Sub-Riemannian structures [1] are stratified in depth, i.e., with respect to the minimum order of Lie brackets required for generating a tangent space from basis vector fields. The complexity of sub-Riemannian (SR) structures grows substantially with increasing step. At present, SR structures of step at most three have been examined in detail [28]. Accordingly, a task of great interest is to systematically investigate SR structures of step 4. The study of the simplest of these structures, namely, a nilpotent SR structure with growth vector (2, 3, 5, 8) (see Example 2 below) was initiated in [911]. Below, we obtain a complete classification of nilpotent SR structures and distributions with growth vector (2, 3, 5, 6). It is shown that all such structures are quotient structures of a nilpotent SR structure with growth vector (2, 3, 5, 8).

1 SUB-RIEMANNIAN QUOTIENT STRUCTURES

In this paper, all Lie algebras are considered over the field \(\mathbb{R}\).

Definition 1. A nilpotent Lie algebra \(\mathfrak{g}\) is called a Carnot algebra if

(i) \(\mathfrak{g}\) is graded:

$$\mathfrak{g} = {{\mathfrak{g}}^{{(1)}}} \oplus \; \cdots \; \oplus {{\mathfrak{g}}^{{(s)}}},$$
$$[{{\mathfrak{g}}^{{(i)}}},{{\mathfrak{g}}^{{(j)}}}] \subset {{\mathfrak{g}}^{{(i + j)}}},\quad {{\mathfrak{g}}^{{(k)}}} = {\text{\{ }}0{\text{\} }}\quad {\text{for}}\quad k > s,$$
(1)

(ii) \(\mathfrak{g}\) is generated by the first component:

$${\text{Lie}}({{\mathfrak{g}}^{{(1)}}}) = \mathfrak{g}.$$
(2)

The corresponding connected simply connected Lie group is called a Carnot group.

Conditions (i) and (ii) are equivalent to the condition

$$[{{\mathfrak{g}}^{{(1)}}},{{\mathfrak{g}}^{{(i)}}}] = {{\mathfrak{g}}^{{(i + 1)}}},\quad {{\mathfrak{g}}^{{(k)}}} = {\text{\{ }}0{\text{\} }}\quad {\text{for}}\quad k > s.$$

Definition 2. The growth vector of a Carnot algebra \(\mathfrak{g}\) is defined as

$$({{n}_{1}},\; \ldots ,\;{{n}_{s}}),\quad {{n}_{j}} = \sum\limits_{i = 1}^j {\text{dim}{{\mathfrak{g}}^{{(i)}}}} .$$

Here, \(({{n}_{1}},\; \ldots ,\;{{n}_{s}})\) is the growth vector of a left-invariant distribution on the Lie group of the Lie algebra \(\mathfrak{g}\) generated by the subspace \({{\mathfrak{g}}^{{(1)}}} \subset \mathfrak{g}\).

Let \(M\) be a smooth manifold. A sub-Riemannian structure on M [1] is a pair \((\Delta ,g)\) consisting of a vector distribution \(\Delta \subset TM\) and a scalar product g in Δ.

Let G be a Lie group and \(\mathfrak{g}\) be its Lie algebra. A left-invariant SR structure on G consists of a left-invariant distribution on G and a left-invariant scalar product in the distribution. This structure is specified by a subspace \(\Delta \subset \mathfrak{g}\) and a scalar product g in \(\Delta \). In this case, we say that \((\Delta ,g)\) is an SR structure in the algebra \(\mathfrak{g}\).

Left-invariant SR structures on Carnot groups arise as nilpotent approximations of general SR structures on smooth manifolds [1].

Definition 3. Let \((\Delta ,g)\) be an SR structure in a Lie algebra \(\mathfrak{g}\), and let \(\mathfrak{i} \subset \mathfrak{g}\) be an ideal such that \(\Delta \cap \mathfrak{i}\) = {0}. Let \(\tilde {\mathfrak{g}} = \mathfrak{g}{\text{/}}\mathfrak{i}\) be a quotient algebra and \(\pi \): \(\mathfrak{g} \to \tilde {\mathfrak{g}}\) be a canonical projection. Define \(\tilde {\Delta } = \pi (\Delta )\) and \(\tilde {g}(\pi (X)\), π(Y)) = g(X, Y) for \(X,Y \in \Delta \). Then (\(\tilde {\Delta }\), \(\tilde {g}\)) is an SR structure in \(\tilde {\mathfrak{g}}\), which will be called a quotient structure of the SR structure \((\Delta ,g)\).

Example 1. Let \({{\mathfrak{g}}^{5}}\) be a free nilpotent Lie algebra of step 3 with two generators (Cartan algebra); this is the Carnot algebra with growth vector (2, 3, 5). There exists a basis \({{\mathfrak{g}}^{5}} = {\text{span}}({{X}_{1}},\; \ldots ,\;{{X}_{5}})\) in which the nonzero Lie brackets are

$$\left[ {{{X}_{1}},{{X}_{2}}} \right] = {{X}_{3}},\quad \left[ {{{X}_{1}},{{X}_{3}}} \right] = {{X}_{4}},\quad \left[ {{{X}_{2}},{{X}_{3}}} \right] = {{X}_{5}}.$$

Consider an SR structure \((\Delta ,g)\) in \({{\mathfrak{g}}^{5}}\) with an orthonormal frame \(({{X}_{1}},{{X}_{2}})\) [8]. Sequentially choosing the subspaces \(\mathbb{R}{{X}_{5}}\), \({\text{span}}({{X}_{4}},{{X}_{5}})\), and \({\text{span}}({{X}_{3}},{{X}_{4}},{{X}_{5}})\) as an ideal \(\mathfrak{i} \subset {{\mathfrak{g}}^{5}}\), we obtain SR quotient structures in the Engel algebra (growth vector (2, 3, 4)) [7], Heisenberg algebra (growth vector (2, 3)) [2], and the two-dimensional commutative algebra \({{\mathbb{R}}^{2}}\) (growth vector (2)).

Example 2. Let \({{\mathfrak{g}}^{8}}\) be a free nilpotent Lie algebra of step 4 with two generators; this is the Carnot algebra with growth vector (2, 3, 5, 8), which will be called a nilpotent (2, 3, 5, 8)-algebra. There exists a basis \({{\mathfrak{g}}^{8}} = {\text{span}}({{X}_{1}}, \ldots ,{{X}_{8}})\) in which the nonzero Lie brackets are given by

$$\left[ {{{X}_{1}},{{X}_{2}}} \right] = {{X}_{3}},\quad \left[ {{{X}_{1}},{{X}_{3}}} \right] = {{X}_{4}},\quad \left[ {{{X}_{2}},{{X}_{3}}} \right] = {{X}_{5}},$$
(3)
$$\begin{gathered} \left[ {{{X}_{1}},{{X}_{4}}} \right] = {{X}_{6}},\quad \left[ {{{X}_{1}},{{X}_{5}}} \right] = \left[ {{{X}_{2}},{{X}_{4}}} \right] = {{X}_{7}}, \\ \left[ {{{X}_{2}},{{X}_{5}}} \right] = {{X}_{8}}. \\ \end{gathered} $$
(4)

Consider an SR structure \((\Delta ,g)\) in \({{\mathfrak{g}}^{8}}\) with an orthonormal frame \(({{X}_{1}},{{X}_{2}})\) [911]. It is easy to see that this SR structure is unique, up to a Lie algebra automorphism, SR structure in \({{\mathfrak{g}}^{8}}\) of rank 2 satisfying the condition \({\text{Lie}}(\Delta ) = {{\mathfrak{g}}^{8}}\); we call it a nilpotent SR (2, 3, 5, 8)-structure. In what follows, we will use a dual basis in the Lie coalgebra \(({{\mathfrak{g}}^{8}}){\text{*}}\):

$${{\omega }_{1}},\; \ldots ,\;{{\omega }_{8}} \in ({{\mathfrak{g}}^{8}}){\text{*}},\quad {{\omega }_{i}}({{X}_{j}}) = {{\delta }_{{ij}}},\quad i,j = 1,\; \ldots ,\;8.$$

The goal of this paper is, given a structure \((\Delta ,g)\), to describe all SR quotient structures for two-dimensional ideals \(\mathfrak{i} \subset {{\mathfrak{g}}^{8}}\). These are exactly nilpotent SR structures with growth vector (2, 3, 5, 6).

It is easy to see that a two-dimensional subspace \(\mathfrak{i} \subset {{\mathfrak{g}}^{8}}\) is an ideal if and only if \(\mathfrak{i} \subset Z(\mathfrak{g})\) = span(X6, \({{X}_{7}},{{X}_{8}})\). Therefore, any quotient algebra \({{\mathfrak{g}}^{8}}{\text{/}}\mathfrak{i}\) by the two-dimensional ideal \(\mathfrak{i}\) is a Carnot algebra with growth vector (2, 3, 5, 6). Let us describe such algebras.

2 CARNOT ALGEBRAS WITH GROWTH VECTOR (2, 3, 5, 6)

Theorem 1. (i) In any Carnot algebra \(\mathfrak{g}\) with growth vector (2, 3, 5, 6), there is a basis \(\mathfrak{g} = {\text{span}}({{X}_{1}},\; \ldots ,\;{{X}_{6}})\) in which all nonzero Lie brackets have the form

$$\left[ {{{X}_{1}},{{X}_{2}}} \right] = {{X}_{3}},\quad \left[ {{{X}_{1}},{{X}_{3}}} \right] = {{X}_{4}},\quad \left[ {{{X}_{2}},{{X}_{3}}} \right] = {{X}_{5}},$$
(5)
$$\begin{gathered} \left[ {{{X}_{1}},{{X}_{4}}} \right] = \alpha {{X}_{6}},\quad \left[ {{{X}_{1}},{{X}_{5}}} \right] = \left[ {{{X}_{2}},{{X}_{4}}} \right] = \beta {{X}_{6}}, \\ \left[ {{{X}_{2}},{{X}_{5}}} \right] = \gamma {{X}_{6}}, \\ \end{gathered} $$
(6)
$$(\alpha ,\beta ,\gamma ) \in {{\mathbb{R}}^{3}}{{\backslash \{ }}0{\text{\} }}.$$

(ii) Any two Carnot algebras with growth vector (2, 3, 5, 6) that have proportional triples \((\alpha ,\beta ,\gamma )\) in (6) are isomorphic to each other. Accordingly, such algebras are denoted by \(\mathfrak{g}_{{\alpha :\beta :\gamma }}^{6}\), \((\alpha :\beta :\gamma ) \in \mathbb{R}{{P}^{2}}\).

(iii) The following isomorphism of Lie algebras holds:

$$\begin{gathered} \mathfrak{g}_{{\alpha :\beta :\gamma }}^{6} \cong {{\mathfrak{g}}^{8}}{\text{/}}(\ker\omega \cap Z({{\mathfrak{g}}^{8}})), \\ \omega = \alpha {{\omega }_{6}} + \beta {{\omega }_{7}} + \gamma {{\omega }_{8}} \ne 0. \\ \end{gathered} $$

Each basis in the algebra \(\mathfrak{g}_{{\alpha :\beta :\gamma }}^{6}\) with multiplication table (5), (6) is associated with a quadratic form Q(xy) = \(\alpha {{x}^{2}} + 2\beta xy + \gamma {{y}^{2}}\). Depending on the sign of its discriminant \({\mathbf{s}} = {\text{sgn}}(\alpha \gamma - {{\beta }^{2}})\) ∈ {0, ±1}, the algebra \(\mathfrak{g}_{{\alpha :\beta :\gamma }}^{6}\) is called

– parabolic if s = 0,

– elliptic if s = 1,

– hyperbolic if \({\mathbf{s}} = - 1\).

Remark 1. Depending on the number s = \({\text{sgn}}(\alpha \gamma \) – β2), the topology of the sets of triples \((\alpha :\beta :\gamma ) \in \mathbb{R}{{P}^{2}}\) is as follows:

$$\begin{gathered} {{A}_{P}} = \{ (\alpha :\beta :\gamma ) \in \mathbb{R}{{P}^{2}}|{\mathbf{s}} = 0\} , \\ {{A}_{E}} = \{ (\alpha :\beta :\gamma ) \in \mathbb{R}{{P}^{2}}|{\mathbf{s}} = 1\} , \\ {{A}_{H}} = \{ (\alpha :\beta :\gamma ) \in \mathbb{R}{{P}^{2}}|{\mathbf{s}} = - 1\} . \\ \end{gathered} $$

Topologically, AP is a circle, AE is an open disk, and \({{A}_{H}} = \mathbb{R}{{P}^{2}}{{\backslash }}({{A}_{P}} \cup {{A}_{E}})\) is a projective plane with a cutout hole, i.e., a Möbius strip.

Below, we give several examples of Carnot algebras with growth vector (2, 3, 5, 6), together with nonzero Lie brackets in the corresponding basis. Recall that the notation \({{N}_{{6,2, * }}}\) for these algebras was used in [12, 13]. A classification of nilpotent Lie algebras of dimension ≤7 was obtained in [12], and all Carnot algebras of dimension ≤7 were described in [13].

Example 3. The parabolic algebra \(\mathfrak{g}_{{1:0:0}}^{6} = {{N}_{{6,2,7}}}\) = span(X1, ..., X6),

$$\begin{gathered} \left[ {{{X}_{1}},{{X}_{2}}} \right] = {{X}_{3}},\quad \left[ {{{X}_{1}},{{X}_{3}}} \right] = {{X}_{4}}, \\ [{{X}_{2}},{{X}_{3}}] = {{X}_{5}},\quad \left[ {{{X}_{1}},{{X}_{4}}} \right] = {{X}_{6}}. \\ \end{gathered} $$

Example 4. The hyperbolic algebra \(\mathfrak{g}_{{1:0:( - 1)}}^{6} = {{N}_{{6,2,5}}}\) = span(X1, ..., X6),

$$\begin{gathered} \left[ {{{X}_{1}},{{X}_{2}}} \right] = {{X}_{3}},\quad \left[ {{{X}_{1}},{{X}_{3}}} \right] = {{X}_{4}}, \\ \left[ {{{X}_{2}},{{X}_{3}}} \right] = {{X}_{5}},\quad \left[ {{{X}_{1}},{{X}_{4}}} \right] = - \left[ {{{X}_{2}},{{X}_{5}}} \right] = {{X}_{6}}. \\ \end{gathered} $$

Example 5. The elliptic algebra \(\mathfrak{g}_{{1:0:1}}^{6} = {{N}_{{6,2,5a}}}\) = span(X1, ..., X6),

$$\begin{gathered} \left[ {{{X}_{1}},{{X}_{2}}} \right] = {{X}_{3}},\quad \left[ {{{X}_{1}},{{X}_{3}}} \right] = {{X}_{4}}, \\ \left[ {{{X}_{2}},{{X}_{3}}} \right] = {{X}_{5}},\quad \left[ {{{X}_{1}},{{X}_{4}}} \right] = \left[ {{{X}_{2}},{{X}_{5}}} \right] = {{X}_{6}}. \\ \end{gathered} $$

Theorem 2. (i) The algebras \(\mathfrak{g}_{{1:0:0}}^{6}\), \(\mathfrak{g}_{{1:0:( - 1)}}^{6}\), and \(\mathfrak{g}_{{1:0:1}}^{6}\) are pairwise nonisomorphic. Any algebra \(\mathfrak{g}_{{\alpha :\beta :\gamma }}^{6}\), (α : β : \(\gamma ) \in \mathbb{R}{{P}^{2}}\), is isomorphic to one of these algebras.

(ii) The number \({\mathbf{s}} = {\text{sgn}}(\alpha \gamma - {{\beta }^{2}}) \in {\text{\{ }}0,\; \pm {\kern 1pt} 1{\text{\} }}\) is an invariant of the Carnot algebra \(\mathfrak{g}\); s is called the signature of \(\mathfrak{g}\).

(ii) The algebras \(\mathfrak{g}_{{\alpha :\beta :\gamma }}^{6}\) are separated by the signature \({\mathbf{s}}\):

$$\begin{gathered} \mathfrak{g}_{{\alpha :\beta :\gamma }}^{6} \cong \mathfrak{g}_{{1:0:0}}^{6} \Leftrightarrow {\mathbf{s}} = 0 \\ (parabolic\;algebra), \\ \end{gathered} $$
(7)
$$\begin{gathered} \mathfrak{g}_{{\alpha :\beta :\gamma }}^{6} \cong \mathfrak{g}_{{1:0:( - 1)}}^{6} \Leftrightarrow {\mathbf{s}} = - 1 \\ (hyperbolic\;algebra), \\ \end{gathered} $$
(8)
$$\begin{gathered} \mathfrak{g}_{{\alpha :\beta :\gamma }}^{6} \cong \mathfrak{g}_{{1:0:1}}^{6} \Leftrightarrow {\mathbf{s}} = 1 \\ (elliptic\;algebra). \\ \end{gathered} $$
(9)

Thus, the signature \({\mathbf{s}} \in {\text{\{ }}0,\; \pm {\kern 1pt} 1{\text{\} }}\) is an invariant separating three classes of isomorphism of the algebras \(\mathfrak{g}_{{\alpha :\beta :\gamma }}^{6}\).

Remark 2. Item (i) in Theorem 2 was first proved in [13]. We proved it independently, together with an algorithm for reducing the multiplication table in the (2, 3, 5, 6)-algebra to the normal form in Examples 3–5.

To find a basis change in the algebra \(\mathfrak{g} = \mathfrak{g}_{{\alpha :\beta :\gamma }}^{6}\) that transforms the basis into one of the normal forms indicated in Examples 3–5, it suffices to reduce the quadratic form Q(x, y) = \(\alpha {{x}^{2}} + 2\beta xy + \gamma {{y}^{2}}\) to a sum of squares, to apply this change to the basis \(({{X}_{1}},{{X}_{2}})\) of the space \({{\mathfrak{g}}^{{(1)}}}\), and to normalize the vector X6 generating the space \({{\mathfrak{g}}^{{(4)}}}\).

3 NILPOTENT (2, 3, 5, 6)-DISTRIBUTIONS AND SUB-RIEMANNIAN STRUCTURES

Definition 4. Let \(\mathfrak{g}\) be a Carnot algebra with growth vector (2, 3, 5, 6). A nilpotent sub-Riemannian (2, 3, 5, 6)-structure is a sub-Riemannian structure \((\Delta ,g)\) in \(\mathfrak{g}\) satisfying the conditions

$$\text{dim}\Delta = 2,\quad {\text{Lie}}(\Delta ) = \mathfrak{g},$$
(10)

or, equivalently, the equality \(\Delta \oplus {{\mathfrak{g}}^{{(2)}}}\) \( \oplus \) \({{\mathfrak{g}}^{{(3)}}} \oplus {{\mathfrak{g}}^{{(4)}}}\) = \(\mathfrak{g}\). The corresponding plane \(\Delta \subset \mathfrak{g}\) is called a nilpotent (2, 3, 5, 6)-distribution in \(\mathfrak{g}\).

The vector (2, 3, 5, 6) is the growth vector of the left-invariant distribution on the Lie group G of the Lie algebra \(\mathfrak{g}\) defined by the plane \(\Delta \subset \mathfrak{g}\).

Theorem 3. Let \(\mathfrak{g}\) be a Carnot algebra with growth vector (2, 3, 5, 6) of signature \({\mathbf{s}} \in {\text{\{ }}0,\; \pm {\kern 1pt} 1{\text{\} }}\). For any nilpotent (2, 3, 5, 6)-distribution \(\Delta \subset \mathfrak{g}\), there exists a frame \(\Delta = {\text{span}}({{X}_{1}},{{X}_{2}})\) such that

$$\begin{gathered} \left[ {{{X}_{1}},{{X}_{2}}} \right] = {{X}_{3}},\quad \left[ {{{X}_{1}},{{X}_{3}}} \right] = {{X}_{4}},\quad \left[ {{{X}_{2}},{{X}_{3}}} \right] = {{X}_{5}}, \\ \left[ {{{X}_{1}},{{X}_{4}}} \right] = {{X}_{6}},\quad \left[ {{{X}_{1}},{{X}_{5}}} \right] = \left[ {{{X}_{2}},{{X}_{4}}} \right] = 0, \\ \left[ {{{X}_{2}},{{X}_{5}}} \right] = {\mathbf{s}}{{X}_{6}}. \\ \end{gathered} $$

Theorem 4. Let \(\mathfrak{g}\) be a Carnot algebra with growth vector (2, 3, 5, 6). For any nilpotent SR (2, 3, 5, 6)-structures in \(\mathfrak{g}\), there exists an orthonormal frame \(({{X}_{1}},{{X}_{2}})\) and a number \(\nu \in [ - 1,\;1]\) for which

$$\left[ {{{X}_{1}},{{X}_{2}}} \right] = {{X}_{3}},\quad \left[ {{{X}_{1}},{{X}_{3}}} \right] = {{X}_{4}},\quad \left[ {{{X}_{2}},{{X}_{3}}} \right] = {{X}_{5}},$$
(11)
$$\begin{gathered} \left[ {{{X}_{1}},{{X}_{4}}} \right] = {{X}_{6}},\quad \left[ {{{X}_{1}},{{X}_{5}}} \right] = \left[ {{{X}_{2}},{{X}_{4}}} \right] = 0, \\ \left[ {{{X}_{2}},{{X}_{5}}} \right] = \nu {{X}_{6}}. \\ \end{gathered} $$
(12)

The number \(\nu \in [ - 1,\;1]\) in (12) is called the canonical parameter of the SR structure (\(\Delta ,g)\). It is easy to see that ν is equal to the ratio of the smaller (in absolute value) eigenvalue of Q to the larger eigenvalue. Moreover, \({\mathbf{s}} = {\text{sgn}}\nu \).

Theorem 5. The canonical parameter \(\nu \in [ - 1,\;1]\) is an invariant of the nilpotent SR (2, 3, 5, 6)-structure.

Let \((\Delta ,g)\) be an SR structure on a manifold \(M\). The corresponding SR distance d [1] transforms M into a metric space. Recall that an isometry between metric spaces \((M,d)\) and \((\tilde {M},\tilde {d})\) is a mapping \(F\): \(M \to \tilde {M}\) such that

$$d(x,y) = \tilde {d}(F(x),F(y)),\quad x,y \in M.$$

The metric group [15] is a Lie group with a left-invariant distance inducing a manifold topology on this group. In particular, any Carnot group is a metric group. It was proved in [15] that

any isometry between metric groups is an analytic mapping;

any isometry between connected nilpotent metric groups is an affine mapping (i.e., the composition of a left shift and an isomorphism).

The following theorem provides a classification of left-invariant SR (2, 3, 5, 6)-structures up to isometries of corresponding Carnot groups.

Theorem 6. Let \(\mathfrak{g}\) and \(\tilde {\mathfrak{g}}\) be nilpotent Carnot (2, 3, 5, 6)-algebras, and let \(G\) and \(\tilde {G}\) be corresponding Carnot groups. Let \((\Delta ,g)\) and \((\tilde {\Delta },\tilde {g})\) be nilpotent SR (2, 3, 5, 6)-structures in g and \(\tilde {\mathfrak{g}}\), \(\nu \) and \(\tilde {\nu }\) be corresponding canonical parameters, and \(d\) and \(\tilde {d}\) be corresponding left-invariant SR metrics on \(G\) and \(\tilde {G}\).

The metric spaces \((G,d)\) and \((\tilde {G},\tilde {d})\) are isometric if and only if \(\nu = \tilde {\nu }\).

Theorem 7. Let \(\mathfrak{g} = {{\mathfrak{g}}^{8}}\) be the (2, 3, 5, 8)-algebra from Example 2 with a basis \(({{X}_{1}}, \ldots ,{{X}_{8}})\) according to (3), (4), and let (\(\Delta ,g)\) be an SR structure in \(\mathfrak{g}\) with orthonormal frame \(({{X}_{1}},{{X}_{2}})\). Let \(\mathfrak{i} = \ker\omega \cap Z(\mathfrak{g})\) with \(\omega = \alpha {{\omega }_{6}} + \beta {{\omega }_{7}} + \gamma {{\omega }_{8}} \ne 0\) be any 2-dimensional subspace of \(Z(\mathfrak{g})\). Then the SR quotient structure \((\tilde {\Delta },\tilde {g})\) is one of the SR (2, 3, 5, 6)-structures in \(\tilde {\mathfrak{g}} = \mathfrak{g}{\text{/}}\mathfrak{i}\):

(i) in the parabolic case s = 0, we have \(\tilde {\mathfrak{g}} \cong \mathfrak{g}_{{1:0:0}}^{6}\);

(ii) in the hyperbolic case \({\mathbf{s}} = - 1\), we have \(\tilde {\mathfrak{g}} \cong \mathfrak{g}_{{1:0:( - 1)}}^{6}\);

(iii) in the elliptic case s = 1, we have \(\tilde {\mathfrak{g}} \cong \mathfrak{g}_{{1:0:1}}^{6}\).

Conversely, any nilpotent (2, 3, 5, 6)-structure in each of these algebras is realized as a quotient structure of a nilpotent (2, 3, 5, 8)-structure \((\Delta ,g)\).

Remark 3. An SR structure in the elliptic algebra \(\mathfrak{g}_{{1:0:1}}^{6}\) with canonical parameter ν = 1 was considered in [14]. For this structure, it was proved that the vertical subsystem of the Hamiltonian system in the Pon-tryagin maximum principle [1] is Liouville integrable. This Hamiltonian system was integrated in [11].

4 CASIMIR FUNCTIONS AND SYMPLECTIC FOLIATIONS

Let us recall some basic concepts of symplectic geometry. Let \(\mathfrak{g}\) be a Lie algebra and \(\mathfrak{g}{\text{*}}\) be the Lie coalgebra (the dual of the space \(\mathfrak{g}\)). Any function \(h \in {{C}^{\infty }}(\mathfrak{g}*)\) is called a Hamiltonian. For any Hamiltonians f and g, the Poisson bracket is defined as

$$\left\{ {f,g} \right\}(\lambda ) = \left\langle {\lambda ,\left[ {d{{f}_{\lambda }},d{{g}_{\lambda }}} \right]} \right\rangle ,\quad \lambda \in \mathfrak{g}{\text{*}}.$$

The Casimir function (on a submanifold \(M \subset \mathfrak{g}{\text{*}}\)) is any Hamiltonian f commuting with all Hamiltonians in the sense of the Poisson bracket:

$$\begin{gathered} \left\{ {f,h} \right\}(\lambda ) = 0,\quad h \in {{C}^{\infty }}(\mathfrak{g}*), \\ \lambda \in \mathfrak{g}{\text{*}}\quad ({\text{accordingly}}\;\lambda \in M). \\ \end{gathered} $$

A symplectic foliation on \(\mathfrak{g}{\text{*}}\) is a partition of \(\mathfrak{g}{\text{*}}\) into symplectic leaves \(L(\lambda )\), which are orbits of the coadjoint action of the Lie group \(G\) of the Lie algebra \(\mathfrak{g}\) in \(\mathfrak{g}{\text{*}}\):

$$L(\lambda ) = \{ (Ad_{x}^{ * })(\lambda )|x \in G\} ,\quad \lambda \in \mathfrak{g}{\text{*}}.$$

The rank of the Poisson bracket {⋅, ⋅} at a point \(\lambda \in \mathfrak{g}{\text{*}}\) is defined as the dimension of the symplectic leaf passing through \(\lambda \) and is denoted by \({\text{rank}}(\lambda )\). The symplectic leaves and Casimir functions are invariants of the Hamiltonian vector field

$$\dot {\lambda } = \mathbf{h}(\lambda ),\quad \lambda \in \mathfrak{g}{\text{*}},$$

for any Hamiltonian h, so they are important for analyzing left-invariant Hamiltonians on the Lie group of the Lie algebra \(\mathfrak{g}\), in particular, for studying left-invariant optimal control problems on this Lie group.

For coalgebras \(\mathfrak{g}{\text{*}}\) of all Carnot algebras \(\mathfrak{g}\) with growth vector (2, 3, 5, 6), we describe their Casimir functions and symplectic foliations. In each of the algebras \(\mathfrak{g} = \mathfrak{g}_{{1:0:0}}^{6}\), \(\mathfrak{g}_{{1:0:( - 1)}}^{6}\), and \(\mathfrak{g}_{{1:0:1}}^{6}\), we use a canonical basis \(\mathfrak{g} = {\text{span}}({{X}_{1}},\; \ldots ,\;{{X}_{6}})\) according to Examples 3–5 and corresponding linear Hamiltonians h1, ..., h6 \( \in \mathfrak{g}{\text{**}}\), \({{h}_{i}}(\lambda ) = \left\langle {\lambda ,{{X}_{i}}} \right\rangle \). In each of the cases presented below, the symplectic leaf \(L(\lambda )\) is a connected component of common level surfaces of the Casimir functions on corresponding submanifolds in \(\mathfrak{g}{\text{*}}\).

4.1 Parabolic Algebra \(\mathfrak{g} = \mathfrak{g}_{{1:0:0}}^{6}\)

(1) If \({{h}_{5}}{{h}_{6}} \ne 0\), then \({\text{rank}}(\lambda ) = 4\).

Casimir functions: \({{h}_{5}}\), \({{h}_{6}}\).

(2) If \({{h}_{6}} = 0\) and \(h_{3}^{2} + h_{4}^{2} + h_{5}^{2} \ne 0\), then \({\text{rank}}(\lambda )\) = 2.

Casimir functions: \({{h}_{4}}\), \({{h}_{5}}\), \(\tfrac{{h_{3}^{2}}}{2} + {{h}_{1}}{{h}_{5}} - {{h}_{2}}{{h}_{4}}\).

(3) If \({{h}_{5}} = 0\) and \(h_{3}^{2} + h_{4}^{2} + h_{6}^{2} \ne 0\), then \({\text{rank}}(\lambda )\) = 2.

Casimir functions: \({{h}_{6}}\), \({{C}_{1}} = {{h}_{3}}{{h}_{6}} - \tfrac{{h_{4}^{2}}}{2}\), C2 = \({{h}_{2}}h_{6}^{2}\)\({{C}_{1}}{{h}_{4}} - \tfrac{{h_{4}^{3}}}{6}\).

(4) If \({{h}_{3}} = \; \cdots \; = {{h}_{6}} = 0\), then \({\text{rank}}(\lambda ) = 0\).

Casimir functions: \({{h}_{1}}\), \({{h}_{2}}\).

4.2 Hyperbolic Algebra \(\mathfrak{g} = \mathfrak{g}_{{1:0:( - 1)}}^{6}\)

(1) If \({{h}_{6}} \ne 0\), then \({\text{rank}}(\lambda ) = 4\).

Casimir functions: \({{h}_{6}}\), \({{h}_{3}}{{h}_{6}} + \frac{{h_{4}^{2} - h_{5}^{2}}}{2}\).

(2) If \({{h}_{6}} = 0\) and \(h_{3}^{2} + h_{4}^{2} + h_{5}^{2} \ne 0\), then \({\text{rank}}(\lambda )\) = 2.

Casimir functions: \({{h}_{4}}\), \({{h}_{5}}\), \(\tfrac{{h_{3}^{2}}}{2} + {{h}_{1}}{{h}_{5}} - {{h}_{2}}{{h}_{4}}\).

(3) If \({{h}_{3}} = \; \cdots \; = {{h}_{6}} = 0\), then \({\text{rank}}(\lambda ) = 0\).

Casimir functions: \({{h}_{1}}\), \({{h}_{2}}\).

4.3 Elliptic Algebra \(\mathfrak{g} = \mathfrak{g}_{{1:0:1}}^{6}\)

(1) If \({{h}_{6}} \ne 0\), then \({\text{rank}}(\lambda ) = 4\).

Casimir functions: \({{h}_{6}}\), \({{h}_{3}}{{h}_{6}} - \tfrac{{h_{4}^{2} + h_{5}^{2}}}{2}\).

(2) If \({{h}_{6}} = 0\) and \(h_{3}^{2} + h_{4}^{2} + h_{5}^{2} \ne 0\), then \({\text{rank}}(\lambda )\) = 2.

Casimir functions: \({{h}_{4}}\), \({{h}_{5}}\), \(\tfrac{{h_{3}^{2}}}{2} + {{h}_{1}}{{h}_{5}} - {{h}_{2}}{{h}_{4}}\).

(3) If \({{h}_{3}} = \; \cdots \; = {{h}_{6}} = 0\), then \({\text{rank}}(\lambda ) = 0\).

Casimir functions: \({{h}_{1}}\), \({{h}_{2}}\).

CONCLUSIONS

The following results were obtained in this paper.

We described the Carnot algebras with growth vector (2, 3, 5, 6), i.e., quotient algebras of a free nilpotent Carnot algebra \({{\mathfrak{g}}^{8}}\) of step 4 with two generators. Previously, it was known that there are three normal forms of such algebras: parabolic, hyperbolic, and elliptic. An invariant separating these algebras was found, namely, the signature \({\mathbf{s}} \in {\text{\{ }}0,\; \pm {\kern 1pt} 1{\text{\} }}\). Additionally, a change of basis transforming the multiplication table into one of three normal forms was described.

We studied the left-invariant sub-Riemannian structures with growth vector (2, 3, 5, 6), i.e., quotient structures of a (unique) sub-Riemannian structure with growth vector (2, 3, 5, 8) by the two-dimensional subspace of the center of the algebra \({{\mathfrak{g}}^{8}}\). A classification of (2, 3, 5, 6)-structures was obtained up to isometries: all of them are uniquely parametrized by a canonical parameter ν ∈ {0, ±1} such that \({\mathbf{s}} = {\text{sgn}}\nu \). A classification of left-invariant (2, 3, 5, 6)-distributions was obtained: in each of the (2, 3, 5, 6)-algebras (parabolic, hyperbolic, elliptic) there exists a unique, up to isomorphism, distribution of this type.

For each Carnot algebra \(\mathfrak{g}\) with growth vector (2, 3, 5, 6), the rank of the Poisson bracket, Casimir functions, and a symplectic foliation in the Lie coalgebra \(\mathfrak{g}{\text{*}}\) were calculated.