Full Text
UDC 513.838+519.4
MATHEMATICS
M. A. AKIVIS
ON CANONICAL DECOMPOSITIONS OF EQUATIONS OF A LOCAL ANALYTIC QUASIGROUP
(Presented by Academician A. D. Aleksandrov on 6 III 1969)
- Let \(X, Y, Z\) be three analytic manifolds of the same dimension \(r\). An analytic mapping \(f: X \times Y \to Z\) is called a local analytic quasigroup \(Q_r\) if, for \(a \in X\), \(b \in Y\), and \(c = f(a,b) \in Z\), the following conditions are satisfied:
a) For every neighborhood \(U_c\) of the point \(c\) there exist neighborhoods \(U_a\) and \(U_b\) of the points \(a\) and \(b\) such that, for any \(x \in U_a\), \(y \in U_b\), the function \(f(x,y)\) is defined and \(f(x,y) \in U_c\).
b) For every neighborhood \(U_a\) of the point \(a\) there exist neighborhoods \(U_b\) and \(U_c\) of the points \(b\) and \(c\) such that, for any \(y \in U_b\) and \(z \in U_c\), the equation \(f(x,y)=z\) has a unique solution with respect to \(x\), and \(x \in U_a\).
c) For every neighborhood \(U_b\) of the point \(b\) there exist neighborhoods \(U_a\) and \(U_c\) of the points \(a\) and \(c\) such that, for any \(x \in U_a\) and \(z \in U_c\), the equation \(f(x,y)=z\) has a unique solution with respect to \(y\), and \(y \in U_b\).
If the manifolds \(X, Y, Z\) coincide, \(X \equiv Y \equiv Z\), and there exists an element \(e\) such that \(f(x,e)=f(e,x)=x\), then the quasigroup \(Q_r\) becomes a local analytic loop \((^1)\). If, moreover, the associativity condition \(f[x,f(y,z)]=f[f(x,y),z]\) is satisfied on the quasigroup \(Q_r\), then it becomes a local \(r\)-parameter Lie group.
In the present paper canonical decompositions are obtained for the analytic function \(z=f(x,y)\) defining the quasigroup \(Q_r\), which pass into the Campbell–Hausdorff formula \((^2)\) when the quasigroup \(Q_r\) becomes a local Lie group. The constructed canonical decompositions are applied to the study of certain special classes of local analytic quasigroups.
- Let \(a\) and \(b\) be arbitrary points of the manifolds \(X\) and \(Y\), and let \(c=f(a,b)\) be a point belonging to the manifold \(Z\). Introduce local coordinates in neighborhoods \(U_a, U_b\), and \(U_c\) of the points \(a, b\), and \(c\). Then in these neighborhoods the mapping \(f: X \times Y \to Z\) is written in the form:
\[ z^i=f^i(x^j,y^k), \quad i,j,k=1,\ldots,r, \tag{1} \]
where \(x^i\) are the coordinates of the point \(x \in U_a\), \(y^i\) are the coordinates of the point \(y \in U_b\), \(z^i\) are the coordinates of the point \(z \in U_c\), and \(f^i\) are analytic functions.
Suppose that the points \(a\) and \(b\) of the manifolds \(X\) and \(Y\) are determined by the coordinates \(x^i=0\) and \(y^i=0\). Then in the neighborhoods \(U_a\) and \(U_b\) the right-hand sides of equations (1) can be represented in the form of convergent power series:
\[ z^i=\sum_{s=0}^{\infty}\underset{(s)}{\Lambda^i}(x^j,y^k), \tag{2} \]
where
\[ \underset{(s)}{\Lambda^i}(x^j,y^k) = \frac{1}{s!}\sum_{p=0}^{s} C_s^p \lambda^i_{j_1\ldots j_p\, r+1\, j_{p+1}\ldots r+j_s} x^{j_1}\ldots x^{j_p}y^{j_{p+1}}\ldots y^{j_s}. \tag{3} \]
homogeneous polynomials of degree \(s\) in \(x^j, y^k\). Their coefficients \(\lambda^i_{j_1\ldots j_p\, r+j_{p+1}\ldots r+j_s}\) are symmetric in the indices \(j_1\ldots j_p\) and \(j_{p+1}\ldots j_s\), and \(C_s^p\) are binomial coefficients.
The local coordinates of points of the neighborhoods \(U_a, U_b\), and \(U_c\) of the analytic manifolds \(X, Y, Z\) admit transformations of the form
\[
\tilde{x}^i=\varphi^i(x^j),\qquad
\tilde{y}^i=\psi^i(y^j),\qquad
\tilde{z}^i=\chi^i(z^j),
\tag{4}
\]
where \(\varphi^i,\psi^i,\chi^i\) are uniquely invertible analytic functions such that \(\varphi^i(0)=0,\ \psi^i(0)=0\). Such transformations are called isotopic coordinate transformations of the quasigroup \(Q_r\). These transformations make it possible to reduce the expansions (2) to a certain simplest form, namely:
Theorem 1. With the aid of transformations (4), the polynomials \(\Lambda^i_{(s)}(x^j,y^k)\) occurring in the expansions (2) can be reduced to the following canonical form:
\[
\Lambda^i_{(0)}=0,\qquad
\Lambda^i_{(1)}=x^i+y^i,\qquad
\Lambda^i_{(2)}=a^i_{jk}x^j y^k,
\]
\[
\Lambda^i_{(s)}=\frac{1}{s!}\sum_{p=1}^{s-1} C_s^p
a^i_{j_1\ldots j_p\, r+j_{p+1}\ldots r+j_s}
x^{j_1}\ldots x^{j_p}y^{j_{p+1}}\ldots y^{j_s},
\tag{5}
\]
where the coefficients of these polynomials satisfy the relations
\[
a^i_{(jk)}=0,\qquad b^i_{(j_1\ldots j_s)}=0,
\tag{6}
\]
where
\[
b^i_{(j_1\ldots j_s)}
=
\sum_{p=1}^{s-1} C_s^p\sigma^{\,s-p}
a^i_{j_1\ldots j_p\, r+j_{p+1}\ldots r+j_s},
\]
and \(\sigma\) is some fixed real number, \(\sigma\ne 0,-1\).
Expansions of equations (2) satisfying the conditions of this theorem will be called canonical expansions of the equations of the quasigroup \(Q_r\). Thus, the quasigroup \(Q_r\) has not one but an entire pencil of canonical expansions, depending on the parameter \(\sigma\). The variables \(x^i\) and \(y^i\) occurring in the canonical expansions of the quasigroup \(Q_r\) will be called its canonical parameters.
The uniqueness of the canonical expansions is confirmed by the following theorem:
Theorem 2. The canonical expansion of the equations of a local analytic quasigroup preserves its form if and only if the variables entering into it are transformed according to the formulas
\[
\tilde{x}^i=\alpha^i_j x^j,\qquad
\tilde{y}^i=\alpha^i_j y^j,\qquad
\tilde{z}^i=\alpha^i_j z^j.
\tag{7}
\]
In this case the coefficients \(a^i_{jk}\) and \(a^i_{j_1\ldots j_p\, r+j_{p+1}\ldots r+j_s}\) of the canonical expansion are transformed according to the tensor law.
The proof of Theorems 1 and 2 is carried out by the method of complete induction on the degree \(s\) of the polynomials \(\Lambda_{(s)}(x,y)\).
- Let us identify the points of the neighborhoods \(U_a, U_b\), and \(U_c\) that have identical coordinates, and denote by the letter \(e\) the point obtained by identifying the points \(a,b,c\). Then the local quasigroup \(Q_r\) will be defined in some neighborhood \(U\) of this point, and the binary quasigroup operation will be an analytic mapping \(f: U\times U\to U\). In this case the point \(e\) will correspond to the identity element of the quasigroup \(Q_r\), since by virtue of relations (5)
\[ f(x,e)=f(e,x)=x. \]
Consequently, the quasigroup \(Q_r\) becomes a local analytic loop. Moreover, by virtue of relations (5) and (6),
\[ f(x,\sigma x)=(1+\sigma)x, \tag{8} \]
where the latter relation is satisfied for one fixed value of the parameter \(\sigma \ne 0,-1\). We shall call condition (8) the normalization condition for the equations of the quasigroup.
- Let \(Q_1\) be a one-parameter quasigroup. By virtue of the first of conditions (6), its canonical decompositions are written in the form
\[ z=x+y+\sum_{s=3}^{\infty}\underset{(s)}{\Lambda}(x,y). \]
If we introduce the notation
\[ \underbrace{a_{1\ldots 1}}_{p}\underbrace{\,{}_{2\ldots 2}}_{q}=a_{pq}, \]
then the polynomials \(\underset{(s)}{\Lambda}(x,y)\), for \(s \geqslant 3\), take the form
\[ \underset{(s)}{\Lambda}(x,y)=\frac{1}{s!}\sum_{p=1}^{s-1} C_s^p a_{p,s-p}x^p y^{s-p}, \tag{9} \]
where
\[ \sum_{p=1}^{s-1} C_s^p \sigma^{s-p} a_{p,s-p}=0. \]
A direct verification shows that if the quasigroup \(Q_1\) becomes a one-parameter Lie group \(G_1\), then all its canonical decompositions take the form
\[ z=x+y. \]
Consequently, if the quasigroup \(Q_r\) becomes a Lie group, then the canonical parameters introduced on \(Q_r\) pass into the ordinary canonical parameters of the local Lie group \(G_r\). All canonical decompositions of the quasigroup \(Q_r\) then pass into the Campbell—Hausdorff formula for the Lie group \(G_r\).
- Consider on the quasigroup \(Q_r\) a one-parameter subquasigroup \(Q_1\). It is defined on \(Q_r\) by the equations:
\[ x^i=x^i(u), \qquad y^i=y^i(v), \qquad f^i[x^j(u),y^h(v)]=z^i(w), \tag{10} \]
where \(w=\varphi(u,v)\). From these relations the following theorem follows easily:
Theorem 3. In order that the first two of equations (10) define on the quasigroup \(Q_r\) a one-parameter subquasigroup \(Q_1\), it is necessary and sufficient that the relations
\[ \lambda \frac{\partial z^i}{\partial x^j}\frac{dx^j}{du}+\mu \frac{\partial z^i}{\partial y^j}\frac{dy^j}{dv}=0, \tag{11} \]
hold, where \(\lambda\) and \(\mu\) are analytic functions of \(u\) and \(v\).
Relations (11) are differential equations of one-parameter subquasigroups on \(Q_r\). These relations make it possible to prove the following theorem:
Theorem 4. In order that on the loop \(Q_r\), for each direction issuing from the point \(e\), there exist a one-parameter subloop \(Q_1\) tangent to this direction, it is necessary and sufficient that the coefficients of at least one of its canonical decompositions satisfy the relations:
\[ \underset{p}{b^i_{(j_1\ldots j_s)}}=\delta^i_{(j_1}\underset{p}{b}_{j_2\ldots j_s)}, \tag{12} \]
where
\[ \underset{p}{b^i_{j_1\ldots j_s}}=a^i_{j_1\ldots j_p\, r+j_{p+1}\ldots r+j_s}. \]
We note that if the condition of Theorem 4 is fulfilled, the coefficients of any of the canonical decompositions of the quasigroup \(Q_r\) will satisfy relations (12).
The three-web \((^3)\) corresponding to the quasigroup \(Q_r\) defined by Theorem 4 was considered by us in \((^4)\) under the name of a transversal-geodesic web.
- The following theorem answers the question of when the one-parameter subloops \(Q_1\) of the loop \(Q_r\) become groups.
Theorem 5. In order that on the loop \(Q_r\), for every direction issuing from the point \(e\), there exist a one-parameter subgroup \(G_1\) tangent to this direction, it is necessary and sufficient that the coefficients of at least one of its canonical decompositions satisfy the conditions
\[ \underset{p}{b}{}^{\,i}_{(j_1\ldots j_s)}=0. \tag{13} \]
The condition contained in Theorem 5 is equivalent to the monoassociativity \((^3)\) of the loop \(Q_r\). Therefore condition (13) is a necessary and sufficient condition for the monoassociativity of the loop \(Q_r\). Note that in \((^1)\) the existence of one-parameter subgroups of a local loop \(Q_r\) was proved only for alternative loops, which form a narrower class than monoassociative loops.
When condition (13) is fulfilled, for arbitrary \(x\) belonging to a sufficiently small neighborhood of the point \(e\) of the loop \(Q_r\), and for sufficiently small real \(\lambda\) and \(\mu\), the relation
\[ f(\lambda x,\mu x)=(\lambda+\mu)x \]
will hold, and this means that the normalization condition (8) is fulfilled on the quasigroup \(Q_r\) for every \(\sigma\ne 0,-1\). Hence it follows:
Theorem 6. In order that all the canonical decompositions of the equations of the quasigroup \(Q_r\) introduced in Theorem 1 coincide with one another, it is necessary and sufficient that this quasigroup be monoassociative.
Moscow Institute
of Steel and Alloys
Received
24 I 1969
CITED LITERATURE
\(^1\) A. I. Mal’tsev, Matem. sborn., 36 (78), No. 3 (1955).
\(^2\) E. B. Dynkin, DAN, 57, No. 4 (1947).
\(^3\) J. Aczel, Advances Math., 1, No. 3 (1965).
\(^4\) M. A. Akivis, Abstracts of Reports, III Baltic Geom. Conf., Palanga, 1968.