1. Introduction
Research Contribution
Accepted: 26 December 2024
The use of the conformal Levi-Civita transformation [15] , [5] , [14] to study planar integrable
mechanical billiards defined with the Hooke and Kepler problems has been first pointed out in [16] and extended in [22] .
In the first part of this note, we discuss some consequences of the K.S. (Kustaanheimo-Stiefel) transformation on integrable four-dimensional Hooke and integrable three-dimensional Kepler billiards. An \(n\)-dimensional mechanical billiard system is integrable if there
exist \(n\) first integrals of the underlying natural mechanical system that satisfy the following conditions: they are functionally independent, in involution, and they remain invariant under reflections at the reflection wall (c.f. [24] ). It is widely known that for the four-dimensional Hooke problem, a centered quadric reflection wall gives an integrable billiard system [6] , [4]
. We shall show that when this reflection wall is invariant under an \(S^{1}\)-symmetry of the K.S. transformation, then its image under the Hopf mapping is one of five special type of quadrics, with the Kepler center as a focus. This is consistent with the results of
[23] and provides a partial explanation of why we conjecture that only these quadrics appear in three-dimensional integrable
Kepler billiards. We can think that the restriction on the type of quadrics is forced by the \(S^{1}\)-invariance of the centered quadric reflection wall lying on the four-dimensional Hooke side. This generalizes the studies in the planar case [22] , [16] to the spatial case of Kepler billiards based on the Levi-Civita transformation. This way we obtain
The Kustaanheimo-Stiefel transformation has been extended to a transformation which simultaneously regularizes both double collisions in the spatial two-center problem first announced in a 1-page note of Stiefel-Waldvogel [18] , which generalized the transformation of Birkhoff used in the planar case. The thesis of Waldvogel [25] provided a much more extensive geometrical study of this transformation. In particular, the relation between this transformation and the Kustaanheimo-Stiefel transformation
has been clarified. Waldvogel later illustrated this theory again in [26] with the use of quaternions. In this article, we
provide a quaternionic formulation of this Birkhoff-Waldvogel transformation in the spatial case, largely inspired by the studies of Waldvogel as well as combining the symplectic viewpoint of [27] . We investigate in part an analogous theory of Levi-Civita planes in this setting, consisting of planes and spheres in the space of quaternions \(\mathbb {H} \cong \R ^{4}\) and a reduction of
this transformation to a dense open subset of \(\mathbb {IH} \cong \R ^{3}\), which already regularizes the double collisions without increasing the dimension of the space. With this we link integrable billiards on both sides, which illustrates some results in [23] with a different method.
We organize this article as follows: In Section 2 , we recall the theory of Kustaanheimo-Stiefel regularization, which largely follows [27] . Then we apply this transformation to link integrable mechanical billiards in Section 3 . The theory of Birkhoff-Waldvogel transformation and the corresponding link on integrable mechanical billiards are discussed in Section 4 .
In this section, we discuss the Kustaanheimo-Stiefel transformation. We follow the quaternionic formulation of [27] .
A quaternion is represented as
\[ z= z_0 + z_1 i+z_2 j+z_3 k, \quad z_0,z_1, z_2,z_3 \in \R \]
in which
\[i^{2}=j^{2}=k^{2}=-1, ij=-ji=k, jk=-kj=i, ki=-ik=j.\]
Addition and multiplication of quaternions are then naturally defined. With these operations, the quaternions form a non-commutative normed division algebra which we denote by \(\mathbb {H}\). For a quaternion \(z= z_0 + z_1 i+z_2 j+z_3 k\), its real part is
given by
\[Re(z)= z_0\]
and its imaginary part is given by
\[Im(z)= z_1 i+z_2 j+z_3 k.\]
Furthermore, the conjugation of \(z\) is defined as
\[\bar {z}=z_0 - z_1 i-z_2 j-z_3 k.\]
The norm of \(z\) is defined as \(|z|:=\sqrt {z \cdot \bar {z}}\).
We denote the set of purely imaginary quaternions by
\[\mathbb {IH}=\{z \in \mathbb {H} \mid Re(z) = 0 \}.\]
We identify \(\mathbb {H}\) with \(\R ^4\) and denote by \(\mathcal {S}^3\) the unit sphere
\[ \{ z \in \mathbb {H} \mid |z|^2= z_0^2 + z_1^2+z_2^2+z_3^2=1 \} \subset \mathbb {H}. \]
Also, we identify \(\mathbb {IH}\) with \(\R ^3\). The unit sphere \(\mathcal {S}^{2}\) therein is
\[ \{z \in \mathbb {IH} \mid |z|^2= z_1^2+z_2^2+z_3^2=1 \} \subset \mathbb {IH}. \]
To introduce the Kustaanheimo-Stiefel transformation, we first recall the Levi-Civita transformation [14]
\[ T^*(\C \setminus \{0 \}) \to T^*(\C \setminus \{0 \}), (z,w) \mapsto \left (q=z \cdot z, p=\dfrac {z}{2 |z|^{2}} \cdot w \right ). \]
It is well-known that this transformation is canonical, and transforms the planar Kepler problem into the planar Hooke problem after making a proper time reparametrization on an energy level. To see this, we start with the shifted Hamiltonian of the Kepler problem and
consider its zero-energy level:
\[ \frac {|p^2|}{2} + \frac {m}{|q|} {-f} = 0. \]
The Levi-Civita transformation pulls this system back to
\[ \frac {|w^2|}{8|z|^2} + \frac {m}{|z|^2} {-f} = 0. \]
We may now multiply this transformed Hamiltonian by \(|z|^2\), which only reparametrizes the flow on this energy-level. We obtain
\[ \frac {|w^2|}{8} + m {-f} |z|^2 = 0. \]
which is the restriction of the Hamiltonian of the planar Hooke problem \(\frac {|w^2|}{8} {-f}|z|^2\) on its \((-m)\)-energy hypersurface.
The whole construction is based on the complex square mapping
\[\C \setminus \{0\} \mapsto \C \setminus \{0\}, \qquad z \mapsto z^{2},\]
which is a \(2\)-to-\(1\) conformal mapping.
A generalization of the complex square mapping with quaternions is the following Hopf mapping
\[ \mathbb {H} \to \mathbb {IH}, \qquad z \mapsto \bar {z}i z. \]
Note that this mapping is well-defined, since
\[Re( \bar {z}i z) = 0, \forall z \in \mathbb {H}.\]
This mapping is “\(S^{1}\)-to-\(1\)”, namely the circle
\[\{\exp (i \theta )z\mid ,z \in \mathbb {H}\setminus \{0\} ,\,\, \theta \in {\R /2\pi \Z } \} \subset \mathbb {H}\]
is mapped under the Hopf mapping to the same point \(\bar {z} i z \in \mathbb {IH}\).
Moreover, this mapping restricts to a mapping \(\mathcal {S}^3 \to \mathcal {S}^2\). This is a mapping with \(S^{1}\)-fibres, and induces the non-trivial Hopf fibration
\[S^{1} \hookrightarrow \mathcal {S}^3 \to \mathcal {S}^2.\]
Associated to the Hopf mapping, the Kustaanheimo-Stiefel mapping is defined as
\[ T^*(\mathbb {H} \setminus \{ 0\}) \to \mathbb {IH} \times \mathbb {H}, (z,w) \mapsto \left (Q= \bar {z}i \cdot z, P= \frac { \bar {z}i }{2 |z|^2} \cdot w \right ). \]
The fibers of the mapping are the circle orbits of the \(S^{1}\)-Hamiltonian action
\[{\theta \cdot } (z,w) \mapsto (\exp (i \theta ) z, \exp (i \theta ) w)\]
on the cotangent bundle \(T^{*} \mathbb {H}\). The bilinear function
\[ BL(z,w):= Re(\bar {z} i w) \]
is the associated moment map.
We define
\[\Sigma :=\{(z, w) |BL(z,w) = 0 \} \subset T^*\mathbb {H} \cong \mathbb {H} \times \mathbb {H},\]
and
\[\Sigma ^1 = \Sigma \setminus \{ z=0 \}.\]
Both are invariant under this \(S^{1}\)-Hamiltonian action.
We define the restricted K.S. mapping as
\[KS:=K.S.|_{\Sigma _{1}}: \Sigma ^1 \to T^*(\mathbb {IH}\backslash \{ 0 \}).\]
For the following lemma from [27] , we present an alternative, simpler proof.
\begin{equation}
\label {eq: one-form 0} {KS^*Re(\bar {P} d Q) = Re(\bar {w} d z)|_{\Sigma ^1}.}
\end{equation}
To see ( 1 ), we compute
\begin{equation}
\label {eq: one-form-1} \begin{split} \bar {P} dQ = - \frac {\bar {w}i \bar {z}^{-1}}{2}((d\bar {z})i z + \bar {z} i dz) \\ = {(- \bar {w} i \bar {z}^{-1} (d \bar {z})iz+ \bar {w}dz)/2.} \end {split}
\end{equation}
\[BL(z,w)=Re(\bar {z}iw)=0\]
is equivalent to
\[Re(z^{-1}i w)=0.\]
Consequently, we also have
\[Re(\bar {w}i \bar {z}^{-1}) = 0.\]
This implies
\[ Re(\bar {w} i \bar {z}^{-1} (d \bar {z})iz) = Re(\bar {w} i \bar {z}^{-1} \cdot Im((d \bar {z})iz)). \]
Since
\[Im((d \bar {z})iz)= -Im( \bar {z} (-i)dz),\]
we have
\begin{equation}
\label {eq: one-form-2} Re(\bar {w} i \bar {z}^{-1} (d \bar {z})iz) = -Re(\bar {w} i \bar {z}^{-1} \cdot Im( \bar {z} (-i)dz)) =-Re(\bar {w} i \bar {z}^{-1} \bar {z} (-i)dz)= -Re(\bar {w} dz),
\end{equation}
\[Re(\bar {w}i \bar {z}^{-1}) = 0.\]
The assertion ( 1 ) is thus obtained by combining the equations ( 2 ) and ( 3 ).
On \(\Sigma ^{1}\), the orbits of the \(S^{1}\)-action mentioned above lie in the direction of the one-dimensional kernel distribution of the 2-form \(Re(d\bar {w}\wedge dz )\). By the theory of symplectic reduction, the 2-form \(Re(d\bar {w}\wedge dz )\) of
\(\Sigma ^1\) gives rise to the reduced symplectic form \(\omega _1\) on the quotient space \(V^1\) of \(\Sigma ^{1}\) by the \(S^{1}\)-action. Thus, the Kustaanheimo-Stiefel mapping induces a symplectomorphism
\[KS_{red}: (V_1, \omega _1) \to (T^*(\mathbb {IH}\backslash \{ 0 \}),Re(d \bar {P}\wedge dQ) ).\]
We have
\[KS=KS_{red} \circ \phi \]
in which \(\phi : \Sigma ^{1} \to V_{1}\) is the quotient map.
\[\frac {\|w\|^2}{8} {-f} \| z\|^2 +m\]
in \(\Sigma ^{1}\) is sent via \(KS\) to a zero-energy orbit of the three-dimensional Kepler Problem in \(T^{*} (\mathbb {IH}\backslash \{ 0 \})\) with Hamiltonian
\[\frac {\|P\|^2}{2} + \frac {m}{\| Q\|} {-f} .\]
after a proper time reparametrization.
\[\left (T^{*} \mathbb {H}, Re(d\bar {w}\wedge dz), H=\frac {\|w\|^2}{8} {-f} \| z\|^2 +m\right ).\]
This follows either from a direct verification, or alternatively from the invariance of \(H\) under the above mentioned (Hamiltonian) \(S^1\)-action. Consequently, the set \(\Sigma ^{1}\) is invariant under its flow.
We consider the restriction of this system on \(\Sigma ^{1}\). Any orbit of this restricted system descends to an orbit in the quotient system in \((V_{1}, \omega _{1}, {H_1})\) so that
\[\phi ^{*}{H_1}=H,\]
which is consequently sent to an orbit via \(KS_{red}\) in the system
\[(T^{*} \mathbb {IH}, Re(d\bar {p}\wedge d q), {K})\]
such that
\[KS_{red}^{*}\, {K}={H_1}.\]
Applying \(\phi ^{*}\) to both sides of this identity, we get
\[H=\phi ^{*} KS_{red}^{*} \, {K}=KS^{*}\, {K}.\]
From this we deduce
\[{K}=\frac {\|P\|^2 \|Q\|}{2} + m {-f} \|Q\|. \]
Now we restrict the system to \(\{{K}=0\}=\{H=0\}\). We observe that the restricted flow can now be time reparametrized (with factor \(\|Q\|^{-1}\)) into the restricted flow on the zero-energy hypersurface of the three-dimensional Kepler Hamiltonian
\[{\frac {\|P\|^2 }{2} + \frac {m}{\|Q\|} {-f} }.\]
The link between Kustaanheimo-Stiefel transformation and the Levi-Civita transformation is given by the Levi-Civita planes. These are planes in \(\mathbb {H}\) generated by two unit quaternions \(v_{1}, v_{2}\) such that \(v_{1} \neq \pm v_{2}\) and satisfy
\[BL(v_{1}, v_{2})=0.\]
The key property of such a plane is that its image is a plane in \(\mathbb {IH}\) and in relevant basis the restriction of the Hopf mapping is equivalent to the complex square mapping
\[\C \to \C , z \mapsto z^{2}.\]
Therefore \(K.S.\) is restricted to \(L. C. \) on the tangent bundle of such a plane. We proceed with the details.
\[ \mathbb {H} \to \mathbb {IH}, \quad z \mapsto \bar {z}iz \]
sends a Levi-Civita plane to a plane passing through the origin in \(\mathbb {IH}\). On the other hand, any plane in \(\mathbb {IH}\) passing through the origin is the image of a \(\mathcal {S}^1\)-family of Levi-Civita planes.
\[|v_{1}|=|v_{2}|=1, \, BL(v_1, v_2)=0\, \hbox {and} \, \langle v_1, v_2 \rangle =0.\]
Then, we have
\[ \bar {v}_1 i v_1 = -\bar {v}_2 i v_2, \]
which follows from the computation
\begin{equation}
\label {eq: LC-plane-1} \begin{split} 2\bar {v}_1 i v_1 + 2\bar {v}_2 i v_2 &= (\bar {v}_2v_1 + \bar {v}_1v_2)(\bar {v}_1i v_2+ \bar {v}_2i v_1) \\ &= 2 \langle v_1, v_2 \rangle (\bar {v}_1i v_2+ \bar {v}_2i v_1)\\ &=0. \end
{split}
\end{equation}
\[BL(v_1, v_2)=0\]
is equivalent to
\[\bar {v}_1i v_2 - \bar {v}_2 i v_1 =0.\]
Thus
\[ (\bar {v}_1 v_2 - \bar {v}_2 v_1)(\bar {v}_1i v_2 - \bar {v}_2 i v_1)= 0 \]
which is equivalent to
\[ \bar {v}_1 i v_1 + \bar {v}_2 i v_2 = \bar {v}_1 v_2 \bar {v}_1 i v_2 + \bar {v}_2 v_1 \bar {v}_2 i v_1. \]
Thus the quaternion \(v_1 + v_2\) in \(V\) is sent via the Hopf mapping to the quaternion
\[ \bar {v}_1i v_1 + \bar {v}_1 i v_2 + \bar {v}_2 i v_1 + \bar {v}_1 i v_1 = 2 \bar {v}_1 i v_2. \]
As a vector in \(\mathbb {IH}\), it is linearly independent of the vector \(\bar {v}_1 i v_1\), which follows from \(\bar {v}_{1} i \neq 0\) and the linear independency of \(v_{1}\) and \(v_{2}\).
As a consequence, the image of \(V\) is the plane passing through the origin, linearly spanned by \(\bar {v}_1 i v_1\) and \(\bar {v}_1 i v_2\).
On the other hand, for any unit quaternion \(w \in \mathbb {IH}\), there exists a \(S^1\)-family of unit vectors \(\{e^{i \theta }v\}\) in \(\mathbb {H}\) whose image under the Hopf map is \(w\). Take a plane \(W\) in \(\mathbb {IH}\) passing through the origin
spanned by two linearly independent unit vectors \(w_1\) and \(w_2\). We can choose the pre-images of \(v_1\) and \(v_2\) in \(\mathbb {H}\) of \(w_1\) and \(w_2\) to be such that \(BL(v_1, v_2) = 0\). Indeed, for \(\bar {v}_1 i v_2 = z_0 + z_1 i + z_2 j +z_4
k\), we have
\[ Re(e^{i \theta }\bar {v}_1 i v_2 ) = z_0\cos \theta - z_1 \sin \theta , \]
thus we can take \(e^{i {\theta _{1}}} v_1\) such that \(z_0\cos {\theta _{1}}- z_1 \sin {\theta _{1}}=0\) in the place of \(v_1\). Thus we get the family of Levi-Civita planes \(\hbox {span} \{e^{i \theta }v_1, e^{i \theta }v_2\}, \theta \in \R /{2
\pi \Z }\) which are sent to \(W\).
\[ T^* \C \to T^* \C , \quad (z,w) \mapsto (z^2, {\dfrac {z}{2 |z|^{2}} \cdot w}) \]
which is the Levi-Civita transformation.
\[ ((a^2- b^2)\bar {v}_1 i v_1 + 2 ab \bar {v}_1 i v_2, \frac {(ac-bd)\bar {v}_1 i v_1 + (ad +bc)\bar {v}_1 i v_2}{2(a^2+b^2)}). \]
From the orthogonality of \(v_1\) and \(v_2\), we obtain
\[ \langle \bar {v}_1 i v_1, \bar {v}_1 i v_2 \rangle = \frac {\bar {v}_1 v_2 + \bar {v}_2 v_1 }{2} = \langle v_1 , v_2 \rangle = 0. \]
Hence we just need to identify \(\bar {v}_1 i v_1\) and \(\bar {v}_1 i v_2\) with the standard orthogonal basis of \(\C \). The conclusion follows after both \(V\) and its image have been identified to \(\C \).
We extend the correspondence shown above to the corresponding billiard systems. This generalizes the correspondence of Hooke and Kepler billiards in the plane [16] , [22] to the spatial (Kepler) case.
A centered quadric in \(\mathbb {H} \cong \R ^{4}\) is called \(S^{1}\)-invariant, if it is invariant under the \(S^{1}\)-action
\[S^{1} \curvearrowright \mathbb {H}, \quad \theta \cdot z \mapsto \exp (i \theta )z.\]
Equivalently, these are quadrics which are pre-images of subsets in \(\mathbb {IH}\) under the Hopf mapping.
A centered quadric in \(\mathbb {H}\) is called non-singular if it does not contain the origin.
For an unbounded non-singular centered quadric in \(\R ^4\) given by
\[ F(z_0,z_1,z_2,z_3) =1 \]
where \(F\) is a quadratic homogeneous function of \(z= (z_0,z_1,z_2,z_3) \in \mathbb {H}\), we define its dual quadric by
\[ -F(z_0,z_1,z_2,z_3) =1. \]
In normal form, for the quadric
\[ \sum _{i= 0}^{3} \sigma _i \frac {\hat {z}_i^2}{a_i^2}=1, \]
where \(\sigma _i \in \{1, -1\}, a_i \in \R \) and \(\{\hat {z}_0\}_{i = 0}^3\) is an orthonormal basis in \(\R ^4\), its dual is
\[ \sum _{i= 0}^{3} - \sigma _i \frac {\hat {z}_i^2}{a_i^2}=1. \]
Indeed for a quadric homogeneous function \(F(z_0, z_1,z_2,z_3)\) there exists a real symmetric \(4 \times 4\) matrix \(A\) and a real orthogonal matrix \(Q\) such that \(z^{T}Az = F\) and \(Q^{T}AQ\) is diagonal, thus its normal form is given by \((Qz)^{T}A Qz
=1\). Clearly, we have \(z^{T}(-A)z = -F\) and \((Qz)^{T}(-A) Qz = -(Qz)^{T}A Qz \).
We first assume that \(\mathcal {E}\) is bounded in \(\mathbb {H}\). Then its image \(\mathcal {F}\) is also bounded in \(\mathbb {IH}\). If all points from \(\mathcal {F}\) have the same distance to the origin, then \(\mathcal {F}\) is a centered sphere in
\(\mathbb {IH}\). Otherwise, there exist a point \(P_{1}\) with least distance, and another distinct point \(P_{2}\) with most distance from \(O\). We consider the line passing through these two points and take a plane in \(\mathbb {IH}\) containing both this line and
the origin. By the above discussion on Levi-Civita planes, the intersection of this plane with the image \(\mathcal {F}\) is an ellipse focused at \(O\). Consequently the indicated line passes through the origin, since for an ellipse this line is the major axis and passes
through the foci. So the distance \(|P_{1} P_{2}|\) is the major axis length of this ellipse.
We consider the family of planes passing through this line. If we take such a plane close to the plane we first took, then by continuity, the intersection of \(\mathcal {F}\) on this plane is again an ellipse focused at \(O\) and the points \(P_1\) and \(P_2\) lie on the
ellipse as pericenter and as apocenter respectively. Thus the ellipses obtained as intersection of \(\mathcal {F}\) on nearby planes from the family are related by a rotation around the line \(P_{1} P_{2}\). Consequently \(\mathcal {E}\) is a spheroid with the line
\(P_{1} P_{2}\) as the symmetric axis.
This argument can be refined to the following local rigidity for (eccentric) ellipses, without assuming that \(\mathcal {F}\) is bounded: Consider the line \(P_{1} O\) and a plane through this line such that the intersection of \(\mathcal {F}\) with it is an ellipse. Then
the intersection of \(\mathcal {F}\) with nearby planes through \(P_{1} O\) are also ellipses, and these ellipses are obtained from each other by rotations along the axis \(P_{1} O\). Indeed all these ellipses need to intersect \(P_{1} O\) at the same point \(P_{2}\),
which necessarily is the apocenter for all of them. This implies this local rigidity for ellipses.
Now we consider the case that \(\mathcal {E}\) is not bounded, thus \(\mathcal {F}\) is not bounded as well. We take a point \(P_{1} \in \mathcal {F}\) which has the least distance from \(O\). Since the centered quadric \(\mathcal {E}\) is not given by a
positive-definite quadric form, its dual quadric \(\tilde {\mathcal {E}}\) is non-empty in \(\mathbb {H}\). The image in \(\mathbb {IH}\) of the dual \(\tilde {\mathcal {E}}\) is \(\tilde {\mathcal {F}}\). We take the point \(\tilde {P}_{1} \in \tilde
{\mathcal {F}}\) which has the least distance from \(O \in \mathbb {IH}\). From Lemma 6 , the three points \(O, P_1, \tilde {P}_1\) lie on the same line.
We consider the family of planes passing through this line. Since \(\mathcal {F}\) is unbounded, there exists a plane in this family which has unbounded intersection with \(\mathcal {F}\). Thus the intersection of \(\mathcal {F} \cup \tilde {\mathcal {F}}\) with
this plane is either a pair of two centrally symmetric parallel lines, a pair of branches of a hyperbola with its focus at \(O\), or a parabola with its focus at \(O\).
In the case of a hyperbola, note that we have the local rigidity just as in the elliptic case: In a nearby plane from this family, the intersection of \(\mathcal {F} \cup \tilde {\mathcal {F}}\) is again a hyperbola focused at \(O\), with \(P_1\) and \(P_2\) as vertices at
each branch. We conclude that \(\mathcal {F}\) is a branch of a circular two-sheeted hyperboloid with a focus at \(O\).
In the case of parallel lines, this local rigidity implies that \(\mathcal {F}\) intersects nearby planes in lines with \(P_{1}\) being the closed point from these lines to \(O\). We conclude \(\mathcal {F}\) is a plane perpendicular to the line \(OP_{1}\).
The only left case is when the intersection of \(\mathcal {F}\) with a plane containing \(OP_1\) is a parabola. This happens when the original quadric \(\mathcal {E}\) is unbounded and degenerate. From the local rigidity of ellipses and hyperbolae, we conclude that if
the intersection with a plane passing through \(OP_{1}\) is a parabola, then the intersections of \(\mathcal {F}\) with nearby planes passing through \(OP_{1}\), we again obtain parabolae. These parabolae are focused at \(O\) and have \(P_1\) as the vertex. Thus,
\(\mathcal {F} \) intersects the nearby planes from this family in parabolae with the focus and the vertex in common. Thus in this case the image \(\mathcal {F}\) is a paraboloid with a focus at \(O\).
\begin{equation}
\label {eq: S1-action-H} \theta \cdot z \mapsto \exp (i \theta )z, \quad z \in \mathbb {H},
\end{equation}
The proof is computational. Up to normalization, a spheroid in \(\mathbb {IH}\) focused at the origin is given by an equation of the form
\[ \frac {q_1^2 - \sqrt {C^2- D^2}}{C^2} + \frac {q_2^2}{D^2} + \frac {q_3^2}{D^2} - 1 = 0, \quad C>D>0. \]
The mapping \(z \mapsto Q= \bar {z}i \cdot z \) pulls this equation back to
\[ G_1 \cdot G_2 = 0 \]
where the factors are
\[ G_1:= Cz_1^2 + Cz_2^2 + Cz_3^2 + Cz_4^2 - 2\sqrt {C^2- D^2} z_1 z_3 - 2\sqrt {C^2- D^2} z_2 z_4 - D^2 \]
and
\[ G_2:= Cz_1^2 + Cz_2^2 + Cz_3^2 + Cz_4^2 + 2\sqrt {C^2- D^2} z_1 z_3 + 2\sqrt {C^2- D^2} z_2 z_4 + D^2. \]
It is readily seen that the equation \(G_{2}=0\) does not admit any real solutions.
In the rotated coordinates \((u_1,u_2,u_3,u_4)\) defined as
\[ z_1 = \frac {u_1+u_2}{\sqrt {2}}, z_2 = \frac {u_3+u_4}{\sqrt {2}}, z_3 = \frac {u_1- u_2}{\sqrt {2}}, z_4 = \frac {u_3-u_4}{\sqrt {2}}, \]
we write
\[ G_1= (C-\sqrt {C^2 - D^2})u_1^2 + (C+\sqrt {C^2 - D^2})u_2^2 + (C-\sqrt {C^2 - D^2})u_3^2 + (C+\sqrt {C^2 - D^2})u_4^2 - D^2=0 \]
and thus by a further normalization we get the desired form ( 6 ).
In [16] , it is noticed that conformal transformations between mechanical systems preserves billiard trajectories. A
generalization of this observation to our current situation is the following:
In the opposite direction, if \(\tilde {v}_{1}\) is reflected to \(\tilde {v}_{2}\) by the reflection at \(q\) off \(\tilde {\mathcal {R}} \subset \mathbb {H} \setminus O\), then for any \(z\) such that \(q=\bar {z} i z\), there exists based vectors \(v_{1},
v_{2}\) at \(z\) such that \((z, v_{1}), (z, v_{2}) \in \Sigma ^{1}\) which is pushed-forward into \((\tilde {v}_{1}, \tilde {v}_{2})\) by the Hopf mapping, such that \(v_{1}\) is reflected to \(v_{2}\) at \(z\) off the pre-image \(\mathcal {R}\) of \(\tilde
{\mathcal {R}}\).
\[BL(z, v_{1})=0.\]
Consider the normal vector \(N_{z}\) to \(\mathcal {R}\) at \(z\). Since \(\mathcal {R}\) is \(S^{1}\)-invariant, we have that \(N_{z}\) is orthogonal to the \(S^{1}\)-symmetric direction, which is given by \(i z\). Consequently, we have
\[BL(z, N_{z})={Re(\bar {z}iN_z)}=-\langle i z, N_{z} \rangle =0.\]
Since \(BL\) is linear in its second variable, we conclude that
\[BL(z, v_{2})=0\]
as well.
The second assertion follows as long as we show that the push-forward of \(N_{z}\) is orthogonal to \(\tilde {\mathcal {R}}\) at \(q=\bar {z} i z\). The push-forward of a vector \(v \in \Sigma \) is \({2}\bar {z} i v\) . Thus
\begin{equation}
\label {eq: equal angle} \langle \bar {z} i v, \bar {z} i N_{v} \rangle ={|z|^2 \langle v, N_{v} \rangle },
\end{equation}
For the opposite direction, if \(\tilde {v}\) is a vector at \(q \neq 0\) and \(z \in \mathbb {H} \setminus O\) such that \(q=\bar {z} i z\), then the vector \(v\) such that \(\bar {z} i v=\tilde {v}\) is a vector at \(z\) which is pushed-forward to \(\tilde
{v}\). With this construction we get at each \(z\) a pair of vectors \(\{v_{1}, v_{2}\}\) from the pair of vectors \(\{\tilde {v}_{1}, \tilde {v}_{2}\}\) at \(q\). There follows directly that
\[(z, v_{1}), (z, v_{2}) \in \Sigma ^{1}.\]
Moreover it follows from the angle-preservation relationship ( 8 ) that if \(\tilde {v}_{1}\) is reflected to \(\tilde {v}_{2}\), then
\(v_{1}\) is reflected to \(v_{2}\).
As part of the proof, we have shown that if an orbit of the four-dimensional Hooke problem satisfies the bilinear relation, then so is its reflection. Therefore we may say that a billiard orbit satisfies the bilinear relation. As only this type of orbits are related to the spatial
Kepler problem, we propose the following definition.
With these definitions we get the following theorem, which generalizes the planar Hooke-Kepler billiard correspondence as has been investigated in [16] and [22] .
This theorem is not limited to the integrable case and thus may be useful to understand the dynamics of non-integrable four-dimensional Hooke and three-dimensional Kepler billiards.
For the integrable case, we know that a four-dimensional Hooke billiard with a centered quadric reflection wall is integrable [6] ,
[4] . We directly obtain the following result, established in [23] via a completely different approach.
The first integrals for the three-dimensional integrable Kepler billiards can be obtained from the first integrals of the four-dimensional Hooke billiards. On the other hand the explicit representations of the first integrals are already obtained in [23] . We here recall:
where \(a\) a is the half distance between the two foci.
In [1] , Birkhoff designed a way to simultaneously desingularize the two double collisions of the particle with the two centers in
the planar problem. This has been subsequently generalized to the spatial problem as first announced in Stiefel and Waldvogel [18] . In [25] , Waldvogel explained that the construction is analogous to the observation that on
the Riemann sphere, Birkhoff's mapping is conjugate to the complex square mapping via a Möbius transformation. The approach was then subsequently applied to the spatial problem. The use of quaternions was introduced in [26] .
The goal of this section is to discuss this transformation in the spatial case with the language of the quaternions and symplectic geometry, with the hope of clarifying the geometry of this transformation even further. Subsequently we apply this transformation to the
problem of integrable billiards. The main fact we will use is that for a separable 2-degree of freedom Hamiltonian system of the form:
\begin{equation}
H = \frac {a(x)P_x^2+b(x)P_y^2}{2} + A(x)+ B(y)
\end{equation}
any coordinate line as a reflection wall results in an integrable mechanical billiard. Indeed one has the independent integrals \(\dfrac {a(x)P_x^2}{2}+A(x)\) and \(\dfrac {b(x)P_y^2}{2}+B(y)\) constant along orbits and under reflections at a coordinate line. The same
result can be obtained by considering the spatial two-center problem in spheroidal elliptic coordinates, as this approach leads to the same class of the separated system after reduction by rotations around the axis containing the centers. However, it is worth mentioning
that the method used here does not require elliptic coordinates; instead, it utilizes spherical coordinates through Birkhoff-Waldvogel's Transformation.
We first recall Waldvogel's view of Birkhoff transformation of the planar two-center problem from [26] . See also [2] for a discussion on the geometry of this transformation.
Consider the mappings
\[ \varphi _1: {\C \cup \{\infty \} \mapsto \C \cup \{\infty \}}, \quad z \mapsto \alpha = 1- \frac {2}{1-z}, \]
\[ L.C. : {\C \cup \{\infty \} \mapsto \C \cup \{\infty \}}, \quad \alpha \mapsto q=\alpha ^2, \]
\[ \varphi _2: {\C \cup \{\infty \} \mapsto \C \cup \{\infty \}}, \quad \mapsto x=1- \frac {2}{1-q}. \]
The mappings \(\varphi _{1}\) and \(\varphi _{2}\) are Möbius transformations on the Riemann sphere \(\C \cup \{\infty \}\). The mapping \(L.C.\) is the complex square mapping, branched at \(0, \infty \) on the Riemann sphere.
The composition of these mappings in the natural order gives rise to
\[ \varphi _2 \circ L.C. \circ {\varphi _1}: {\C \cup \{\infty \} \mapsto \C \cup \{\infty \}}, \quad z \mapsto x = \frac {z{+}z^{-1}}{2}. \]
This is Birkhoff's transformation, used to simultaneously regularize both double collisions with two Kepler centers placed at \(-1, 1 \in \C \).
This suggests the following construction for the spatial two-center problem.
We define the base Birkhoff-Waldvogel mapping as the composition
\[ \begin {split} \phi _2 \circ \hbox {Hopf} \circ \phi _1: &\mathbb {H} \cup \{\infty \} \mapsto \mathbb {IH} \cup \{\infty \},\\ &z \mapsto x=i-4\| z-i \|^4\left ((z- \bar {z} -2i)\|z-i\|^2 +2(z-i)i(\bar {z}+i)\right )^{-1}
\end {split} \]
where
\[ \begin {split} \phi _1: & \mathbb {H} \cup \{\infty \} \mapsto \mathbb {H} \cup \{\infty \},\\ & z \mapsto \alpha =i - \frac {2}{z-i}, \end {split} \]
\[ \begin {split} \hbox {Hopf}: & \mathbb {H} \cup \{\infty \} \mapsto \mathbb {IH} \cup \{\infty \}\\ & \alpha \mapsto q=\bar {\alpha } i \alpha \end {split} \]
and
\[ \begin {split} \phi _2: &\mathbb {IH} \cup \{\infty \} \to \mathbb {IH} \cup \{\infty \},\\ &q \mapsto x = i - \frac {2}{q-i}. \end {split} \]
In coordinates, we have
\begin{equation}
\label {eq: BW-entrywise-formula} \begin{split} &x_1= \frac {1}{2}\left (z_1+ \frac {z_1(z_0^2+1)}{z_1^2 +z_2^2 +z_3^2}\right )\\ &x_2 = \frac {1}{2}\left (z_2+ \frac {z_2(z_0^2-1) + 2 z_0 z_3}{z_1^2 +z_2^2 +z_3^2}\right )\\
&x_3 = \frac {1}{2}\left (z_3+ \frac {z_3(z_0^2-1) - 2 z_0 z_2}{z_1^2 +z_2^2 +z_3^2}\right ). \end {split}
\end{equation}
By restriction and properly lifting the mappings to the cotangent bundles, we get the unrestricted Birkhoff-Waldvogel mapping
\[ \begin {split} \widetilde {B.W.}:=\Phi _2 \circ K.S. \circ \Phi _1: & {(\mathbb {H} \setminus \{i, -i\}) \times \mathbb {H} \to (\mathbb {H} \setminus \{i, -i\}) \times \mathbb {H}} , (z, w) \mapsto (x,y) \end {split} \]
where
\[ \begin {split} \Phi _1: & (\mathbb {H} \setminus \{i, {-i}\}) \times \mathbb {H} \to (\mathbb {H} \setminus \{{0,i}\}) \times \mathbb {H}, \\ &(z, w) \mapsto \left (\alpha = i - \frac {2}{z-i}, \beta = \frac {\overline {(z -
i)}w \overline {(z- i)}}{2} \right ), \end {split} \]
\[ \begin {split} K.S.: & (\mathbb {H} \setminus {\{0,i\}}) \times \mathbb {H} \to (\mathbb {H} \setminus {\{0,i\}}) \times \mathbb {H} \\&(\alpha ,\beta ) \mapsto \left (q= \bar {\alpha }i \alpha , p= \frac { \bar {\alpha
}i\beta }{2 |\alpha |^2} \right ), \end {split} \]
\[ \begin {split} \Phi _2: &(\mathbb {H} \setminus \{{0}, i\}) \times \mathbb {H} \to (\mathbb {H} \setminus \{i, {-i}\}) \times \mathbb {H} \\&(q, p) \mapsto \left (x= i - \frac {2}{q-i}, y= \frac {\overline {(q- i)}p \overline
{(q- i)}}{2}\right ). \end {split} \]
Explicitly, the unrestricted Birkhoff-Waldvogel mapping \(\widetilde {B.W.}\) is given by \((z, w) \mapsto (x, y)\) with
The mappings \(\Phi _{1}, \Phi _{2}\) are constructed in a way that the transformations on positions are natural generalizations of \(\phi _{1}, \phi _{2}\), while the transformations on momenta are obtained as contragradients. The mapping \(K.S.\) is the usual
Kustaanheimo-Stiefel transformation.
In \((\mathbb {H}\setminus \{ i \}) \times \mathbb {H}\) we define the subsets
\[ \hat {\Lambda }:= \{(z,w) \in (\mathbb {H} \setminus (\R \cup \{i\} {\cup \{-i\}})) \times \mathbb {H} \mid Re((\bar {z}-i)w(\bar {z}+i)) = 0 \} \]
and
\[ \hat {\Sigma } := \{ (\alpha ,\beta ) \in (\mathbb {H}\setminus (\{e^{i \theta }\} {\cup \{0\}})) \times \mathbb {H} \mid BL(\alpha , \beta ) = Re(\bar {\alpha }i\beta ) = 0 \}. \]
Then we have the following:
To show this we first show
Since \(K.S.(\{ e^{i \theta } \})=i\) and \(\phi _{2}(i)=\infty \), this shows in particular that \(\R \subset \mathbb {H}\) represents physical infinity of the physical space \(\mathbb {IH}\).
\[ \begin {split} z= i- \frac {2}{\alpha -i} &= i-\frac {2}{\cos \theta - i (1- \sin \theta )}\\ & = i - \frac {2(\cos \theta + i (1- \sin \theta ))}{(\cos \theta - i (1- \sin \theta ))(\cos \theta + i (1- \sin \theta ))}\\ &
= i-\frac {\cos \theta + i (1- \sin \theta ) }{1- \sin \theta } \\ & = \frac {\cos \theta }{\sin \theta -1}. \end {split} \]
We thus have
\[ \left \{ z=\frac {\cos \theta }{\sin \theta -1}\mid \theta \in \R /2\pi \Z \right \} =\R . \]
\begin{equation}
\label {eq: BL-equiv-condition} \begin{split} Re(\bar {\alpha } i \beta ) = 0 &\Leftrightarrow Re \left (-i - \frac {2}{\bar {z}+i}\right )i \left (\frac {(\bar {z}+i)w(\bar {z}+i)}{2}\right ) = 0 \\ &\Leftrightarrow Re((1-2(\bar
{z}+i)^{-1}i)(\bar {z}+i)w(\bar {z}+i)){=0}\\ &\Leftrightarrow Re((\bar {z}+i)^{-1}(\bar {z} +i -2i)(\bar {z}+i)w(\bar {z}+i)) {=0} \\ &\Leftrightarrow Re((\bar {z} - i)w(\bar {z}+i))=0. \end {split}
\end{equation}
\[ z= \frac {2}{i - \alpha } + i \]
and
\[ w = 2 (\bar {z}+i)^{-1} \beta (\bar {z} + i )^{-1} = 2 \alpha \beta \alpha . \]
Thus, the first part of the lemma follows.
For \((q,p) \in (\mathbb {IH}\setminus \{i\}) \times \mathbb {IH}\), the conjugation of its image \((x,y)\) by \(\Phi _2\) is obtained as
\[ \bar {x} = -i - \frac {2}{-q+i}= -(i - \frac {2}{q-i}) = - x \]
and
\[ \bar {y}= - \frac {(q-i)p(q-i)}{2} = - \frac {(-q+i)p(-q+i)}{2} = - \frac {(\overline {q-i})p(\overline {q-i})}{2} = -y, \]
thus \((x, y) \in (\mathbb {IH}\setminus \{i\}) \times \mathbb {IH}\).
On the other hand, for any \((x, y) \in (\mathbb {IH}\setminus \{i\}) \times \mathbb {IH}\), the conjugate of its pre-image \((p,q)\) is obtained as
\[ \bar {q} =\frac {2}{-i + x} + i = - \left ( \frac {2}{i-x} +i \right ) = -q \]
and
\[ \bar {p}= -2 xyx = -p, \]
thus the pre-image \((p,q)\) belongs again to \((\mathbb {IH} \setminus \{i\}) \times \mathbb {IH}\).
\[ BL(\alpha , \beta )=0 \Leftrightarrow Re(\bar {\alpha } i \beta )= Re(p)=0. \]
On the other hand, for any \((q,p) \in (\mathbb {IH}\setminus \{i\}) \times \mathbb {IH}\), we can take \((\alpha , \beta ) \in \hat {\Sigma }\) such that \(K.S.(\alpha , \beta ) = (q,p)\). Indeed,for any \((q,p) \in \mathbb {IH} \times \mathbb
{IH}\), there exists an \(S^1\)-family \(\{(e^{i \theta _1}\alpha , e^{i \theta _1}\beta ) \}\) satisfying \(BL(\alpha , \beta ) =0\).
\[ B.W.: \hat {\Lambda } \to (\mathbb {IH}\setminus \{i,-i\}) \times \mathbb {IH} \quad (z,w) \mapsto (x, y), \]
where
\[ \begin {split} &x=i-|z-i|^{2}(2|z-i|^{-2}(zi\bar {z}+ \bar {z} -z +i)-\bar {z} +z -2i)^{-1}\\ &y = \frac {1}{|i - 2(z-i)^{-1}|^2}((z-i)^{-1}(\bar {z}+i)^{-1}i (\bar {z}+i)-i(\bar {z}+i)^{-1}+(z-i)^{-1}i+2(z-i)^{-1}(\bar
{z}+i)^{-1})\\ &\times w(1-(\bar {z}+i)(z-i)^{-1}+2i(z-i)^{-1}) \end {split} \]
is surjective.
\[ \begin {split} \Phi _1^{*}(Re(\bar {\beta } d \alpha )) &= Re \left ( \frac {(z-i)\bar {w}(z-i)}{2} \cdot (-2d(z-i)^{-1}) \right ) \\ &= Re \left ( \frac {(z-i)\bar {w}(z-i)}{2} \cdot 2(z-i)^{-1}(d(z-i)) (z-i)^{-1} \right )\\
&=Re(\bar {w} dz) \end {split} \]
Similarly, we get
\[ \Phi _2^{*}(Re(\bar {y} dx)) = Re(\bar {p} dq). \]
We now recall the fact
\[ K.S.|_{\Sigma }^* (Re(\bar {p} d q))= Re(\bar {\beta } d \alpha ). \]
Since \(\hat {\Sigma } \subset \Sigma \), we have
\[ K.S.|_{\hat {\Sigma }}^* (Re(\bar {p} d q)) = Re(\bar {\beta } d \alpha ). \]
By combining these facts, we obtain
\[ B.W.^{*}(Re(\bar {y}dx)) = Re(\bar {w} dz). \]
We now apply this mapping to the two center problem in \(\R ^{3} \cong \mathbb {IH}\), with the two centers at \(\pm i \in \mathbb {IH}\). We start with the shifted-Hamiltonian of the two-center problem
\[ H - f = \frac {|y|^2}{2} + \frac {m_1}{|x-i|} + \frac {m_2}{|x+i|} -f \]
and consider its 0-energy hypersurface. By multiplying the above equation by \(|x-i||x+i|\), we obtain
\[ |x-i||x+i|(H-f)= \frac {|y|^2|x-i||x+i|}{2} + m_1|x+i|+ m_2|x-i| -f|x-i||x+i|. \]
With the following identities
\[ \begin {split} &B.W.^{*} (|x-i|) = \frac {|z-i|^2}{|\bar {z} -z|}\\ &B.W.^{*} (|x+i|)= \frac {|z+i|^2}{|\bar {z} -z|}\\ &B.W.^{*} (|y|^2) = \frac {|\bar {z}-z|^4 |w|^2}{4|z-i|^2|z+i|^2} \end {split} \]
we obtain
\begin{equation}
\label {eq: def tilde K} \tilde {K}=\frac {|w|^2|\bar {z}-z|^2}{8} + m_1 \frac {|z+ i|^2}{|\bar {z} -z|} + m_2\frac {|z- i|^2}{|\bar {z} -z|} -f \frac {|z-i|^2|z+i|^2}{|\bar {z}-z|^2}=0,
\end{equation}
\begin{equation}
\label {eq: def K} K:= \frac {|w|^2}{8} + m_1 \frac {|z+ i|^2}{|\bar {z} -z|^3} + m_2\frac {|z- i|^2}{|\bar {z} -z|^3} -f \frac {|z-i|^2|z+i|^2}{|\bar {z}-z|^4}=0.
\end{equation}
Note that the Hamiltonian ( 13 ) is regular at the physical double collisions \(\{z=\pm i\}\). The physical collisions are therefore regularized. Its
singular set \(\{z \in \R \}\) corresponds to \(\infty \) of the original system, and is not contained in any finite energy level (Lem. 15 ).
\begin{equation}
\label {eq: plane-IH} k_2 x_2 + k_3 x_3 = 0
\end{equation}
\[\begin {cases}\label {eq: family-of-spheres-and-planes} & (\sin \theta z_0 - \cos \theta )^2 + (z_1^2 +z_2^2 +z_3^2) \sin ^{2} \theta = 1 \\ & k_2 (z_2 \cos \theta + z_3 \sin \theta )+ k_3(z_3 \cos \theta - z_2 \sin \theta ) =
0. \\ \end {cases}\]
For each \(\theta \not \equiv 0, \pi \, (mod \,\, 2 \pi )\), Equation ( ?? ) describes a
two-dimensional sphere as the intersection of a three-dimensional sphere with a hyperplane in \(\mathbb {H}\). We call them Birkhoff spheres. We denote them by \(S_{\theta , \kappa }\) respectively.
For \(\theta \equiv 0, \pi \, (mod \,\, 2 \pi )\), Equation ( ?? ) describes the plane
\[\begin {cases} & z_0 = 0 \\ & k_2 z_2 + k_3 z_3 = 0, \end {cases}\]
which we call a Birkhoff plane and we denote it by \(\pi _{\kappa }\). In both cases, the angle \(\kappa \) is the unique angle which satisfies
\[\cos \kappa =z_{2}, \sin \kappa =z_{3}.\]
Moreover, the mapping \(B. W. \) is restricted to the Birkhoff mapping on the cotangent bundle of a Birkhoff plane.
\[ k_1 q_2 + k_2 q_3 = 0 \]
in \(\mathbb {IH}\). The pre-image of this plane by the Hopf map is the family of Levi-Civita planes given by
\begin{align}
\left \{ \begin{aligned} & \alpha _0 \cos \theta + \alpha _1 \sin \theta =0\\ & k_2 (\alpha _2 \cos \theta + \alpha _3 \sin \theta )+ k_3(\alpha _3 \cos \theta - \alpha _2 \sin \theta ) = 0 \\ \end {aligned} \right .
\end{align}
in which \(\theta \in \R /2\pi \Z \) is an angle parametrizing the \(S^{1}\)-symmetry of the Hopf mapping. The pre-image of this family of Levi-Civita planes by \(\phi _{1}\) is
\begin{align}
\left \{ \begin{aligned} & \cos \theta (-z_0) + \sin \theta \left (z_1 - 1 + (z_0^2 + (z_1 -1)^2 + z_2^2 + z_3^2)^{2}/2 \right ) =0,\\ & k_2 (\cos \theta z_2 + \sin \theta z_3)+ k_3(\cos \theta z_3 - \sin \theta z_2) = 0, \\ \end
{aligned} \right .
\end{align}
which is equivalent to Eq. ?? . For the last assertion, the restriction of the \(B.W.\) mapping to a Birkhoff plane is the composition of
planar mappings each of them can be identified with \(\varphi _{2}, L.C., \varphi _{1}\) respectively. Indeed, the restriction of \(\phi _1\) to the \(ij\)-plane is obtained as
\[ \phi _1(z_1i + z_2j) = (1 - 2((z_1-1)^2+z_2)^{-1}(1-z_1))i + 2((z_1-1)^2+z_2)^{-1}z_2 j \]
which is equivalent to the Möbius transformation on \(\mathbb {C} \cup \{ \infty \}\) given by
\[ \varphi _1(z_0 + z_1 i) = 1-2((z_0 - 1)^2 + z_1^2)^{-1}(1-z_0) - 2(z_0 - 1)^2 + z_1^2)^{-1}z_1 i \]
up to some basis changes. One can generalize this identification to any planes in \(\mathbb {IH}\) containing the \(i\)-axis by rotating the plane with respect to the \(i\)-axis. Analogously, we can identify the restriction of \(\phi _2\) to a plane in \(\mathbb {IH}\)
containing the \(i\)-axis. Finally, we recall the argument from Proposition 5 and use the equivalence between the restriction of the Hopf mapping to the \(ij-\)plane in
\(\mathbb {IH}\) and the complex square mapping. The conclusion follows.
It is desirable to relate \(S_{\theta , \kappa }\) and \(\pi _{\kappa }\), as they are related by the symmetry of the Birkhoff-Waldvogel mapping. We also would like to introduce natural coordinates to analyze the transformed system. For this purpose, we have the
following lemma:
\[z= (r \cos \psi ) \, i+(r \sin \psi \cos \kappa ) \, j+(r \sin \psi \sin \kappa ) \, k,\]
and \(z_{\theta } \in S_{\theta , \kappa }\) be related to \(z\) by the action of the \(S^{1}\)-symmetry of the Birkhoff-Waldvogel mapping by shifting the corresponding angle by \(\theta \). Then we have
We may thus use \((r, \psi , \kappa , \theta )\) as coordinates for points in \(\mathbb {H} \setminus O\) with the help of Eq. ( 17 ). The
mapping \((r, \psi , \kappa , \theta ) \mapsto z:=z_{\theta }\) is seen to be 2-to-1, as both \((r, \psi , \kappa , \theta )\) and \((r, \psi , \kappa , \theta +\pi )\) is sent to the same point \(z \in \mathbb {H}\).
We compute \(\tilde {K}\) in Eq. ( 12 ) with these coordinates. We denote by \((P_{r}, P_{\psi }, P_{\kappa }, P_{\theta })\) the
corresponding conjugate momenta. We set \(P_{\theta }=0\), which is equivalent to the condition \(Re((\bar {z}+i)w (\bar {z}-i))=0\). This follows from Eq.( 11 ). We then obtain after this restriction the formula
\begin{align}
\label {eq: formula for tilde K in spherical coordinates} \tilde {K}=&\dfrac {r^{2} P_{r}^{2}}{2}+\dfrac { P_{\psi }^{2}}{2}+\dfrac {P_{\kappa }^{2}}{ 2 \sin ^{2} \psi } -\dfrac {2 r^2 P_{\kappa }^{2}}{(r^2-1)^2} \\&+4 f \cos ^2
\psi +(m_1-m_2) \cos \psi +\dfrac {(f r^2-( m_1+ m_2) r/2+f) (r^2+1)}{r^2}.
\end{align}
We have
\begin{equation}
\label {eq: separation in two parts for K} \tilde {K}=\tilde {K}_{1}+\tilde {K}_{2},
\end{equation}
\begin{equation}
\label {eq: expressions of each parts for K} \begin{aligned} &\tilde {K}_{1}(r, P_{r}, P_{\kappa })=\dfrac {r^{2} P_{r}^{2}}{2} -\dfrac {2 r^2 P_{\kappa }^{2}}{(r^2-1)^2} +\dfrac {(f r^2-( m_1+ m_2) r/2+f) (r^2+1)}{r^2} ;\\
&\tilde {K}_{2}(\psi , P_{\psi }, P_{\kappa })= \dfrac { P_{\psi }^{2}}{2}+\dfrac {P_{\kappa }^{2}}{ 2 \sin ^{2} \psi }+4 f \cos ^2 \psi +(m_1-m_2) \cos \psi . \end {aligned}
\end{equation}
The angle \(\kappa \) does not appear in this formula, reflecting the rotational invariance of the system around the axis of centers in \(\mathbb {IH}\). We may thus fix \(P_{\kappa }=C\). The further reduced Hamiltonian is
\begin{equation}
\label {eq: separation in two parts} \tilde {K}_{red}=\tilde {K}_{red, 1}+\tilde {K}_{red, 2},
\end{equation}
\begin{equation}
\label {eq: expressions of each parts} \begin{aligned} &\tilde {K}_{red, 1}(r, P_{r})=\dfrac {r^{2} P_{r}^{2}}{2} -\dfrac {2 r^2 C^{2}}{(r^2-1)^2} +\dfrac {(f r^2-( m_1+ m_2) r/2+f) (r^2+1)}{r^2} ;\\ &\tilde {K}_{red, 2}(\psi ,
P_{\psi })= \dfrac { P_{\psi }^{2}}{2}+\dfrac {C^{2}}{ 2 \sin ^{2} \psi }+4 f \cos ^2 \psi +(m_1-m_2) \cos \psi . \end {aligned}
\end{equation}
Both \(\tilde {K}_{red, 1}(r, P_{r})\), \(\tilde {K}_{red, 2}(\psi , P_{\psi })\) are 1 degree of freedom systems. The theory of [22] applies. Any finite combination of coordinate lines \(\{r=cst.\}\) and \(\{\psi =cst.\}\) in the \((r, \psi )\)-plane are integrable reflection walls.
It follows from Eq. ( 17 ) that each fibre of the \(B. W.\)-mapping intersects the subspace \(\mathbb {IH}\) in two points when \(r \neq 1\), and
lie completely in this subspace when \(r =1\). In this latter case, only the combination of the angles \(\theta +\kappa \) appears in the formula, meaning that in this case the \(\kappa \)-orbit is the same as the \(\theta \)-orbit. This is reflected in the formula ( 18 ) for \(\tilde {K}\), which is singular at \(\{r=1\}\) if \(C \neq 0\). Indeed
it is not hard to check that this set corresponds to the \(i\)-axis in the physical space \(\mathbb {IH}\). This follows from Eq.(
10 ). Otherwise, it is also singular at \(\psi =0, \pi \,(mod\, 2 \pi )\), corresponding again to the \(i-\)axis. With this in mind, we consider the restriction of the system to the set
\[\tilde {D}=\{z=z_{1} i + z_{2} j + z_{3} k \in \mathbb {IH} \setminus : |z| \neq 0, 1, (z_{2}, z_{3})\neq (0,0)\}\]
with (orthogonal) spherical coordinates \((r, \psi , \kappa )\), given by Eqs. ( 20 ), ( 21 ).
By a quadric of class R in \(\mathbb {IH}\) we mean a spheroid or a circular hyperboloid of two sheets there-in with foci at the two Kepler centers \(\pm i\) following [23] . Restricting the system in \(\mathbb {IH}\) to the Birkhoff planes and making use of [22] , we obtain that the above mentioned system is equivalent to the two-center billiards in \(\R ^{3}\) with any combinations of quadrics of class R as reflection walls.
This provides an alternative way to show the integrability of these two-center billiards, and generalizes [22] to dimension 3.
\[ H = \frac {|{w}|^2}{2} + m_0 |{z}|^2 + \frac {m_1}{|{z}-i|} + \frac {m_2}{|{z}+i|}, \]
with \(m_0, m_1, m_2 \in \R \) as parameters.
The same procedure shows that this system is separable after reduction in the same coordinates as above. Consequently, Thm. 22 also holds with the Lagrange problem as the
underlying system, as well as for other similar systems separable after reduction in these coordinates.
We here recall the explicit representation of the first integrals for the integrable Lagrange billiard obtained in [23] :
\begin{align*}
&E = \frac {\dot {x}^2+\dot {y}^2+\dot {z}^2}{2} + m_0 (x^2+y^2+z^2) + \frac {m_1}{\sqrt {(x-1)+y^2+z^2}} + \frac {m_2}{\sqrt {(x+1)+y^2+z^2}},\\ &L_{yz} = \dot {y}z-\dot {z}y,\\ &E_{sph}= \frac {1}{2}\left ( 2\dot {x}^2+\dot
{y}^2+\dot {z}^2 + ((\sqrt {2}x+1)\dot {y}- \sqrt {2}y \dot {x})^2+ (\sqrt {2}z\dot {x} - (\sqrt {2}x+1)\dot {z})^2+ (y\dot {z}-z\dot {y})^2 \right )\\ & \quad \quad + m_0 (2x^2+y^2+z^2) + \frac {m_1(1+\sqrt {2}x)}{\sqrt {(x-1/\sqrt
{2})^2+y^2+z^2}} +\frac {m_2(1-\sqrt {2}x)}{\sqrt {(x+1/\sqrt {2})^2+y^2+z^2}}.
\end{align*}
Setting \(m_0=0\) yields the first integrals of the two-center problem stated in Theorem B .
Acknowledgement We thank the referees and the editor for their suggestions and corrections. A.T. and L.Z. are supported by DFG ZH 605/1-1, ZH 605/1-2. A.T. is supported in part by DFG FR 2637/4-1. L.Z is also supported in part by the Heisenberg
programme (ZH 605/4-1).
[1] G. D. Birkhoff, The restricted problem of three bodies, Rend. Circ. Matem. Palermo , 39: 265-334, (1915).
[2] K. Cieliebak, U. Frauenfelder, and L. Zhao, \(J^+ \)-invariants for planar two-center Stark-Zeeman systems, Ergodic Theory Dynam. System , 43(7): 2258-2292, (2023).
[3] L. Euler, De motu corporis ad duo centra virium fixa attracti, Nov. Comm. Acad. Imp. Petropolitanae , 10: 207-242, (1767).
[4] Y. N. Fedorov, An ellipsoidal billiard with a quadratic potential, Funct. Anal. Appl. , 35(3): 199-208, (2001).
[5] E. Goursat, Les transformations isogonales en mécanique, C. R. Math. Acad. Sci. Paris , 108: 446-450, (1887).
[6] C. G. J. Jacobi, Vorlesungen über Dynamik , Verlag von Georg Reimer, Berlin (1866).
[7] H. J. Korsch and J. Lang, A new integrable gravitational billiard, J. Phys. A: Math. Gen. , 24: 45-52, (1991).
[8] V. V. Kozlov and D. V. Treschev, Billiards. A genetic introduction to the dynamics of systems with impacts. Translation of Mathematical Monographs, 89 , American Mathematical Society, Providence, (1991).
[9] V. V. Kozlov, Some integrable extensions of Jacobi's Problem of geodesics on an ellipsoid, J. Appl. Math. Mech. , 59(1): 1-7, (1995).
[10] P. Kustaanheimo, Spinor Regularization of Kepler Motion, Ann. Univ. Turkuens. , Ser. A, 73, (1964).
[11] P. Kustaanheimo and E. Stiefel, Perturbation Theory of Kepler Motion based on Spinor Regularisation, J. reine angew. Math. , 218: 204-219, (1965).
[12] J.-L. Lagrange, Recherches sur le mouvement d'un corps qui est attiré vers deux centres fixes, Second mémoire, VIII, Miscellanea Taurinensia , t. IV(1): 67-121, (1766-1769).
[13] T. Levi-Civita, Sur la résolution qualitative du problème restreint des trois corps, Verhandlungen des dritten Internationalen Mathematiker-Kongresses in Heidelberg vom 8. bis 13. August 1904 , B. G.
Teubner, Leipzig, (1904).
[14] T. Levi-Civita, Sur la régularisation du problème des trois corps, Acta. Math. , 42: 204-219, (1920).
[15] C. MacLaurin, Treatise of Fluxions. In two books. Ruddimans, Edinburgh, (1742).
[16] A. A. Panov, An elliptic billiard system with Newtonian potential, Mat. Zametki , 55(3): 139-140, (1994).
[17] H. Poritsky, The Billard Ball Problem on a Table With a Convex Boundary–An Illustrative Dynamical Problem, Ann. Math. , 51(2): 446-470, (1950).
[18] E. Stiefel and J. Waldvogel, Problème restreint des trois corps. Généralisation de la régularisation de Birkhoff pour le mouvement du mobile dans l'espace à trois dimensions, C.R.
Acad. Sc. Paris , 260: 805, (1965).
[19] E. Stiefel and G. Scheifele, Linear and Regular Celestial Mechanics , Springer-Verlag Berlin-Heidelberg, (1971).
[20] P. Saha, Interpreting the Kustaanheimo Stiefel transform in gravitational dynamics, Month. Not. Royal Astron. Soc. , 400(1): 228-231, (2009).
[21] A. Takeuchi and L. Zhao, Projective Integrable Mechanical Billiards, Nonlinearity , 37(1): 015011, (2023).
[22] A. Takeuchi and L. Zhao, Conformal Transformations and Integrable Mechanical Billiards, Adv. Math. , 436: 109411, (2024).
[23] A. Takeuchi and L. Zhao. Integrable Mechanical Billiards in Higher-Dimensional Space Forms, Regul. Chaotic Dyn. , 29(3): 405-434, (2024).
[24] A. P. Veselov, Confocal surfaces and integrable billiards on the sphere and in the Lobachevsky space, J. Geom. Phys. , 7(1): 81-107, (1990).
[25] J. Waldvogel, Die Verallgemeinerung der Birkhoff-Regularisierung für das räumliche Dreikörper-problem, Bull. Astron. , Série 3, Tome II, Fasc. 2, (1967), 295-341, (1965).
[26] J. Waldvogel, Quaternions for regularizing celestial mechanics: the right way, Cel. Mech. Dyn. Astron. , 102(1): 149-162, (2008).
[27] L. Zhao, Kustaanheimo-Stiefel regularization and the quadrupolar conjugacy, Regul. Chaot. Dyn. , 20: 19-36, (2008).
[28] L. Zhao, Projective Dynamics and an Integrable Boltzmann Billiard Model, Comm. Cont. Math. , 24(10): 2150085, (2022).
1. Introduction
Theorem A.
Consider a surface of revolution in \(\R ^3\), by revolving a conic with a focus at the origin about its principal
axis. Then reflecting spatial Kepler orbits (attracted to/repelled from the origin) off such a surface of revolution gives rise to an integrable mechanical billiard. Theorem B.
Consider a surface of revolution in \(\R ^3\) by revolving a conic with foci at the two Kepler centers around the axis joining the centers. Then reflecting orbits of the spatial two-center problem off such a surface of revolution is an integrable mechanical
billiard. Moreover, taking a finite combination of these surfaces does not destroy the integrability of the resulting two-center mechanical billiard systems. 2.
The Kustaanheimo-Stiefel Transformation
Proof.
Proposition 2.
Any zero-energy orbit of the four-dimensional Hooke problem with the shifted Hamiltonian
Proof.
Definition 3.
A Levi-Civita plane is a plane in \(\mathbb {H}\) spanned by two linearly independent unit quarternions \(v_1, v_2 \in \mathbb {H}\) satisfying \(BL(v_1, v_2)=0\). Proposition 4.
The Hopf mapping
Proof.
Proposition 5.
There exists an identification with \(\C \) of a Levi-Civita plane \(V\) together with its image under the Hopf mapping, such that under this identification, the
restriction of the K.S. mapping to \(T^*V\) is given by
Proof.
3. Application to integrable Hooke and Kepler billiards
Lemma 6.
For an unbounded non-singular centered quadric \(\mathcal {E}\) and its dual quadric \(\tilde {\mathcal {E}}\) in \(\mathbb {H}\), we denote their images in
\(\mathbb {IH}\) by the Hopf mapping by \(\mathcal {F}\) and \(\tilde {\mathcal {F}}\) respectively. Let \(P \in \mathcal {F}\) be the point of \(\mathcal {F}\) with the least distance from \(O \in \mathbb {IH}\). Let \(\tilde {P} \in \tilde {\mathcal
{F}}\) be the point of \(\tilde {\mathcal {F}}\) with the least distance from \(O \in \mathbb {IH}\). Then the three points \(O, P, \tilde {P}\) are collinear. Proof.
Proposition 7.
The image of any \(S^{1}\)-invariant, non-singular, centered quadric in \(\mathbb {H}\) under the Hopf mapping is either a plane, or
a centered sphere, or a spheroid, or a sheet of a two-sheeted circular hyperboloid, or a paraboloid in \(\mathbb {IH}\), with always a focus at the origin \(O \in \mathbb {IH}\) in the latter three cases. These surfaces correspond precisely to those obtained by revolving a
Kepler orbit (a conic with focus at \(O\)) about its principal axis.
Proof.
Corollary 8.
Any combination of confocal \(\mathcal {S}^1\)-invariant centered spheroids or two-sheeted circular hyperboloids in \(\mathbb {H}\) is sent to a combination of confocal spheroid or a sheet of a two-sheeted circular hyperboloid. Proof.
Proposition 9.
If a centered quadric in \(\mathbb {H}\) is invariant under the \(\mathcal {S}^1\)-action on \(\mathbb {H}\) given by Proof.
Proposition 10.
Let \(\mathcal {R}\) be an \(S^{1}\)-invariant hypersurface in \(\mathbb {H} \setminus O\) and \(\tilde {\mathcal {R}} \subset \mathbb {IH}\) its image under the Hopf mapping. Let \(v_{1}\) be an incoming vector at a point \(z \in
\mathcal {R}\) such that \((z, v_{1}) \in \Sigma ^{1}\) with the outgoing vector \(v_{2}\) after reflection. Then \((z, v_{2}) \in \Sigma ^{1}\). Assume that the Hopf mapping pushes \((v_{1}, v_{2})\) into \((\tilde {v}_{1}, \tilde {v}_{2})\). Then
\(\tilde {v}_{1}\) is reflected to \(\tilde {v}_{2}\) by the reflection at \(q=\bar {z} i z\) off \(\tilde {\mathcal {R}}\).
Proof.
Definition 11.
The subsystem of a four-dimensional Hooke billiard consisting only of orbits satisfying the bilinear relation is called the restricted four-dimensional Hooke billiard . Definition 12.
A spatial Kepler billiard and a four-dimensional Hooke billiard are called in correspondence , if the reflection wall of the Hooke problem in \(\mathbb {H}\) is the pre-image of the reflection wall of the Kepler
problem in \(\mathbb {IH}\) by the Hopf map. Theorem 13.
Any billiard orbit of the spatial Kepler billiard is the image of an \(S^{1}\)-family of billiard orbits of the corresponding restricted four-dimensional Hooke billiard. In the opposite
direction, the image of any orbit of the restricted four-dimensional Hooke billiard under the Hopf mapping is an orbit of the corresponding spatial Kepler billiard. Theorem A.
Consider a surface of revolution in \(\R ^3\), by revolving a conic with a focus at the origin about its principal axis. Then reflecting spatial Kepler orbits (attracted to/repelled from the origin) off such a surface of revolution gives rise to an integrable
mechanical billiard. 4.
The two-center problem and integrable billiards
Lemma 14.
The image of the mapping \(\Phi _1\) with the restricted domain \(\hat {\Lambda }\) is \(\hat {\Sigma }\). Additionally, the image of the mapping \(\Phi _2\) restricted to
\((\mathbb {IH}\setminus \{i, {0}\})\times \mathbb {IH}\) is \((\mathbb {IH}\setminus \{i,{-i}\}) \times \mathbb {IH}\). Proof.
Proof.
Lemma 16.
The image of the \(K.S\) mapping restricted to \(\hat {\Sigma } \) is \((\mathbb {IH}\setminus \{i, {0}\}) \times \mathbb {IH}\). Proof.
Proposition 17.
The restricted Birkhoff-Waldvogel mapping
Proposition 18.
\(B.W.^* (Re(d \bar {y} \wedge dx)) = Re(d \bar {w}\wedge dz) |_{{\hat {\Lambda }}}\). Proof.
Proposition 19.
Consider a plane in \(\mathbb {IH}\) containing the \(i\)-axis given by the equation Proof.
Lemma 20.
Let \(z \in {\pi _{\kappa }}\) be expressed as
Proof.
Proposition 21.
A mechanical billiard system in \(\mathbb {IH}\), defined by the restriction of \(\tilde {K}\) and any finite combination of concentric spheres and any cones symmetric around the \(i\)-axis with the vertex at the origin is integrable.
Proof.
Theorem B.
Consider a surface of revolution in \(\R ^3\) by revolving a conic with foci at the two Kepler centers around the axis joining the centers. Then reflecting orbits of the spatial two-center problem on such a surface of
revolution R is an integrable mechanical billiard. Moreover, taking a finite combination of these surfaces does not destroy the integrability of the resulting two-center mechanical billiard systems. Remark 23.
The Lagrange problem in \(\R ^3\) is given by the Hamiltonian
References
99