1: \chapter[\protect\uppercase{Methodology}]{Methodology}\label{ch.ch3}
2: \addtocontents{toc}{\protect\vspace*{0.2in}}
3:
4: In this chapter, first we review two existing schemes of critical points analysis for systems with one polyad number in $\S$ 3.1, 3.2. Then in $\S$ 3.3, additional questions concerning arbitrary DOF and multiple polyad numbers are addressed, in order to formulate a generalized version of the critical points analysis. The method establishes that near a stable critical point in the reduced phase space, classical trajectories are quasiperiodic. These critical points therefore indicate the existence of regular modes of vibration.
5:
6: \section[Critical Points Analysis of Single $m:n$ Resonance]{\underline{Critical Points Analysis of Single $m:n$ Resonance}}
7: \addtocontents{toc}{\protect\vspace*{7pt}}
8:
9: This analysis was developed by Kellman \et for the effective Hamiltonian consisting of two zero-order modes coupled by a single resonance. Below, a brief overview is given on aspects that will be used in Chapter 4. For a more detailed description, the reader is referred to \cite{KellmanReview,SvitakMNRes}.
10:
11: \subsection{3.1.1 The $m:n$ Resonance Hamiltonian}
12: \addtocontents{toc}{\protect\vspace*{5pt}}
13:
14: In many triatomic molecules, the coupling between two vibrational modes $1$ and $2$ (not necessarily normal modes) can be approximated by an $m:n$ type resonance. Both $m$ and $n$ are positive integers. Eqn. (\ref{FermiExample}) is one example with $m:n=1:2$. For the general $m:n$ resonance the effective Hamiltonian takes the following form:
15: \begin{eqnarray}
16: \hat{H}_{mn}=\hat{H}_0(n_1, n_2) + V_{mn} \mbox{[} {(\hat{a}_1^\dagger)}^{m}{(\hat{a}_2)}^{n}+{(\hat{a}_2^\dagger)}^{n}{(\hat{a}_1)}^{m} \mbox{]} \label{mnresquantum}
17: \end{eqnarray}
18: \noindent The second term on the right hand side corresponds to a matrix element between ZOS $\vert n_1, n_2 \rangle$ and $\vert n_1+m, n_2-n \rangle$. This coupling destroys both $n_1$, $n_2$ as exact quantum numbers, but preserves one polyad number
19: \begin{eqnarray}
20: P_{mn}=\frac{n_1}{m} +\frac{n_2}{n}
21: \end{eqnarray}
22:
23: Using eqn. (\ref{Heisenberg}), a classical Hamiltonian in action-angle variables ($I_1, \phi_1, I_2, \phi_2$) is obtained from $\hat{H}_{mn}$. Then let $\sigma$ be the largest common factor between $m$ and $n$, the following canonical transformation is carried out:
24: \begin{align}
25: I & = \frac{\sigma}{2}\left(\frac{I_1}{m} + \frac{I_2}{n}\right), & \theta & = \frac{m\phi_1+n\phi_2}{\sigma} \\
26: I_z & = \frac{\sigma}{2}\left(\frac{I_1}{m} - \frac{I_2}{n}\right), & \Psi & = \frac{m\phi_1-n\phi_2}{\sigma} \label{IzPsi}
27: \end{align}
28:
29: $I$ is the constant of motion differing from $P_{mn}$ by a constant, while $\theta$ is a cyclic angle. $I_z$ can be regarded as a measure for the extent of mixing between the zero-order oscillators 1 and 2, and $\Psi$ as their relative phase angle. The classical Hamiltonian becomes
30: \begin{eqnarray}
31: H_{mn}=H_0(I, I_z)+2V_{mn} (I+I_z)^{\frac{m}{2}} (I-I_z)^{\frac{n}{2}} \cos[\sigma \Psi] \label{mnres}
32: \end{eqnarray}
33:
34: The 2 DOF Hamiltonian of (\ref{mnres}) is integrable, since both $H_{mn}$ and $I$ are constants of motion. The classical phase space is reduced to 1 DOF with the equations of motion:
35: \begin{align}
36: \dot{I}_z & = -\frac{\partial H}{\partial \Psi}, & \dot{\Psi} & = \frac{\partial H}{\partial I_z}
37: \end{align}
38:
39: \subsection{3.1.2 The Polyad Phase Sphere and Critical Points}
40: \addtocontents{toc}{\protect\vspace*{5pt}}
41:
42: The reduced phase space ($I_z, \Psi$) has the same topology as the surface of a 2-dimensional sphere \cite{XiaoPre}. The sphere is called the {\it Polyad Phase Sphere} (PPS) \fn {Alternatively, ($I_z, \Psi$) could be regarded as a special SOS in the full phase space ($I, \theta, I_z, \Psi$). Instead of energy, here $I$ is held constant. The dividing surface is defined by a constant $\theta$.}. As shown in Fig.~\ref{phasesphere}, the angle arccos$\lbrack {I_z}/{I} \rbrack$ is the longitude of the PPS, while $\Psi$ is its latitude. The north pole ($I_2=0$) and the south pole ($I_1=0$) are the mode 1 and 2 overtones, respectively. At these two points, $\Psi$ becomes unphysical, since the phase angle of an oscillator is ill defined when the action vanishes.
43:
44: \newpage \begin{figure}[hbtp]
45: %==\vspace{3.32in}
46: \begin{center}\includegraphics[width=5.98in]{phasesphere.eps}\end{center}
47: \cpn{Coordinates on the PPS}{Coordinates on the PPS. The rectangular plane in panel (a) is a Mecartor projection of the spherical surface in panel (b). Identical points are labeled by A-D in both panels to aid visualization. \label{phasesphere}}
48: \end{figure} \newpage
49:
50: In an integrable Hamiltonian, each eigenstate of $\hat{H}_{mn}$ is associated with an invariant torus in ($I$, $\theta$, $I_z$, $\Psi$) via {\it Einstein-Brillion-Keller} (EBK) quantization \cite{EBK}. On the PPS (which eliminates $I$ and $\theta$), each torus appears as a closed semiclassical trajectory. In practice, this trajectory can be well approximated (typically within 1 cm$^{-1}$) by simply solving for points on the PPS with the same energy as the quantum state \cite{HCP}.
51:
52: On the PPS, all semiclassical trajectories are organized by the critical points in the reduced phase space:
53: \begin{align}
54: \dot{I}_z & =-\frac{\partial H}{\partial \Psi}=0 \\
55: \dot{\Psi} &= \frac{\partial H}{\partial I_z} = 0
56: \end{align}
57: When there is no resonance ($V_{mn}=0$), all trajectories are parallel to the equator of the PPS, because $H_{0}$ has no dependence on $\Psi$. The only critical points are then the north and south poles. This corresponds to the trivial case where the eigenstates are assigned by the zero-order quantum numbers ($n_1,n_2$). When the resonance is turned on, new critical points may emerge and the old ones may change their stabilities. The semiclassical trajectories, as well as the quantum states they represent, change accordingly. A bifurcation of the critical points signals the birth, death and/or transformation of the vibrational modes.
58:
59: Fig.~\ref{sphereHCP} shows a sample PPS for the HCP molecule ($m:n=1:2$). Here modes 1 and 2 refer to the normal C-P stretch and normal H-C-P bend, respectively. In this particular polyad $P=n_1+n_2/2 = 11$ there are 12 eigenstates, and their trajectories (labeled 0-11) are evenly spread over the surface of the PPS. The most prominent structure on the PPS is a separatrix (dashed line) with the unstable critical point {\bf $\overline{[SN]}$} (``{\bf X}") at the center of its ``figure eight" shape. The separatrix is so-named because it separates the phase space into three regions: (1) occupied by trajectories 0-7 which surround the stable critical point {\bf $[B]$}; (2) occupied by trajectories 9,11 which surround the stable critical point {\bf $[SN]$}; and (3) occupied by levels 8,10 which surround the stable critical point {\bf $[r]$} at the north pole. Each level can be assigned to two quantum numbers: one is the polyad number $P$, the other determined by the critical point its trajectory surrounds. Since the surface of the PPS is divided, the assignment is not uniform for all 12 states.
60:
61: \newpage \begin{figure}[hbtp]
62: %==\vspace{2.71in}
63: \begin{center}\includegraphics[width=5.07in]{fig09.eps}\end{center}
64: \cpn{PPS and semiclassical trajectories} {PPS and semiclassical trajectories from \cite{HCP}. (a) a view of the PPS. The points labeled $[r]$, $[B]$ and $[SN]$ are stable critical points, while $\overline{[SN]}$ is an unstable critical point. The dashed line is the separatrix. Panel (b) presents a cut along the great circle defined by $\Psi=0, \pi$, where $\theta \in [0,2\pi]$ is a parameter around the great circle. Panel (b) also shows the relative energy of eigenstates in this polyad. \label{sphereHCP}}
65: \end{figure}
66:
67: \newpage \begin{figure}[hbtp]
68: %==\vspace{2.01in}
69: \begin{center}\includegraphics[width=4.77in]{fig13.eps}\end{center}
70: \cpn{Gap in the energy pattern} {Gap in the energy pattern for the eigenstates in Fig~\ref{sphereHCP}, from \cite{HCP}. Plotted are $\Delta E = E_1-E_2$ vs. $E_{avg} = (E_1+E_2)/2$ between pairs of levels whose label are in the parentheses on the graph. \label{fig3.4}}
71: \end{figure} \newpage
72:
73: In the full phase space, although $I$, $I_z$ and $\Psi$ are fixed at the critical points, the cyclic angle $\theta$ is not. Instead, its value (modulo $2\pi$) precesses between $\lbrack 0, 2\pi \rbrack$ at a constant frequency. Hence, a critical point in the reduced phase space corresponds to a PO in the full phase space.
74:
75: \subsection{3.1.3 Spectral Patterns}
76: \addtocontents{toc}{\protect\vspace*{5pt}}
77:
78: Patterns in the quantum spectra are reflected in the semiclassical trajectories on the PPS. First, the ratio between $n_1$ and $n_2$ for each state corresponds to the time-averaged $I_z$ of its trajectory. In Fig.~\ref{sphereHCP}, for example, levels 9,11 with $I_1 \ll I_2$ are expected to have a strong bending character. These levels are identified in experimental spectra by their large rotational constant \cite{HCPMikami}.
79:
80: Second, the separatrix on the PPS acts like a barrier in phase space. The classical frequency traversing the top of the barrier is expected to drop to zero. The quantum equivalence of this frequency is the energy difference $\Delta E$ between adjacent levels. The pattern of $\Delta E$ therefore exhibits a dip when a separatrix is crossed.
81:
82: When there are more than 2 regions on the PPS (e.g. Fig.~\ref{sphereHCP}), $\Delta E$ should be taken only between eigenstates within the same region on the PPS. If the states are sorted by energy alone, levels 9, 11 are intermingled with 8,10. As shown in Fig.~\ref{fig3.4}, this choice creates a ``zigzag" pattern in panel (a). The smooth dip is recovered in panel (b), when the energy differences are taken within the same zone. This resorting procedure was first discussed by Svitak \et in \cite{SvitakPattern1}.
83:
84: \subsection{3.1.4 Catastrophe Map}
85: \addtocontents{toc}{\protect\vspace*{5pt}}
86:
87: If $H_{eff}$ includes up to quadratic terms in $H_0$ and $V_{mn}$ is a constant, all possible PPS structures for a given $m:n$ can be further summarized by just two independent parameters with the help of catastrophe theory in mathematics \cite{catastrophe}. The PPS up to a scaling factor can be reconstructed from these parameters. This 2-parameter space (called {\it catastrophe map}) is divided into zones for any $m:n$ system \cite{SvitakMNRes}, and within each zone the PPS have the same {\it qualitative} structure. As an example, Fig.~\ref{catmap} displays the catastrophe map and representative PPS for $m:n=1:1$ \fn{3D models of these spheres are also included on the accompanying CD-ROM.}. In zone {\bf I}, (cases 1, 5, 7, 8 and 9) the spheres share an undivided structure, while in zone {\bf II} (cases 3, 4 and 6) the spheres are each divided by a separatrix. In going from spheres 1-2-3-4, the bifurcation occurs at sphere 2 where its representative point crosses from {\bf I} (normal mode dynamics) to {\bf II} (local mode dynamics).
88:
89: \newpage \begin{figure}[hbtp]
90: %==\vspace{3.26in}
91: \begin{center}\includegraphics[width=5.66in]{fig11a.eps}\end{center}
92: \cpn{Catastrophe map of $1:1$ resonance system} {Catastrophe map of $1:1$ resonance system and associated PPS, adapted from \cite{BEC}. Panel (a) is the catastrophe map with two independent parameters being $\mu$ and $\Lambda$. Panel (b) displays the PPS corresponding to the representative points labeled 1-9 on panel (a). \label{catmap}}
93: \end{figure} \newpage
94:
95: One limitation of the catastrophe map is that its extension to include high-order terms is nontrivial. As shown in $\S$ 5.2 of \cite{SvitakThesis}, the addition of a single cubic term in $H_0$ adds substantial complexity to the catastrophe map. When the high-order terms are indeed not ignorable, a simpler alternative using the PPS and spectral patterns alone, since they contain the same amount of information.
96:
97: \subsection{3.1.5 Summary}
98: \addtocontents{toc}{\normalfont\normalsize{\noindent Chapter \hfill Page}\protect\vspace*{30pt}}
99:
100: The steps discussed in $\S$ 3.1.1 - 3.1.4 for the single resonance analysis are summarized in Fig.~\ref{integrableflow}. The 2-dimensional reduced phase space is directly visualized with the PPS. On the PPS, each quantum state corresponds to a semiclassical trajectory. The trajectory can be assigned quantum numbers by the stable critical point it surrounds. A separatrix (associated with an unstable critical point) causes a ``dip" in the energy gap pattern ($\Delta E$ versus $E_{avg}$) when the trajectory traverses the separatrix. All possible divisions on the PPS with the same $m:n$ resonance can be further classified by two parameters on the catastrophe map.
101:
102: \newpage \begin{figure}[hbtp]
103: %==\vspace{3.98in}
104: \begin{center}\includegraphics[width=3.98in]{xiaochart.eps}\end{center}
105: \cpn{Critical points analysis of the $m:n$ resonance Hamiltonian} {Critical points analysis of the $m:n$ resonance Hamiltonian, adapted from Fig.~2 of \cite{xiaocatmap} with modifications. \label{integrableflow}}
106: \end{figure} \newpage
107:
108: \section[Large-Scale Bifurcation Analysis]{\underline{Large-Scale Bifurcation Analysis}}
109: \addtocontents{toc}{\protect\vspace*{7pt}}
110:
111: %== The Principle and Example
112:
113: In a non-integrable Hamiltonian, the main distinction in the classical phase space structure is between the regular and the chaotic regions. Even today, it remains a poorly understood field. The most difficult cases have multiple resonances acting together, preventing reduction of the dynamics to less than 3 DOF. Lu and Kellman proposed the {\it large-scale bifurcation analysis} as an extension to these non-integrable systems with one polyad number \cite{Zi-MinH2O1,Zi-MinH2O2}.
114:
115: The main assumption here is that {\it ``The large-scale bifurcation structure is defined by the lowest-order periodic orbits and their bifurcations"} \cite{Zi-MinH2O1}. Especially, when the reduced phase space has 2 DOF, regions with different types of dynamics can be visually recognized on an SOS. Each regular region surrounds a ``periodic orbit" on the SOS. Here the word ``periodic" should not be confused with the continuous $T$ in eqn. (\ref{PODefinition}) for a PO. It refers to the trajectory that appears on the SOS at a few discrete points (as opposed to filling a continuous curve/area). The period is the integer number of steps between the returns. Those with period 1 are also known as {\it fixed points} on the SOS \fn{In existing literature, ``fixed points" and ``critical points" are often used interchangeably. In this thesis, ``fixed points" is used in the context of a discrete mapping (such as an SOS). ``Critical points" refer to stationery points in a {\it continuous} dynamical system.}.
116:
117: Consider a 3 DOF system with one polyad number, such as the Baggott H$_2$O Hamiltonian in \cite{Baggott}. The polyad number enables one to rewrite the Hamiltonian in a 4-dimensional reduced phase space ($I_1, \psi_1, I_2, \psi_2$), plus a conserved action $I_3$ and a cyclic angle $\psi_3$. Dynamics in the reduced phase space can be visualized using a series of SOS. Without loss of generality, let the energy and coordinate $\psi_2$ be held constant in the construction of this SOS, and the crossings of trajectories be recorded in ($I_1, \psi_1$) space. A fixed point on the resulting SOS has all four action-angle variables ($I_1, \psi_1, I_2, \psi_2$) constant -- therefore it must be a critical point of the reduced phase space:
118: \begin{eqnarray}
119: \dot{I}_1=\dot{\psi}_1=\dot{I}_2=\dot{\psi}_2=0 \label{largescalebifur}
120: \end{eqnarray}
121: \noindent Unless $\psi_3$ has zero frequency, these critical points are closed PO in the full phase space.
122:
123: These fixed points on the SOS therefore can be found by solving the simultaneous analytic equations (\ref{largescalebifur}). It avoids numerical integration of many individual trajectories, as well as the subsequent problem of classifying their behavior.
124:
125: Ref.~\cite{Zi-MinH2O1} solved the bifurcation structure of critical points for several triatomic systems. Fig.~\ref{LuKellman} shows the results in H$_2$O. In the limit $P \rightarrow 0$, there are 3 branches of critical points corresponding to the 3 normal modes in H$_2$O. In the lower right corner of Fig.~\ref{LuKellman}, the normal bend family is oriented vertically from the origin, while the two normal O-H stretch families (on top of each other in this figure) are along the short diagonal segment between the origin and point A. As $P$ is increased, resonances cause the normal modes to bifurcate (at points A, B, B', etc.) into new families of critical points. These critical points were then used to successfully assign all eigenstates in polyad $P=8$ to quantum numbers consistent with their vibrational dynamics \cite{Zi-MinH2O2}.
126:
127: \newpage \begin{figure}[hbtp]
128: %==\vspace{3.07in}
129: \begin{center}\includegraphics[width=3.92in]{lukellman.eps}\end{center}
130: \cpn{Large-scale bifurcation structure in H$_2$O} {Large-scale bifurcation structure in H$_2$O, reproduced from Fig.~3 in \cite{Zi-MinH2O1}. \label{LuKellman}} \end{figure} \newpage
131:
132: Two subsequent studies \cite{EzraH2O,WigginsH2O} examined the finer details of the phase space structures in the same system. Their agreement with \cite{Zi-MinH2O1} showed the validity of using critical points to identify the large-scale phase space structures.
133:
134: \section[Generalized Critical Points Analysis]{\underline{Generalized Critical Points Analysis}}
135: \addtocontents{toc}{\protect\vspace*{7pt}}
136:
137: In $\S$ 3.1 and 3.2, the importance of critical points is illustrated for both integrable and nonintegrable Hamiltonians. Two aspects are worthy of emphasizing:
138:
139: \noindent \,\,\, (1) \,\,\, The existence of at least one polyad number is crucial for this analysis. In reducing the DOF of the classical Hamiltonian, the cyclic angle(s) not explicit in the reduced phase space provides time evolution for the critical points in the full phase space. In contrast, a critical point defined in the {\it full} phase space usually conveys little information about the dynamics. For example, although two coupled anharmonic oscillators may exhibit a rich range of dynamical behavior, this is not apparent from examination of the equilibrium point (usually with no motion in either oscillator). With a {\it single} polyad number, critical points in the reduced phase space are PO in the full phase space. These POs form the ``skeletons" of phase space \cite{EzraH2O}. With {\it multiple} polyad numbers, the critical points are expected to have the same importance, although they now correspond to invariant tori in the full phase space.
140:
141: \noindent \,\,\, (2) \,\,\, The critical points are found by solving analytically defined equations. Because the method does not rely on numerical integration of Hamilton's equations, it circumvents the problem induced by unstable/chaotic trajectories. Unlike most existing nonlinear methods, the equations can be extended to arbitrary number of DOF without significant change.
142:
143: Nevertheless, so far we only considered systems with 1 polyad number and up to 3 DOF. The following three points need to be addressed in order to extend the analysis to multiple polyad numbers and arbitrary number of DOF.
144:
145: \begin{enumerate}
146: \item In a 2 DOF system, the consistency between critical points and large-scale phase space structure may be verified by direct inspection, such as through SOS. These visual aids are increasingly costly in higher dimensions. Although it was suggested that the large-scale bifurcation analysis could be extended to $> 3$ DOF systems with one polyad number \cite{KellmanReview}, it remains unclear how the ``periodic orbits" defined on a SOS (see $\S$ 3.2) can be extended to arbitrary DOF. A dimensionality-independent description of the dynamics related to a critical point is strongly preferred.
147:
148: \item With multiple polyad numbers, the critical points generally have multiple non-commensurable frequencies associated with the cyclic angles. Motion at these critical points is quasiperiodic in the full phases space, instead of being closed PO. To what extent would this difference affect the classical and quantum dynamics of the molecule?
149:
150: \item In references \cite{EzraH2O,Zi-MinH2O2}, the eigenstate assignment was performed through visually identifying the localization of the Husimi distribution function of the eigenstates. As both the computation of these semiclassical wavefunctions and the visual assignment become impractical in higher dimensions, a more general consideration of how to assign wavefunction localization is necessary.
151: \end{enumerate}
152:
153: The next three subsections $\S$ 3.3.1-3.3.3 discuss these questions in their order. The result is a more generalized version of the critical points analysis, which will be used in Chapter 4 on the pure bending subsystem of C$_2$H$_2$.
154:
155: \subsection{3.3.1 Reduced Phase Space Trajectory Near Critical Points}
156: \addtocontents{toc}{\protect\vspace*{5pt}}
157:
158: First, we consider an effective Hamiltonian of the most general form. Let the Hamiltonian have a total of $N$ modes, $M$ linearly independent resonance vectors, and ($N-M$) polyad numbers. The classical Hamiltonian after a suitable canonical transformation has $2M$ action-angle variables spanning the reduced phase space:
159: $$\vec{X}=\{ x_i \} = \{ J_1, \ldots, J_M, \Psi_1, \ldots, \Psi_M \}$$
160: \noindent and ($N-M$) constants of motion and their conjugate cyclic angles
161: $$\{ P_{M+1}, \ldots , P_N, \theta_{M+1}, \ldots , \theta_{N} \}$$
162:
163:
164: Hamilton's equations of motion in the reduced phase space can be written in the following matrix form
165: \begin{eqnarray}
166: \frac{d}{dt} \vec{X} = \left( \begin{array}{c} -\frac{\partial H}{\partial x_{M+1}} \\ \cdots \\ -\frac{\partial H}{\partial x_{2M}} \\ \, \\ \frac{\partial H}{\partial x_1} \\ \cdots \\ \frac{\partial H}{\partial x_M} \end{array} \right) = \left( \begin{array}{cc} 0 & -E_M \\ E_M & 0 \end{array} \right) \left( \begin{array}{c} \frac{\partial H}{\partial x_1} \\ \cdots \\ \frac{\partial H}{\partial x_M} \\ \, \\ \frac{\partial H}{\partial x_{M+1}} \\ \cdots \\ \frac{\partial H}{\partial x_{2M}} \end{array} \right) \label{matrixEOM} \end{eqnarray}
167: \noindent with $E_M$ being the $M \times M$ unit matrix. A critical point $\vec{X}_0$ in the reduced phase space is defined by $2M$ simultaneous equations:
168: \begin{eqnarray}
169: \left( \frac{\partial H}{\partial x_i} \right)_{X_0} = 0 \label{GeneralCP}
170: \end{eqnarray}
171:
172: The linear stability of $\vec{X}_0$ is defined by the behavior of the {\it linearized} equations of motion in the neighborhood. Let the point be
173: \begin{eqnarray}
174: \vec{X}= \vec{X}_0 + \{ d x_1, \ldots, d x_i,\ldots, d x_{2M} \} = \vec{X}_0+ d \vec{X}
175: \end{eqnarray}
176: \noindent The linearized equations of motion are obtained by expanding ${\partial H}/{\partial x_i}$ on the right hand side of eqn. (\ref{matrixEOM}) into a Taylor series, and keeping only terms linear to the displacement
177: \begin{eqnarray}
178: \left( \frac{\partial H}{\partial x_i} \right)_{X} = \left(\frac{\partial H}{\partial x_i} \right)_{X_0} + \sum_j \left( \frac{\partial^2 H}{\partial x_i \partial x_j} \right)_{X_0} d x_j = \sum_j \left( \frac{\partial^2 H}{\partial x_i \partial x_j} \right)_{X_0} d x_j \label{Taylor}
179: \end{eqnarray}
180: \noindent Then eqn. (\ref{matrixEOM}) is reduced to the linearized form:
181: \begin{eqnarray}
182: \frac{d}{dt} \vec{X} = \left( \begin{array}{cc} 0 & -E_M \\ E_M & 0 \end{array} \right) \left( \frac{\partial^2 H}{ \partial x_i \partial x_j } \right)_{X_0} d \vec{X} = A \cdot d \vec{X} \label{linearEOM}
183: \end{eqnarray}
184: \noindent which is a set of homogeneous ordinary differential equations. The standard procedure of solving them requires first finding the $2M$ eigenvalues $\lambda_i$ and eigenvectors $\vec{V}_i$ of matrix $A$ \cite{Diff}. The $\lambda_i$ and their respective $\vec{V}_i$ satisfy
185: \begin{eqnarray} A \cdot \vec{V}_i = \lambda_i \ \vec{V}_i \end{eqnarray}
186: \noindent If none of the $\lambda_i$ is zero, the solutions have the following form:
187: \begin{eqnarray} \vec{X}(t)= \vec{X}_0 + \sum_{i=1}^{2M} a_i \mbox{\,\,} e^{\lambda_i t} \mbox{\,\,} \vec{V}_i \label{linearEOMsolution} \end{eqnarray}
188: \noindent With $a_i$ being arbitrary complex coefficients. The time evolution of $\vec{X}(t)$ therefore is separable into $2M$ directions, each indicated by the vector $\vec{V}_i$.
189: %% There are additional constraints of a_i so that the trajectories are all real.
190:
191: The linear stability of $\vec{X}_0$ is defined in terms of eqn. (\ref{linearEOM}), through the eigenvalues $\lambda_i$. In a Hamiltonian system, the conservation of phase space volume (Liouville's theorem) leads to the result that $\lambda_i$ always appear in the form of conjugate quadruplets ($\pm a \pm b i$), for which there are four cases described below.
192:
193: \noindent \,\,\, {\bf i.} \,\,\, When a pair of $\lambda_i$ is purely imaginary ($a=0$), all solutions in eqn. (\ref{linearEOMsolution}) would oscillate in the subspace spanned by $\vec{V}_i$ with a characteristic frequency determined by $\vert \lambda_i \vert$. The linear stability in this direction is known as {\it stable}, {\it elliptic} or (E).
194:
195: \noindent \,\,\, {\bf ii.} \,\,\, When a pair of $\lambda_i$ is real ($b=0$), in the $\vec{V}_i$ subspace all solutions in eqn. (\ref{linearEOMsolution}) would be attracted to or repelled from $\vec{X}_0$ exponentially with time. This direction is known as linearly {\it unstable}, {\it hyperbolic} or (H). The names elliptic and hyperbolic originated from the shape of these linearized trajectories (Fig.~\ref{EH}).
196:
197: \noindent \,\,\, {\bf iii.} \,\,\, When $a \neq 0, b \neq 0$, the solution contains {\it both} oscillating and exponential attraction/repulsion components in the subspace spanned by the four $\vec{V}_i$ corresponding to the quadruple $\lambda_i$. In two of the four directions the nearby trajectory ``spirals" into the critical point, while in the other two directions it ``spirals" out of the critical point. This stability type is called {\it mixed} or (M) \cite{Zi-MinH2O1}.
198:
199: \noindent \,\,\, {\bf iv.} \,\,\, When a pair of $\lambda_i = 0$, the stability type is degenerate (D). In this case, the linearized equations eqn. (\ref{linearEOM}) become insufficient, and higher-order terms in the Taylor expansion are needed to evaluate the stability near a critical point.
200:
201: If all the eigenvalues fall into category {\bf i.}, then the linearized trajectories defined by eqn. (\ref{linearEOMsolution}) oscillate with $M$ distinctive frequencies. {\it Hence, near an all-stable critical point, the linearized equations of motion are quasiperiodic.} These linearized trajectories are expected to resemble the trajectories of the nonlinear Hamiltonian $H_{eff}$ for at least a finite time.
202:
203: %Even in a non-integrable system, it is still possible to approximate the dynamics using these quasiperiodic motions, when they do not exactly match the actual classical trajectories \cite{ShirtsReinhardt}
204:
205: \subsection{3.3.2 The Presence of Multiple Cyclic Angles}
206: \addtocontents{toc}{\protect\vspace*{5pt}}
207: % == MEK: Expand upon below == %
208:
209: At a critical point, all the canonical variables are fixed except the ($N-M$) cyclic angles $\theta_i$. When there is more than one polyad number, the trajectory does not close onto itself within a finite time. Otherwise, unless any of their frequencies becomes zero or commensurable with another, the full phase space trajectory is quasiperiodic and restricted to an ($N-M$) dimensional invariant torus. Fig.~\ref{fullphasespace} illustrates the case with $N=2, M=1$, which is integrable. The reduced phase space ($J, \Psi$) is a projection of the full phase space ($J, \Psi, \theta$). Critical points in ($J, \Psi$) trace out PO in the full phase space (blue line). A trajectory near the stable critical point (green oval at bottom) is a PO in the reduced phase space, and a quasiperiodic motion in the full phase space.
210:
211: \newpage \begin{figure}[hbtp]
212: %==\vspace{5.06in}
213: \begin{center}\includegraphics[width=5.13in]{slice.eps}\end{center}
214: \cpn{Dynamics in the reduced and full phase spaces} {Dynamics in the reduced and full phase spaces, a schematic illustration. The reduced Hamiltonian is taken to resemble Fig.~\ref{dbwellbifur} in action-angle variables ($J, \Psi$). The angle $\theta$ is the cyclic angle, which evolves (modulo $2\pi$) between $\lbrack 0, 2\pi \rbrack$. \label{fullphasespace}}
215: \end{figure} \newpage
216:
217: Intuitively, the role these critical points play in the phase space should not change whether there are one or more cyclic angles. As an example, consider the case of HCP where only 2 of the 3 normal modes are coupled by a Fermi resonance \cite{HCP}. Excitation in the spectator mode $3$ (C-H stretching) can be treated as a parameter in the effective Hamiltonian. So strictly speaking, there are {\it two} polyad numbers and cyclic angles:
218: \begin{align}
219: P_1 & =n_1 + \frac{n_2}{2}, & \theta_1 &= 2\phi_1+\phi_2 \\
220: P_2 & =n_3, & \theta_2 &= \phi_3
221: \end{align}
222: \noindent Yet, in assigning e.g. the $n_3=1$ states, one could simply use the critical points found in this manifold, in spite of the fact that the frequency of $\theta_2$ is not zero in these polyads.
223:
224: We argue that the cyclic angles $\theta_i$ in general represent a trivial aspect of the dynamics. As far as quantum assignment is concerned, this is evident if one considers the semiclassical quantization procedure. There $\theta_i$ appear only in a pre-factor with the form $\Pi e^{i P_i \theta_i}$ in the resulting wavefunctions \cite{TaylorCHBrClF}. In classical mechanics, also note that $\theta_i$ do not have physical meaning on their own, since the polyad number $P_i$ are not uniquely defined (Appendix A).
225: % although it may change the nature of classical motion (periodic versus quasiperiodic).
226: %The resulting wavefunction only contains the angle variables $\Psi_i$, while the action variables are averaged out in the process.
227:
228: %% relative equilibria crap here
229:
230: Critical points in a reduced phase space, especially those with non-zero frequencies in the cyclic coordinate, have been known as {\it relative equilibria} in mathematical literature \cite{Marsden,REDyn}. Near a relative equilibrium, classical dynamics in the full phase space can be separated into two parts: the {\it group orbit}, which is motion along the cyclic angles, and motion in the reduced phase space \cite{Gaeta}. The latter is a multidimensional ``slice" transverse to the group orbit \cite{REHamiltonian}. The slice contains all the ``essential dynamics" \cite{Marsden}, in the sense that the full dynamics can be reconstructed from a point on the slice and appropriate initial conditions. This provides a further argument against making a distinction between systems with one and multiple polyad numbers.
231:
232: %The Hamiltonian Noether Theorem indicates that each constant of motion generates a group of infinitesimal transformations under which the Hamiltonian is invariant (Chapter 5.3 of \cite{Jose}) . For example, the conservation of energy and linear, angular momentum result from the homogeneity of time, space and the isotropy of space, respectively.
233: %For Noether's Theorem also see Chapter 12.7 of \cite{Goldstein}
234:
235: In the field of chemistry, relative equilibria theory has been used to classify rotationally excited molecular spectra \cite{REZhilinskii,RERoberts}. The total angular momentum $J$ plays the same role as the polyad numbers in this thesis. A stable relative equilibrium corresponds to the molecule rotating with a fixed shape. Vibrational modes are defined by the {\it normal form} of the Hamiltonian in the neighborhood. As $J$ is increased, the bifurcations of relative equilibria correspond to predictions of the (as yet unobserved) rovibrational spectral patterns.
236:
237: \subsection{3.3.3 Semiclassical Localization Near Critical Points}
238: \addtocontents{toc}{\protect\vspace*{5pt}}
239:
240: An eigenstate may be assigned meaningful quantum numbers based on the critical point, if its representation in the same ($\Psi_i, J_i$) space is localized near the critical point with a well-ordered nodal pattern. The semiclassical wavefunctions can be obtained through either phase space representations (e.g. the Wigner or Husimi function \cite{Wigner}), or the $\Psi_i$ space {\it semiclassical quantization} proposed by Voth and Marcus in \cite{Marcus1985}. Examples from both methods are illustrated in Fig.~\ref{localization}. In panels (a) and (b), the localization occurs around the normal mode critical points at ($n_1=n_2=3.5, \psi_1=\frac{\pi}{2}$ and $\frac{3\pi}{2}$). In panels (c) and (d), the localization is around the critical points at ($\psi_a = \pm \pi, \psi_b=\pm \pi$).
241:
242: \newpage \begin{figure}[hbtp]
243: %==\vspace{2.0in}
244: \begin{center}\includegraphics[width=4.55in]{lulocalization.eps}\end{center}
245: %==\vspace{2.514in}
246: \begin{center}\includegraphics[width=4.55in]{jacobsonlocalization.eps}\end{center}
247: \cpn{Semiclassical localization near critical points} {Semiclassical localization near critical points. Panels (a), (b) are reproduced from Fig.~3 of \cite{Zi-MinH2O2}, displaying the two projections of the Husimi function of the same eigenstate in action ($n_1, n_2$) and angle ($\psi_1, \psi_2$) space, respectively. Panels (c) and (d) are reproduced from Fig.~5 of \cite{Jacobson15000}, which display the angle-space representation of two different wavefunctions both localized around ($\psi_a = \pm \pi, \psi_b=\pm \pi$). \label{localization}}
248: \end{figure} \newpage
249:
250: Consider a local minimum or maximum (together referred to as {\it extremum}) in the reduced phase space ($\Psi_i, J_i$). This extremum point is necessarily a critical point. Then if there is a quantum eigenstate whose energy is nearby, intuitively one expects the semiclassical representation of the eigenstate in either ($\Psi_i, J_i$) or ($\Psi_i$) space to localize near the critical point, simply because of the limited volume of accessible phase space into which it can expand. This is illustrated schematically in Fig.~\ref{localillu} in the case of a minimum.
251:
252: \newpage \begin{figure}[hbtp]
253: %==\vspace{3.28in}
254: \begin{center}\includegraphics[width=3.94in]{localization.eps}\end{center}
255: \cpn{Localization near a minimum in the reduced phase space} {Localization near a minimum in the reduced phase space, which is also an elliptic critical point of the Hamiltonian. An eigenstate whose energy is close to the minimum must be localized in the nearby phase space. \label{localillu}}
256: \end{figure} \newpage
257:
258: There is no apparent reason why the same argument should not be valid for all choices of semiclassical representation as well as for arbitrary DOF, except for the following two scenarios. The localization may be disrupted by quantum tunneling when the local extremum is not sufficiently prominent, or there are other local extrema nearby with similar energy.
259:
260: In $\S$ 3.3.2, it was shown that the linearized motion near an all-stable critical point is quasiperiodic. If (1) it is a good approximation for the real classical trajectories in this region and (2) the region is large enough to support one quantizing invariant torus, then the semiclassical wavefunctions may be localized around the torus, with $M$ quantum numbers assigned by EBK quantization. Therefore, one could expect the all-stable critical points to correspond to quantum modes around which semiclassical wavefunctions localize. Other critical points with partial linear stability (while unstable in some directions) may also become the center of localization under favorable circumstances.
261:
262: \subsection{3.3.4 Summary}
263: \addtocontents{toc}{\protect\vspace*{5pt}}
264:
265: From $\S$ 3.3.1 - 3.3.3, we can draw the following conclusions about critical points in the reduced phase space:
266:
267: \begin{enumerate}
268: \item Along the stable directions of a critical point, the linearized classical trajectories nearby are quasiperiodic.
269: \item The presence of multiple cyclic angles is not expected to affect the essential part of the classical dynamics or semiclassical localization.
270: \item If a critical point is also a local extremum in the reduced phase space of $H_{eff}$, then it is expected to be a center of localization for semiclassical eigenfunctions.
271: \end{enumerate}
272:
273: Therefore, the critical points can be used to assign vibrational modes to the quantum spectra. The change in their number and/or stability should correspond to the change in birth, death and transformations of the vibrational modes.
274:
275: % Marsden: page 304: ``The Hamiltonian system induced on the reduced space represents, in a sense, the ``essential dynamics"; the explicitly known dynamics is factored out in the reduction process."