ N 0 ∣ r k = n } \nu=\min\{k\in\mathbb{N}_0\mid r_k=n\} ν = min { k ∈ N 0 ∣ r k = n } such that r ν = ⋯ = r ν + 1 = n r_\nu=\cdots=r_{\nu+1}=n r ν = ⋯ = r ν + 1 = n .
The number of functionally independent invariants of order k k k on a regular domain is given by the formula d k = 2 k + 3 − r k d_k=2k+3-r_k d k = 2 k + 3 − r k .
A maximal set I k I_k I k of functionally independent invariants of order at most k k k forms a universal differential invariant of order k k k .
On an appropriate regular domain, a finite set of differential invariants together with an invariant differentiation operator locally generates higher-order invariants by means of functional operations and differentiation. For a single independent variable, such an operator has the form δ = λ D t \delta= \lambda D_t δ = λ D t .
The coefficient function λ ( t , x ( ν ) , y ( ν ) ) \lambda(t,x^{(\nu)},y^{(\nu)}) λ ( t , x ( ν ) , y ( ν ) ) is determined from the commutation condition [ e i ( ∞ ) , δ ] = 0 [e_i^{(\infty)},\delta]=0 [ e i ( ∞ ) , δ ] = 0 :
e i ( ν ) ( λ ) − λ D t ( ξ i ) = 0 , i = 1 , … , n . e_i^{(\nu)}(\lambda)-\lambda D_t(\xi_i)=0,\qquad i=1,\ldots,n. e i ( ν ) ( λ ) − λ D t ( ξ i ) = 0 , i = 1 , … , n .
Once a complete set of functionally independent second-order differential invariants I 1 , … , I d I_1,\ldots,I_d I 1 , … , I d is found, a regular invariant system of two second-order ordinary differential equations can be locally specified by the relations
Φ 1 ( I 1 , … , I d ) = 0 , Φ 2 ( I 1 , … , I d ) = 0. \Phi_1(I_1,\ldots,I_d)=0,\qquad\Phi_2(I_1,\ldots,I_d)=0. Φ 1 ( I 1 , … , I d ) = 0 , Φ 2 ( I 1 , … , I d ) = 0.
The solvability of such a system for the highest-order derivatives (normal form) is determined by the ranks of the prolongations.
Proposition 13. A Lie algebra of vector fields g \mathfrak{g} g admits a local normal regular invariant system
x ¨ = F ( t , x , y , x ˙ , y ˙ ) , y ¨ = G ( t , x , y , x ˙ , y ˙ ) \ddot{x}=F(t,x,y,\dot{x},\dot{y}),\qquad\ddot{y}=G(t,x,y,\dot{x},\dot{y}) x ¨ = F ( t , x , y , x ˙ , y ˙ ) , y ¨ = G ( t , x , y , x ˙ , y ˙ ) if and only if r 1 = r 2 r_1=r_2 r 1 = r 2 on the corresponding regular domain.
In addition to the regular invariant systems mentioned above, there may also exist singular invariant systems described using Lie determinants, though we do not consider such cases in the present work.
Parameterized invariants of a generating realization of the algebra s l ( 2 , C ) \mathfrak{sl}(2,\mathbb{C}) sl ( 2 , C )
Consider an example of constructing an invariant system of the form (1) for the first ε \varepsilon ε -realization of the algebra s l ( 2 , C ) \mathfrak{sl}(2,\mathbb{C}) sl ( 2 , C ) in Table 1 , corresponding to the zero subalgebra:
∂ 1 , ε x 1 2 ∂ 1 + ( 1 − 2 ε x 1 x 2 ) ∂ 2 + 2 ε x 1 ∂ 3 , x 1 ∂ 1 − x 2 ∂ 2 + ∂ 3 . \partial_1,\quad\varepsilon x_1^2\partial_1+(1-2\varepsilon x_1x_2)\partial_2+2\varepsilon x_1\partial_3,\quad x_1\partial_1-x_2\partial_2+\partial_3. ∂ 1 , ε x 1 2 ∂ 1 + ( 1 − 2 ε x 1 x 2 ) ∂ 2 + 2 ε x 1 ∂ 3 , x 1 ∂ 1 − x 2 ∂ 2 + ∂ 3 .
To simplify the computations, we apply the change of coordinates
x ~ 1 = x 1 , x ~ 2 = e x 3 , x ~ 3 = ε x 2 e x 3 . (16) \widetilde{x}_1=x_1,\qquad\widetilde{x}_2=e^{x_3},\qquad\widetilde{x}_3=\varepsilon x_2e^{x_3}. \tag*{(16)} x 1 = x 1 , x 2 = e x 3 , x 3 = ε x 2 e x 3 . (16)
Its Jacobian equals
det ∂ ( x ~ 1 , x ~ 2 , x ~ 3 ) ∂ ( x 1 , x 2 , x 3 ) = − ε e 2 x 3 . \det\frac{\partial(\widetilde{x}_1,\widetilde{x}_2,\widetilde{x}_3)}{\partial(x_1,x_2,x_3)}=-\varepsilon e^{2x_3}. det ∂ ( x 1 , x 2 , x 3 ) ∂ ( x 1 , x 2 , x 3 ) = − ε e 2 x 3 .
Hence, for ε ≠ 0 \varepsilon\ne0 ε = 0 , this is a local biholomorphism. It is not globally injective on C 3 \mathbb{C}^3 C 3 , since the exponential function is not globally injective. To bring the vector fields to canonical form, the change of coordinates is accompanied by the change of basis
e ~ 1 = e 1 , e ~ 2 = e 3 , e ~ 3 = 1 ε e 2 . \widetilde{e}_1=e_1,\qquad\widetilde{e}_2=e_3,\qquad\widetilde{e}_3=\frac{1}{\varepsilon}e_2. e 1 = e 1 , e 2 = e 3 , e 3 = ε 1 e 2 .
As a result, we obtain the canonical case R 1 R_1 R 1 from (12):
∂ x ~ 1 , x ~ 1 ∂ x ~ 1 + x ~ 2 ∂ x ~ 2 , x ~ 1 2 ∂ x ~ 1 + 2 x ~ 1 x ~ 2 ∂ x ~ 2 + x ~ 2 ∂ x ~ 3 . \partial_{\widetilde{x}_1},\qquad\widetilde{x}_1\partial_{\widetilde{x}_1}+\widetilde{x}_2\partial_{\widetilde{x}_2},\qquad\widetilde{x}_1^2\partial_{\widetilde{x}_1}+2\widetilde{x}_1\widetilde{x}_2\partial_{\widetilde{x}_2}+\widetilde{x}_2\partial_{\widetilde{x}_3}. ∂ x 1 , x 1 ∂ x 1 + x 2 ∂ x 2 , x 1 2 ∂ x 1 + 2 x 1 x 2 ∂ x 2 + x 2 ∂ x 3 .
Fix the independent variable t t t and the two dependent variables as t : = x ~ 1 t := \tilde{x}_1 t := x ~ 1 , x : = x ~ 2 x := \tilde{x}_2 x := x ~ 2 , y : = x ~ 3 y := \tilde{x}_3 y := x ~ 3 . Throughout this example, we consider the regular domain x ≠ 0 x \ne0 x = 0 . Denote the canonical generators by
X 1 = ∂ t , X 2 = t ∂ t + x ∂ x , X 3 = t 2 ∂ t + 2 t x ∂ x + x 2 ∂ y . X_1 = \partial_t,\qquad X_2 = t\partial_t + x\partial_x,\qquad X_3 = t^2\partial_t + 2tx\partial_x + x^2\partial_y. X 1 = ∂ t , X 2 = t ∂ t + x ∂ x , X 3 = t 2 ∂ t + 2 t x ∂ x + x 2 ∂ y .
For this realization, the rank stabilizes already at the zeroth prolongation: on the domain x ≠ 0 x \ne0 x = 0 , we have r 0 = r 1 = r 2 = 3 r_0 = r_1 = r_2 = 3 r 0 = r 1 = r 2 = 3 .
Constructing the first prolongations of the operators and solving the determining system yields two absolute first-order differential invariants:
I 1 = y − 1 2 x ˙ , J 1 = x ˙ y − y ˙ x − y 2 . I_1 = y - \frac{1}{2}\dot{x},\qquad J_1 = \dot{x}y - \dot{y}x - y^2. I 1 = y − 2 1 x ˙ , J 1 = x ˙ y − y ˙ x − y 2 .
One directly verifies that
pr ( 1 ) X i ( I 1 ) = 0 , pr ( 1 ) X i ( J 1 ) = 0 , i = 1 , 2 , 3. \operatorname{pr}^{(1)}X_i(I_1)=0,\qquad\operatorname{pr}^{(1)}X_i(J_1)=0,\qquad i=1,2,3. pr ( 1 ) X i ( I 1 ) = 0 , pr ( 1 ) X i ( J 1 ) = 0 , i = 1 , 2 , 3.
Moreover,
det ∂ ( I 1 , J 1 ) ∂ ( x ˙ , y ˙ ) = x 2 ≠ 0. \det\frac{\partial(I_1,J_1)}{\partial(\dot{x},\dot{y})}=\frac{x}{2}\ne0. det ∂ ( x ˙ , y ˙ ) ∂ ( I 1 , J 1 ) = 2 x = 0.
Since d 1 = 5 − r 1 = 2 d_1=5-r_1=2 d 1 = 5 − r 1 = 2 , the invariants I 1 , J 1 I_1,J_1 I 1 , J 1 form a complete local functionally independent set on the first-order jet space.
For this realization, an invariant differentiation operator is δ = x D t \delta=xD_t δ = x D t . Indeed,
X 1 ( x ) = 0 , X 2 ( x ) = x , X 3 ( x ) = 2 t x . X_1(x)=0,\qquad X_2(x)=x,\qquad X_3(x)=2tx. X 1 ( x ) = 0 , X 2 ( x ) = x , X 3 ( x ) = 2 t x .
Thus x x x is not an absolute invariant: on the domain x ≠ 0 x\ne0 x = 0 , it is a relative invariant with multipliers ( 0 , 1 , 2 t ) (0,1,2t) ( 0 , 1 , 2 t ) . These multipliers define a local cocycle rather than a constant character of the group S L ( 2 , C ) \mathrm{SL}(2,\mathbb{C}) SL ( 2 , C ) . Since for the coefficients ξ i \xi_i ξ i of ∂ t \partial_t ∂ t we have
X i ( x ) = x D t ( ξ i ) , i = 1 , 2 , 3 , X_i(x)=xD_t(\xi_i),\qquad i=1,2,3, X i ( x ) = x D t ( ξ i ) , i = 1 , 2 , 3 ,
it follows that [ pr ( ∞ ) X i , x D t ] = 0 [\operatorname{pr}^{(\infty)}X_i,xD_t]=0 [ pr ( ∞ ) X i , x D t ] = 0 . At x = 0 x=0 x = 0 , the operator x D t xD_t x D t degenerates, so the condition x ≠ 0 x\ne0 x = 0 is essential.
Applying δ \delta δ to I 1 I_1 I 1 and J 1 J_1 J 1 produces second-order absolute invariants:
I 2 = x ( y ˙ − 1 2 x ¨ ) , J 2 = x ( x ¨ y − x y ¨ − 2 y ˙ y ) , I_2=x\left(\dot{y}-\frac{1}{2}\ddot{x}\right),\qquad J_2=x\left(\ddot{x}y-x\ddot{y}-2\dot{y}y\right), I 2 = x ( y ˙ − 2 1 x ¨ ) , J 2 = x ( x ¨ y − x y ¨ − 2 y ˙ y ) ,
for which
det ∂ ( I 2 , J 2 ) ∂ ( x ¨ , y ¨ ) = x 3 2 ≠ 0. \det\frac{\partial(I_2,J_2)}{\partial(\ddot{x},\ddot{y})}=\frac{x^3}{2}\ne0. det ∂ ( x ¨ , y ¨ ) ∂ ( I 2 , J 2 ) = 2 x 3 = 0.
Hence I 1 , J 1 , I 2 , J 2 I_1,J_1,I_2,J_2 I 1 , J 1 , I 2 , J 2 are functionally independent. Since d 2 = 7 − r 2 = 4 d_2=7-r_2=4 d 2 = 7 − r 2 = 4 , they form a complete local set of differential invariants of order at most two.
We also note that for this realization, higher-order invariants are locally generated from I 1 , J 1 I_1,J_1 I 1 , J 1 by repeated application of δ \delta δ and functional operations: at each subsequent order, the two new derivatives δ k − 1 I 1 \delta^{k-1}I_1 δ k − 1 I 1 and δ k − 1 J 1 \delta^{k-1}J_1 δ k − 1 J 1 are independent with respect to the highest-order derivatives for x ≠ 0 x\ne0 x = 0 .
Since the ranks of the first and second prolongations coincide for the realization R 1 R_1 R 1 , the corresponding regular invariant system of Newtonian type can be written in normal form. Consider first the absolute invariant equations
I 2 = F ( I 1 , J 1 ) , J 2 = G ( I 1 , J 1 ) , (17) I_2=F(I_1,J_1),\qquad J_2=G(I_1,J_1), \tag*{(17)} I 2 = F ( I 1 , J 1 ) , J 2 = G ( I 1 , J 1 ) , (17)
where F F F and G G G are arbitrary local holomorphic functions. Solving for the highest-order derivatives yields the system
{ x ¨ = 2 y ˙ − 2 x F ( I 1 , J 1 ) , y ¨ = − 2 y x 2 F ( I 1 , J 1 ) − 1 x 2 G ( I 1 , J 1 ) , (18) \begin{cases}
\ddot{x}=2\dot{y}-\dfrac{2}{x}F(I_1,J_1),\\
\ddot{y}=-\dfrac{2y}{x^2}F(I_1,J_1)-\dfrac{1}{x^2}G(I_1,J_1), \tag*{(18)}
\end{cases} ⎩ ⎨ ⎧ x ¨ = 2 y ˙ − x 2 F ( I 1 , J 1 ) , y ¨ = − x 2 2 y F ( I 1 , J 1 ) − x 2 1 G ( I 1 , J 1 ) , (18)
which is defined on the domain x ≠ 0 x\ne0 x = 0 .
To clarify the nature of its invariance, denote the left-hand sides of system (18) by
H 1 = x ¨ − 2 y ˙ + 2 F x , H 2 = y ¨ + 2 y F x 2 + G x 2 . H_1=\ddot{x}-2\dot{y}+\dfrac{2F}{x},\qquad H_2=\ddot{y}+\dfrac{2yF}{x^2}+\dfrac{G}{x^2}. H 1 = x ¨ − 2 y ˙ + x 2 F , H 2 = y ¨ + x 2 2 y F + x 2 G .
Then
I 2 − F = − x 2 H 1 , J 2 − G = x y H 1 − x 2 H 2 . I_2-F=-\dfrac{x}{2}H_1,\qquad J_2-G=xyH_1-x^2H_2. I 2 − F = − 2 x H 1 , J 2 − G = x y H 1 − x 2 H 2 .
The determinant of the transition matrix between the pairs of residuals ( H 1 , H 2 ) (H_1,H_2) ( H 1 , H 2 ) and ( I 2 − F , J 2 − G ) (I_2-F,J_2-G) ( I 2 − F , J 2 − G ) equals x 3 / 2 x^3/2 x 3 /2 . Thus, on the domain x ≠ 0 x\ne0 x = 0 , they define the same invariant set, although H 1 H_1 H 1 and H 2 H_2 H 2 themselves need not be individual absolute invariants.
In the real case, the resulting class of systems reproduces the corresponding known class from [41 ] . For comparison, we use the invariants given there:
ω 1 = 2 y − x ˙ , ω 2 = 4 x y ˙ − x ˙ 2 . \omega_1=2y-\dot{x},\qquad\omega_2=4x\dot{y}-\dot{x}^2. ω 1 = 2 y − x ˙ , ω 2 = 4 x y ˙ − x ˙ 2 .
The relation between the two sets of invariants is
I 1 = ω 1 2 , J 1 = − ω 1 2 + ω 2 4 . I_1=\dfrac{\omega_1}{2},\qquad J_1=-\dfrac{\omega_1^2+\omega_2}{4}. I 1 = 2 ω 1 , J 1 = − 4 ω 1 2 + ω 2 .
In particular, in the normalization of [41 ] , the corresponding system can be written as
{ x ¨ = 2 y ˙ + 1 x F 26 ( ω 1 , ω 2 ) , y ¨ = y x 2 F 26 ( ω 1 , ω 2 ) + 1 x 2 G 26 ( ω 1 , ω 2 ) . \begin{cases}
\ddot{x}=2\dot{y}+\dfrac{1}{x}F_{26}(\omega_1,\omega_2),\\
\ddot{y}=\dfrac{y}{x^2}F_{26}(\omega_1,\omega_2)+\dfrac{1}{x^2}G_{26}(\omega_1,\omega_2).
\end{cases} ⎩ ⎨ ⎧ x ¨ = 2 y ˙ + x 1 F 26 ( ω 1 , ω 2 ) , y ¨ = x 2 y F 26 ( ω 1 , ω 2 ) + x 2 1 G 26 ( ω 1 , ω 2 ) .
It is equivalent to system (18) upon relabeling the arbitrary functions via
F 26 ( ω 1 , ω 2 ) = − 2 F ( ω 1 2 , − ω 1 2 + ω 2 4 ) , F_{26}(\omega_1,\omega_2)=-2F\left(\dfrac{\omega_1}{2},-\dfrac{\omega_1^2+\omega_2}{4}\right), F 26 ( ω 1 , ω 2 ) = − 2 F ( 2 ω 1 , − 4 ω 1 2 + ω 2 ) ,
G 26 ( ω 1 , ω 2 ) = − G ( ω 1 2 , − ω 1 2 + ω 2 4 ) . G_{26}(\omega_1,\omega_2)=-G\left(\dfrac{\omega_1}{2},-\dfrac{\omega_1^2+\omega_2}{4}\right). G 26 ( ω 1 , ω 2 ) = − G ( 2 ω 1 , − 4 ω 1 2 + ω 2 ) .
Thus, the construction of the canonical invariant system (18) reproduces the known result [41 ] ; its parameterization and limit transition are considered below.
Construction of parameterized invariant systems
To construct an ε \varepsilon ε -parameterized invariant system, we apply the transformation locally inverse to (16) , written in terms of the variables t = x ~ 1 t=\tilde{x}_1 t = x ~ 1 , x = x ~ 2 x=\tilde{x}_2 x = x ~ 2 , y = x ~ 3 y=\tilde{x}_3 y = x ~ 3 and the variables T : = x 1 T:=x_1 T := x 1 (independent variable), X : = x 2 X:=x_2 X := x 2 and Y : = x 3 Y:=x_3 Y := x 3 (dependent variables):
T = t , X = y ε x , Y = ln x . (19) T=t,\qquad X=\dfrac{y}{\varepsilon x},\qquad Y=\ln x. \tag*{(19)} T = t , X = ε x y , Y = ln x . (19)
Here ln x \ln x ln x denotes a chosen local holomorphic branch of the logarithm on the domain x ≠ 0 x\ne0 x = 0 . In the real setting, this substitution corresponds to the domain x > 0 x>0 x > 0 . The transformation is defined only for ε ≠ 0 \varepsilon\ne0 ε = 0 ; therefore, the limit transition will be performed after rewriting the system in the common coordinates ( T , X , Y ) (T, X, Y) ( T , X , Y ) . Local biholomorphic equivalence of realizations for ε ≠ 0 \varepsilon\ne0 ε = 0 does not by itself establish equivalence at ε = 0 \varepsilon= 0 ε = 0 .
Using x = e Y x=e^Y x = e Y and y = ε e Y X y=\varepsilon e^Y X y = ε e Y X , and denoting X ′ = d X d T X'=\frac{dX}{dT} X ′ = d T d X , Y ′ = d Y d T Y'=\frac{dY}{dT} Y ′ = d T d Y , X ′ ′ = d 2 X d T 2 X''=\frac{d^2X}{dT^2} X ′′ = d T 2 d 2 X , Y ′ ′ = d 2 Y d T 2 Y''=\frac{d^2Y}{dT^2} Y ′′ = d T 2 d 2 Y , since T = t T=t T = t , derivatives with respect to t t t coincide with derivatives with respect to T T T , and we obtain
x ˙ = e Y Y ′ , x ¨ = e Y ( Y ′ ′ + Y ′ 2 ) , y ˙ = ε e Y ( X Y ′ + X ′ ) , y ¨ = ε e Y ( X ( Y ′ ′ + Y ′ 2 ) + 2 Y ′ X ′ + X ′ ′ ) . \dot{x}=e^Y Y', \qquad\ddot{x}=e^Y(Y''+Y'^2), \qquad\dot{y}=\varepsilon e^Y(XY'+X'), \qquad\ddot{y}=\varepsilon e^Y\left(X(Y''+Y'^2)+2Y'X'+X''\right). x ˙ = e Y Y ′ , x ¨ = e Y ( Y ′′ + Y ′2 ) , y ˙ = ε e Y ( X Y ′ + X ′ ) , y ¨ = ε e Y ( X ( Y ′′ + Y ′2 ) + 2 Y ′ X ′ + X ′′ ) .
Let I ε I_\varepsilon I ε and J ε J_\varepsilon J ε denote the expressions for the invariants I 1 I_1 I 1 and J 1 J_1 J 1 from the preceding subsection after passing to the coordinates ( T , X , Y ) (T,X,Y) ( T , X , Y ) . They take the form
I ε = e Y ( ε X − 1 2 Y ′ ) , J ε = − ε e 2 Y ( X ′ + ε X 2 ) . I_\varepsilon=e^Y\left(\varepsilon X-\frac{1}{2}Y'\right), \qquad J_\varepsilon=-\varepsilon e^{2Y}\left(X'+\varepsilon X^2\right). I ε = e Y ( εX − 2 1 Y ′ ) , J ε = − ε e 2 Y ( X ′ + ε X 2 ) .
For each fixed ε ≠ 0 \varepsilon\ne0 ε = 0 , the functions I ε I_\varepsilon I ε and J ε J_\varepsilon J ε are absolute differential invariants of the parameterized realization. However, J ε → 0 J_\varepsilon\to0 J ε → 0 as ε → 0 \varepsilon\to0 ε → 0 , so the limit of the pair ( I ε , J ε ) (I_\varepsilon,J_\varepsilon) ( I ε , J ε ) does not provide two functionally independent invariants. To obtain a nontrivial limit, consider the normalized invariants
A ε = − 2 I ε = e Y ( Y ′ − 2 ε X ) , B ε = − J ε ε = e 2 Y ( X ′ + ε X 2 ) . A_\varepsilon=-2I_\varepsilon=e^Y(Y'-2\varepsilon X), \qquad B_\varepsilon=-\frac{J_\varepsilon}{\varepsilon}=e^{2Y}(X'+\varepsilon X^2). A ε = − 2 I ε = e Y ( Y ′ − 2 εX ) , B ε = − ε J ε = e 2 Y ( X ′ + ε X 2 ) .
For each fixed ε ≠ 0 \varepsilon\ne0 ε = 0 , the factors − 2 -2 − 2 and − 1 / ε -1/\varepsilon − 1/ ε are constants with respect to the jet-space variables, so A ε A_\varepsilon A ε and B ε B_\varepsilon B ε remain absolute invariants.
In the common jet-space coordinates, taking ε → 0 \varepsilon\to0 ε → 0 yields
A 0 = e Y Y ′ , B 0 = e 2 Y X ′ . A_0=e^Y Y', \qquad B_0=e^{2Y}X'. A 0 = e Y Y ′ , B 0 = e 2 Y X ′ .
These functions are first-order absolute invariants of the limit realization whose generators are given below. They are functionally independent, since
det ∂ ( A 0 , B 0 ) ∂ ( Y ′ , X ′ ) = e 3 Y ≠ 0. \det\frac{\partial(A_0,B_0)}{\partial(Y',X')}=e^{3Y}\ne0. det ∂ ( Y ′ , X ′ ) ∂ ( A 0 , B 0 ) = e 3 Y = 0.
In the new coordinates, the invariant differentiation operator δ = x D t \delta=xD_t δ = x D t has the form δ ε = e Y D T \delta_\varepsilon=e^Y D_T δ ε = e Y D T ; in particular, for the limit realization we have δ 0 = e Y D T \delta_0=e^Y D_T δ 0 = e Y D T .
In this example, by a limit transition, or contraction, of differential invariants we mean the limit of a suitably normalized family of invariants in common coordinates on the jet space, together with verification of the functional independence of its limit elements. Without additional structure, such a transition does not define a canonical contraction of an abstract algebra of invariants.
In the variables T T T , X X X , Y Y Y , system (18) for ε ≠ 0 \varepsilon\ne0 ε = 0 takes the form
Y ′ ′ = − Y ′ 2 + 2 ε ( X Y ′ + X ′ ) − 2 e − 2 Y F ( I ε , J ε ) , X ′ ′ = − 2 ε X ( X Y ′ + X ′ ) − 2 Y ′ X ′ − e − 3 Y ε G ( I ε , J ε ) . (20) \begin{aligned}
Y''&=-Y'^2+2\varepsilon(XY'+X')-2e^{-2Y}F(I_\varepsilon,J_\varepsilon),\\
X''&=-2\varepsilon X(XY'+X')-2Y'X'-\frac{e^{-3Y}}{\varepsilon}G(I_\varepsilon,J_\varepsilon). \tag*{(20)}
\end{aligned} Y ′′ X ′′ = − Y ′2 + 2 ε ( X Y ′ + X ′ ) − 2 e − 2 Y F ( I ε , J ε ) , = − 2 εX ( X Y ′ + X ′ ) − 2 Y ′ X ′ − ε e − 3 Y G ( I ε , J ε ) . (20)
Notice that in the parameterized system, the arbitrary function F F F enters only the equation for Y ′ ′ Y'' Y ′′ , while the function G G G enters only the equation for X ′ ′ X'' X ′′ .
In the second equation, the contraction parameter ε \varepsilon ε appears in the denominator, so the existence of the limit must be checked for the chosen function G G G . Suppose F ( I , J ) F(I,J) F ( I , J ) and G ( I , J ) G(I,J) G ( I , J ) are fixed (i.e., not explicitly dependent on ε \varepsilon ε ) holomorphic functions near the limit values of their arguments. A sufficient local condition for regularity is
G ( I , 0 ) = 0. G(I,0)=0. G ( I , 0 ) = 0.
Then locally G ( I , J ) = J H ( I , J ) G(I,J)=JH(I,J) G ( I , J ) = J H ( I , J ) for some holomorphic function H H H , and therefore
G ( I ε , J ε ) ε = − e 2 Y ( X ′ + ε X 2 ) H ( I ε , J ε ) \frac{G(I_\varepsilon,J_\varepsilon)}{\varepsilon}=-e^{2Y}\left(X'+\varepsilon X^2\right)H(I_\varepsilon,J_\varepsilon) ε G ( I ε , J ε ) = − e 2 Y ( X ′ + ε X 2 ) H ( I ε , J ε )
has a finite limit. In particular, this condition is satisfied for the fixed function G ( I , J ) = J G(I,J)=J G ( I , J ) = J .
One may also consider separately a family of functions G ε ( I , J ) G_\varepsilon(I,J) G ε ( I , J ) depending explicitly on the parameter. For example, if G ε = ε G ^ ε G_\varepsilon=\varepsilon\widehat{G}_\varepsilon G ε = ε G ε and the expression G ^ ε ( I ε , J ε ) \widehat{G}_\varepsilon(I_\varepsilon,J_\varepsilon) G ε ( I ε , J ε ) has a regular limit, then the corresponding term in the system also has a regular limit. This is a different formulation from choosing a fixed function G G G .
Choosing the fixed functions F ( I , J ) = I F(I,J)=I F ( I , J ) = I and G ( I , J ) = J G(I,J)=J G ( I , J ) = J , i.e., F ( I ε , J ε ) = I ε F(I_\varepsilon,J_\varepsilon)=I_\varepsilon F ( I ε , J ε ) = I ε and G ( I ε , J ε ) = J ε G(I_\varepsilon,J_\varepsilon)=J_\varepsilon G ( I ε , J ε ) = J ε , upon taking the limit ε → 0 \varepsilon\to0 ε → 0 in system (20) we obtain the system
{ Y ′ ′ = − Y ′ 2 + e − Y Y ′ , X ′ ′ = − 2 X ′ Y ′ + e − Y X ′ . (21) \begin{cases}
Y''=-Y'^2+e^{-Y}Y',\\
X''=-2X'Y'+e^{-Y}X'. \tag*{(21)}
\end{cases} { Y ′′ = − Y ′2 + e − Y Y ′ , X ′′ = − 2 X ′ Y ′ + e − Y X ′ . (21)
This is just one of the possible choices of functions and the corresponding limit system. The limit generators of the associated family of realizations have the form
Q 1 = ∂ T , Q 2 = ∂ X , Q 3 = T ∂ T − X ∂ X + ∂ Y . Q_1=\partial_T,\qquad Q_2=\partial_X,\qquad Q_3=T\partial_T-X\partial_X+\partial_Y. Q 1 = ∂ T , Q 2 = ∂ X , Q 3 = T ∂ T − X ∂ X + ∂ Y .
They form the realization R 1 \mathfrak{R}_1 R 1 of the algebra g 3.4 − 1 \mathfrak{g}_{3.4}^{-1} g 3.4 − 1 from the list (12) . Thus, in this example, a transition to g 3.4 − 1 \mathfrak{g}_{3.4}^{-1} g 3.4 − 1 rather than to g 3.1 \mathfrak{g}_{3.1} g 3.1 has been carried out.
The invariance of the limit system can be verified directly. Denote its left-hand sides by
L 1 = Y ′ ′ + Y ′ 2 − e − Y Y ′ , L 2 = X ′ ′ + 2 X ′ Y ′ − e − Y X ′ . L_1=Y''+Y'^2-e^{-Y}Y',\qquad L_2=X''+2X'Y'-e^{-Y}X'. L 1 = Y ′′ + Y ′2 − e − Y Y ′ , L 2 = X ′′ + 2 X ′ Y ′ − e − Y X ′ .
Then
Q 1 ( 2 ) L a = Q 2 ( 2 ) L a = 0 , a = 1 , 2 , Q_1^{(2)}L_a=Q_2^{(2)}L_a=0,\qquad a=1,2, Q 1 ( 2 ) L a = Q 2 ( 2 ) L a = 0 , a = 1 , 2 ,
Q 3 ( 2 ) L 1 = − 2 L 1 , Q 3 ( 2 ) L 2 = − 3 L 2 . Q_3^{(2)}L_1=-2L_1,\qquad Q_3^{(2)}L_2=-3L_2. Q 3 ( 2 ) L 1 = − 2 L 1 , Q 3 ( 2 ) L 2 = − 3 L 2 .
Hence, the equation manifold L 1 = L 2 = 0 L_1=L_2=0 L 1 = L 2 = 0 is invariant under the listed generators.
In the example under consideration, the parameterized generators in the coordinates ( T , X , Y ) (T,X,Y) ( T , X , Y ) are given by
Q 1 ε = ∂ T , Q 2 ε = ε T 2 ∂ T + ( 1 − 2 ε T X ) ∂ X + 2 ε T ∂ Y , Q 3 ε = T ∂ T − X ∂ X + ∂ Y . Q_1^\varepsilon=\partial_T,\qquad Q_2^\varepsilon=\varepsilon T^2\partial_T+(1-2\varepsilon TX)\partial_X+2\varepsilon T\partial_Y,\qquad Q_3^\varepsilon=T\partial_T-X\partial_X+\partial_Y. Q 1 ε = ∂ T , Q 2 ε = ε T 2 ∂ T + ( 1 − 2 εT X ) ∂ X + 2 εT ∂ Y , Q 3 ε = T ∂ T − X ∂ X + ∂ Y .
Their coefficients and the coefficients of their prolongations up to second order converge regularly to the corresponding coefficients of Q 1 , Q 2 , Q 3 Q_1,Q_2,Q_3 Q 1 , Q 2 , Q 3 . For F ( I , J ) = I F(I,J)=I F ( I , J ) = I and G ( I , J ) = J G(I,J)=J G ( I , J ) = J , the right-hand sides of the parameterized system also have a regular limit together with the derivatives required to verify the infinitesimal criterion. Thus, the limit transition conditions for this family are verified, and the direct check above further confirms the invariance of the limit system.
Concluding remarks
In this paper, we have investigated limit transitions between local realizations of complex Lie algebras and invariant systems of ordinary differential equations. The starting ingredients of the approach are the known chain of contractions of three-dimensional Lie algebras [24 ] , the correspondence between local transitive realizations and stationary subalgebras, the criterion ker R = core g ( h ) \ker R=\operatorname{core}_{\mathfrak{g}}(\mathfrak{h}) ker R = core g ( h ) for a transitive realization, and the construction of regular intransitive realizations by promoting parameters of a family of transitive realizations to additional coordinates. The definition of differential invariants and the criterion r 1 = r 2 r_{1}=r_{2} r 1 = r 2 for a regular normal invariant system were presented in [41 ] .
Using the algebra s l ( 2 , C ) \mathfrak{sl}(2,\mathbb{C}) sl ( 2 , C ) as an example, we constructed the chain of Inönü–Wigner contractions
s l ( 2 , C ) ⟶ g 3.4 − 1 ⟶ g 3.1 \mathfrak{sl}(2,\mathbb{C}) \longrightarrow\mathfrak{g}_{3.4}^{-1} \longrightarrow\mathfrak{g}_{3.1} sl ( 2 , C ) ⟶ g 3.4 − 1 ⟶ g 3.1
obtained a list of local realizations and their limit forms, and showed that the proposed modification of Shirokov’s method—consisting in formally lifting invariant parameters—enables the systematic construction of intransitive realizations. The entire algorithm reduces to matrix operations—exponentiation, inversion, and taking limits—without the need to solve differential equations.
Tables 1, 2, and 3 juxtapose the chosen subalgebras, parameterized vector fields, and their limit expressions. In the relevant cases, local changes of variables and bases are also provided for comparison with canonical realizations. The statement regarding faithfulness of a limit transitive realization applies provided the corresponding limit subalgebra exists and is compatible with the limit action.
Table 1 exhibits an example of splitting : two complex realizations that are locally equivalent for ε ≠ 0 \varepsilon\ne0 ε = 0 yield inequivalent faithful limit realizations of the algebra g 3.4 − 1 \mathfrak{g}_{3.4}^{-1} g 3.4 − 1 . For these limit realizations, the ranks of the action of the derived ideal differ, being 2 and 1, respectively. An example of class merging was not constructed separately in this work, and therefore merging is not claimed as a demonstrated result here.
For one generating realization of s l ( 2 , C ) \mathfrak{sl}(2,\mathbb{C}) sl ( 2 , C ) on the regular domain x ≠ 0 x \ne0 x = 0 , complete local sets of functionally independent first- and second-order absolute differential invariants and an invariant differentiation operator were computed. Using them, a second-order invariant system was written in normal form and its parameterized family was constructed.
In the example considered, the direct limit of one of the first-order invariants vanishes; after normalization, two functionally independent invariants of the limit action are obtained. For the choice F ( I , J ) = I F(I,J)=I F ( I , J ) = I , G ( I , J ) = J G(I,J)=J G ( I , J ) = J , a limit transition of the systems to a realization of the algebra g 3.4 − 1 \mathfrak{g}_{3.4}^{-1} g 3.4 − 1 was carried out. The existence of the limit depends on the behavior of the arbitrary functions and does not follow solely from the contraction of the realizations.
Future research should aim at completing the chains of transitions, classifying limit stabilizers, analyzing changes in faithfulness and rank, and establishing the functional completeness of contracted differential invariants. Particular attention is warranted for the study of singular strata, integrability and globalization conditions for local actions, compatibility of contractions with reduction, conditional and hidden symmetries, and the development of equivalence criteria for classes of systems.
Potential applications include constructing coupled dynamical models with two degrees of freedom (in particular, Ermakov systems, generalized oscillators, and central-force problems) and transferring exact or asymptotic solutions across controlled limits. Furthermore, the results obtained can serve as a foundation for symbolic automation of computations, searching for contraction-compatible reductions of partial differential equations, and analyzing limit regimes in Galilean and relativistic models.
Acknowledgments
The authors are grateful to Doctor of Physical and Mathematical Sciences V. M. Boyko and Candidate of Physical and Mathematical Sciences B. H. Feshchenko for helpful ideas and remarks during the preparation of this article. The authors also thank the anonymous referees for their careful and detailed analysis of the paper and for comments that helped improve it.
This work was supported by the Simons Foundation (SFI-PD-Ukraine-00014586, M.S., M.N.); M.N. also gratefully acknowledges financial support from the National Research Foundation of Ukraine (project no. 2025.07/0405).