0810.1019/D2.tex
1: 
2: 
3: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
4: \chapter{Conservative mechanics on manifolds}\label{c.pmani}
5: 
6: 
7: \bigskip
8: \at{adapt this introduction, and also the summary in Section 1.5 
9: of QML}
10: 
11: We consider closed 2-forms in manifolds and their associated Poisson 
12: algebras. This naturally leads to symplectic geometry and a symplectic 
13: formulation of the dynamics of quantum mechanics.
14: It also leads to classical Hamiltonian and Lagrangian mechanics,
15: including constraints.
16: 
17:  
18: 
19: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
20: \section{Poisson algebras from closed 2-forms}\label{s.sympl}
21: 
22: In general, a classical, conservative dynamical system is
23: described in terms of motion on a manifold $\Mz $, called the 
24: {\bfi{phase space}}, such that some algebra of 
25: functions on it has a Poisson algebra structure; more precisely,
26: $C^\infty(\Mz )$ is
27: equipped with a Lie product $\lp$ that is antisymmetric and satisfies
28: the Jacobi identity and the Leibniz identity. Such a manifold is
29: called a {\bfi{Poisson manifold}}; see, e.g., 
30: \sca{Vaisman} \cite{Vai} or \sca{da Silva \& Weinstein} \cite{daSW} 
31: Poisson manifolds provide a
32: general setting for the study of the dynamics of classical
33: conservative mechanical systems by differential geometric methods;
34: for a more comprehensive discussion of different aspects see 
35: \sca{Marsden  \& Ratiu} \cite{marsdenratiu}, 
36: \sca{Ratiu} \cite{courseratiu} and
37: \sca{Morrison} \cite{morrison98}.
38: 
39: Every symplectic manifold is a Poisson manifold since the symplectic
40: structure gives rise to a natural Poisson bracket. 
41: In the symplectic case, a Hamiltonian is a function of some
42: coordinates $q_i$ and the conjugated momenta $p_i$. In such cases, the
43: phase space is even-dimensional. In more general cases described, 
44: e.g., by Lie-Poisson algebras, the phase space need not be a symplectic 
45: manifold. Indeed, symplectic manifolds are always even-dimensional 
46: while the manifold $SO(3)$ of the spinning top (see Section 
47: \ref{s.obs.rot} and Section \ref{s.rigid}) \at{combine these?}
48: has dimension 3.
49: 
50: \bigskip
51: Many Poisson algebras of relevance in classical mechanics may be 
52: constructed via a uniform construction based on a 
53: closed 2-form characterizing the kinematics of the system of interest.
54: The description is then completed by specifying the dynamics through
55: a Hamiltonian in the resulting Poisson algebra, and by selecting an 
56: initial state describing the preparation of the system.
57: In this section, we discuss the general construction principle.
58: 
59: Let $\omega$ be a closed 2-form on a differential geometry $\Ez$.
60: We call a scalar field $f\in \Ez$ \bfi{compatible} with
61: $\omega$ if there is a vector field $X_f$ such that\MMt{\footnote{
62: Equation \gzit{e.obs} can be
63: understood better via the "stick" tensor type notation. The left hand
64: side $df$ corresponds to "$|\cdot$" (covector times scalar), and the
65: right hand side $X_f \omega$ corresponds to "\_$||$" (rovector
66: contracted with a bilinear form). Both side are therefore a covector = 
67: linear form.
68: } % end footnote
69: } % end MMt
70: \lbeq{e.obs}
71:    df ~=~ X_f \, \omega ~;
72: \eeq
73: any such $X_f$ is called a \bfi{Hamiltonian vector field} 
74: associated with $f$.
75: \at{Note somewhere that $\ad_f=X_fd$.}
76: We write $\Ez(\omega)$ for the set of all scalar fields $f\in \Ez$ 
77: which are compatible with $\omega$.
78: In general, $X_f$ need not exist for all $f$, and if it 
79: exists, it need not be unique. Thus $\Ez(\omega)$ may be a 
80: proper subspace of $\Ez$; this situation is typical for examples
81: arising from constrained Hamiltonian mechanics.
82: 
83: \begin{prop}
84: Let $\omega$ be a symplectic form. Then every scalar field $f$ is 
85: compatible with $\omega$, 
86: \lbeq{e.comp}
87: X_f=df\omega^{-1},
88: \eeq
89: and $\Ez(\omega)=\Ez$.
90: 
91: \end{prop}
92: 
93: \bepf
94: Since $\omega$ is a symplectic form, $\omega$ is 
95: nondegenerate and has an inverse satisfying \gzit{e.Ainv}.
96: The defining condition for $X_f$ can therefore be solved uniquely 
97: for $X_f$, for all $f\in \Ez$, resulting in \gzit{e.comp}.
98: \epf
99: 
100: A vector field $ X$ is called \bfi{locally Hamiltonian} 
101: (with respect to $\omega$) if the linear form $ X\omega$ is closed, 
102: and \bfi{Hamiltonian} (with respect to $\omega$) if $ X\omega$ is 
103: exact (and hence closed). Thus, for any $f\in\Ez(\omega)$, the vector 
104: field $X_f$ is Hamiltonian with respect to $\omega$,
105: 
106: \begin{prop}\label{p4.1}
107: If $X,Y$ are locally Hamiltonian vector fields with respect to
108: the closed 2-form $\omega$ then $X\lp Y$ is Hamiltonian,
109: and
110: \lbeq{e.locham}
111: (X \lp Y)\omega = d(XY\omega).
112: \eeq
113: In particular, the locally Hamiltonian vector fields and the
114: Hamiltonian vector fields form Lie subalgebras of $\Wz$.
115: \end{prop}
116: 
117: \bepf
118: Since $\omega$ and $X\omega$ are closed, \gzit{e.ld22} implies
119: that $L_X\omega=Xd\omega+d(X\omega)=0$. Again by \gzit{e.ld22},
120: $d(XY\omega) = L_X(Y\omega)-Xd(Y\omega) =  L_X(Y\omega)
121: = (L_XY)\omega+YL_X\omega=(L_XY)\omega=(X\lp Y)\omega$,
122: using the closedness of $Y\omega$ and the product rule \gzit{e.ld3}.
123: This proves \gzit{e.locham}. The concluding statement is an immediate 
124: consequence.
125: \epf
126: 
127: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
128: \begin{thm} \label{t2.2}
129: For every closed 2-form $\omega$ over the manifold $ \Mz $, the set
130: $\Ez(\omega)$ is a Poisson algebra, with Lie product given by
131: \lbeq{e.liep}
132:    f \lp g := X_f dg = X_fX_g\omega  = -X_gX_f\omega  = -X_g df.
133: \eeq
134: A Hamiltonian vector field associated with $f\lp g$ is given by 
135: \lbeq{e.hampl}
136:   X_{f \slp g}:=X_f\lp X_g.
137: \eeq
138: In particular, if $\omega$ is a symplectic form then 
139: \lbeq{e.lies}
140: f\lp g =df\omega^{-1}dg.
141: \eeq
142: \end{thm}
143: 
144: \bepf
145: We first show that $\Ez(\omega)$ is a subalgebra of the algebra 
146: $\Ez$. 
147: If $f,g\in\Ez(\omega)$ and $\lambda\in \Cz$ then $\lambda f, f\pm g,
148: fg \in\Ez(\omega)$ since we may take
149: \[
150:    X_{\lambda f} = \lambda X_f,~~~X_{f \pm g} = X_f \pm X_g,~~~
151:  X_{fg} = f X_g + gX_f.
152: \]
153: We next show that $f \lp g$ is well-defined. Indeed, if $X_f,X_f'$ 
154: are two Hamiltonian vector fields associated with $f$ then
155: $\omega(X_f'-X_f,Y)=0$, hence $f \lp g$ does not depend on the
156: choice of the Hamiltonian vector fields associated with $f$ and $g$.
157: 
158: Proposition \ref{p4.1} implies that \gzit{e.hampl}
159: is a Hamiltonian vector field for $f\lp g$; therefore
160: $f\lp g\in\Ez(\omega)$.
161: 
162: The operation $\lp$ defined by \gzit{e.liep} is bilinear, 
163: antisymmetric, and satisfies the Leibniz identity.
164: To conclude that $\Ez(\omega)$ is a Poisson algebra it therefore 
165: suffices to show that the Jacobi identity holds. This follows 
166: since, with $X:=X_f,Y:=X_g$,
167: \[
168: \bary{lll}
169: (f \lp g) \lp h &=& X_{f\slp g} d h = (X_f \lp X_g) d h
170: = (X\lp Y) d h = L_{X\slp Y} h \\
171: &=& [L_X,L_Y] h
172: = L_X L_Y h - L_Y L_X h 
173: = X_f d (X_g d h) -  X_g d (X_f d h)\\
174: &=& f\lp (g \lp h) - g\lp (f \lp h)
175: = (f\lp h)\lp g + f\lp (g \lp h).
176: \eary
177: \]
178: Finally, if $\omega$ is a symplectic form, \gzit{e.comp} implies that 
179: the Lie product \gzit{e.liep} can be rewritten in the form 
180: \gzit{e.lies}.
181: \epf
182: 
183: Note that the Lie product can be extended to the case where one argument
184: is in $\Ez(\omega)$ and the other may be an arbitrary quantity from 
185: $\Ez$:
186: \[
187: f \lp g = X_f dg \for f\in \Ez(\omega),~g\in\Ez,
188: \]
189: \[
190: f \lp g = - X_g df \for f\in \Ez,~g\in\Ez(\omega).
191: \]
192: Thus if $f$ is compatible with $\omega$, the Lie product is defined 
193: even when $g$ is not compatible with $\omega$.
194: 
195: 
196: 
197: \bigskip
198: In the manifold case, the above theorem defines, for each
199: closed 2-form $\omega$ on an $\Fz$-manifold $\Mz $, a Poisson algebra 
200: $\Ez(\omega)$ which is the set of functions $f \in \Ez= C^\infty(\Mz )$ 
201: which are compatible with $\omega$.
202: In the special case where $\omega$ is symplectic, we have seen that
203: $\Ez(\omega)=\Ez$; thus we may define the \bfi{Poisson bracket} 
204: \lbeq{e.pbracket}
205: \{f,g\}:= g \lp f = dg\omega^{-1}df
206: \eeq
207: of $f,g\in \Ez$. This is the traditional Poisson bracket
208: associated with the \bfi{symplectic space} $(\Mz ,\omega)$.
209: 
210: The affine functions, which map $\xi\in \Mz $ to $\xi u+\gamma$
211: for some $u\in \Fz$ and some $\gamma\in\Cz$, satisfy
212: \[
213: (\xi u +\gamma) \lp (\xi v +\gamma') = u \omega{-1} v \in \Cz,
214: \]
215: hence form a Lie subalgebra, which is a Heisenberg algebra.
216: This provides a faithful classical Poisson representation of general
217: Heisenberg algebras.
218: 
219: 
220: \begin{expl}~
221: We continue the discussion of Example \ref{ex.Rn}, where scalar fields
222: (resp. vector fields) are the smooth 
223: complex-valued (resp. row vector valued) functions
224: on a nonempty, open subset $\Mz $ of the space $\Rz^{\times n}$ 
225: of rovectors of length $n$. In this case, it is natural to identify
226: $\Wz^*$ with the vector space $C^\infty(\Mz ,\Cz^n)$ of covector-valued 
227: fields via
228: \[
229: (X\zeta)(x)=X(x)\zeta(x)
230: \]
231: for $X\in \Wz= C^\infty(\Mz ,\Cz^{\times n})$ and 
232: $\zeta\in\Wz^*= C^\infty(\Mz ,\Cz^n)$. In particular, the gradient 
233: $df=\partial f$ appears naturally as an element of $\Wz^*$, consistent 
234: with our abstract development. 
235: 
236: Now let $\theta$ be a distinguished linear form. Then we can define its 
237: \bfi{Jacobian}, 
238: the $x$-dependent square array\MMt{\footnote{
239: an object of type $||$ in the stick notation
240: }% end footnote 
241: } % end MMt
242: $\partial \theta$ whose entries are the partial derivatives 
243: \[
244: \partial_j\theta_k(x) ~=~ \Pdrv{\theta_k(x)}{x^j} ~.
245: \]
246: We now consider the exact 2-form $\omega=-d\theta$ (the minus
247: sign is traditional). We have
248: \lbeq{e.dtheta}
249: YX\omega = Yd(X\theta)-Xd(Y\theta)+(X\lp Y)\theta \for \omega=-d\theta
250: \eeq
251: since $YX\omega = -YXd\theta=-Y(L_X\theta-d(X\theta))
252: =-YL_X\theta+Yd(X\theta) = -L_X(Y\theta)+(L_XY)\theta + Yd(X\theta)$
253: by \gzit{e.ld22b} and \gzit{e.ld21}.
254: It is not difficult to show that now
255: \lbeq{e.omega1}
256: (X\omega)(x) = \sum X^j(x)\omega_{jk}(x),
257: \eeq
258: where 
259: \lbeq{e.omega2}
260: \omega_{jk}(x) = \partial_k\theta_j(x)-\partial_j\theta_k(x)
261: \eeq
262: are the components of the antisymmetric expression 
263: $(\partial \theta)^T-\partial \theta$ in the Jacobian of $\theta$.
264: 
265: As a consequence, $\omega=-d\theta$ is nondegenerate precisely when 
266: $(\partial \theta)^T-\partial \theta$ is nonsingular.
267: If this holds, $\omega$ is a symplectic form, and all our results apply.
268: This matches the present development with that found in standard 
269: treatises such as \sca{Marsden \& Ratiu} \cite{MarR}.
270: 
271: \end{expl}
272: 
273: 
274: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
275: \section{Conservative Hamiltonian dynamics}\label{s.hamphase}
276: 
277: We now apply the results of Section \ref{s.sympl} to 
278: classical Hamiltonian mechanics of conservative systems. 
279: The phase space of a classical system 
280: is the set of all states that may be attained in some realization of 
281: the system. We begin with the unconstrained case, 
282: where the phase space is a cotangent bundle over a manifold $\Mz $, 
283: and than extend the discussion to the constrained case, where the 
284: phase space has a more complicated structure.
285: 
286: To avoid technicalities, we only treat the case where the manifold
287: can be described by a single chart, so that it can be treated as an
288: open subset of some topological vector space. However, using standard
289: techniques from differential geometry, it is not difficult 
290: to lift the discussion to arbitrary manifolds. Thus, in the following,
291: the \bfi{configuration space}
292: $\Mz _c$ is a nonempty, open subset of a convenient vector space $\Fz$
293: over $\Rz$. Thinking of $\Mz _c$ as a chart of a general manifold, 
294: everything we say here extends in a standard way to arbitrary 
295: $\Fz$-manifolds in place of $\Mz _c$.
296: 
297: We write the bilinear pairing between elements $q$ from $\Fz$ and 
298: elements $p$ from the dual space $\Fz^*$ as product 
299: $p\cdot q = q \cdot p$.
300: We extend this product linearly to the compexifications $\Cz\Fz$
301: of $\Fz$ and $\Cz\Fz^*$ of $\Fz^*$, and extend it further pointwise to 
302: $\Cz\Fz$-valued or $\Cz\Fz^*$-valued functions.
303: 
304: 
305: In this section, we consider the case of unconstrained dynamics
306: Here $\Ez=C^\infty(\Mz )$ and $\Wz=C^\infty(\Mz ,\Cz\Fz\times\Cz\Fz^*)$ 
307: are the spaces of scalar fields and vector fields, respectively, on the 
308: \bfi{cotangent bundle} $\Mz  = T^*\Mz _c:=\Mz _c\times\Fz^*$ of $\Mz_c$.
309: 
310: The reader may think of the Euclidean space $\Fz=\Fz^*=\Rz^n$ of 
311: vectors with $n$ real components and bilinear pairing 
312: $p\cdot q = \sum_k p_k q_k$.  As discussed in Section 
313: \ref{s.cao}, this accounts for the mechanics of point particles.
314: For field theories, $\Fz$ is an infinite-dimensional function space.
315: 
316: A classical, conservative, \bfi{unconstrained mechanical system} 
317: is defined by a \bfi{Hamiltonian} $H\in\Ez$ and considering the 
318: full cotangent bundle $\Mz $ as the \bfi{phase space} of the system.
319: The point $x =(q,p)\in \Mz $ is called the \bfi{state} 
320: with \bfi{position} $q\in \Mz _c$ and \bfi{momentum} $p\in\Fz^*$.
321: The \bfi{energy} of the system in the state $(q,p)$ is 
322: the value $H(q,p)$ of the Hamiltonian at $(q,p)$. 
323: \at{The Hamiltonian of a physically realistic 
324: system must be bounded below.}
325: 
326: The state of the system varies with \bfi{time} $t$, which we 
327: consider to be a number in the interval $[\ul t,\ol t]$,
328: where $\ul t$ is the \bfi{initial time} and $\ol t>\ul t$ is the 
329: \bfi{final time} for which the system is considered. 
330: The time dependence is modeled by a \bfi{trajectory}, a state-valued, 
331: arbitrarily often differentiable function of time, mapping 
332: $t\in[\ul t,\ol t]$ to $(p(t),q(t))\in \Mz $. The position $q(t)$ 
333: and the momentum $p(t)$ at time $t$ are constrained by the
334: \bfi{Hamilton equations} in \bfi{state form},
335: \lbeq{e.hamc}
336:   \dot q = \frac{dq}{dt} = \partial_p H,~~~ 
337:   \dot p = \frac{dp}{dt} = -\partial_q H.
338: \eeq
339: Here $\partial_p = \partial/\partial p$ and 
340: $\partial_q = \partial/\partial q$ denote the gradient with respect 
341: to momentum $p$ and position $q$, respectively. Note that if $f$ is a 
342: scalar field then $\partial_p f(q,p)\in\Cz\Fz$ and 
343: $\partial_q  f(q,p)\in \Cz\Fz^*$.
344: 
345: The Hamiltonian equations automatically imply
346: the \bfi{conservation of energy}:
347: $\frac{d}{dt} H(q,p)
348: =\partial_q H \cdot \dot q + \partial_p H \cdot \dot p = 0$.
349: 
350: The Hamiltonian equations may be derived from a \bfi{variational 
351: principle}. We define the \bfi{action} as the functional on smooth 
352: paths in $\Mz$ defined by 
353: \lbeq{e.action}
354: I(q,p) := \int_{\ul t}^{\ol t} dt~ 
355: \Big(p(t)\cdot \dot q(t) - H(q(t),p(t))\Big),
356: \eeq
357: and consider small variations $\delta q$ and $\delta p$ of the arguments
358: $q$ and $p$, respectively. Since we do not make further 
359: use of the principle, we assume without the discussion that the 
360: integral can be manipulated as accustomed from the finite-dimensional 
361: case, where $\Fz=\Rz^n$.
362: \at{The action needs a Banach space setting. Treat only the flat case
363: since this is only motivation!
364: later add a boundary term 
365: $I_b(q(\ul t),q(\ol t),\dot q(\ul t),\dot q(\ol t))$
366: and discuss the resulting boundary conditions?
367: In the case without boundary conditions, apparently the 
368: variation $h$ must be required to vanish at the end points?}
369: For variations vanishing
370: at $t=\ul t$ and $t=\ol t$, we have, up to higher order terms,
371: \[
372: \bary{lll}
373: I(q+\delta q,p) - I(q,p) 
374: &\approx& \D\int_{\ul t}^{\ol t} dt~ 
375: \Big(p(t)\cdot \delta \dot q(t)
376: -\partial_q H(q(t),p(t))\cdot \delta q\Big)\\
377: &=& \D\int_{\ul t}^{\ol t} dt~ 
378: \Big(-\dot p(t)\cdot \delta q(t)
379: -\partial_q H(q(t),p(t))\cdot\delta q\Big),
380: \eary
381: \]
382: \[
383: I(q,p+\delta p) - I(q,p) \approx  \int_{\ul t}^{\ol t} dt~ 
384: \Big(\delta  p(t)\cdot \dot q(t)
385: -\partial_p H(q(t),p(t))\cdot\delta p\Big),
386: \]
387: so that the path $(q,p)$ is a stationary point of the action if
388: and only if the extended Hamiltonian equations \gzit{e.hamc} hold.
389: 
390: \bigskip
391: A vector field $X\in\Wz$ is a pair of functions $X=(X^q,X^p)\in 
392: C^\infty(\Mz ,\Cz\Fz)\times C^\infty(\Mz ,\Cz\Fz^*)$; 
393: its value at the state $(q,p)$ is $X(q,p)=(X^q(q,p),X^p(q,p))$. 
394: Associated with each vector field $X$ is the derivation $Xd$ defined by
395: \[
396: Xdf := X^q\cdot \partial_q f + X^p \cdot \partial_p f.
397: \]
398: Using the mapping $d$ defined in this way, it is easily checked that
399: we have a commutative differential geometry.
400: In particular, a general linear form $\zeta$
401: is described by a pair of functions $(\zeta_q,\zeta_p)\in 
402: C^\infty(\Mz ,\Cz\Fz^*)\times C^\infty(\Mz ,\Cz\Fz)$
403: such that
404: \lbeq{e.zeta}
405: X\zeta = X^q\cdot\zeta_q  + X^p \cdot \zeta_p.
406: \eeq
407: 
408: %%%%%%%%%%%%%%%%%%%%%%%
409: \begin{thm}
410: Let $\theta = (p,0)$ be the linear form defined by
411: \lbeq{e.thetap}
412: (X\theta)(q,p):=X^q(q,p) \cdot p ~.
413: \eeq
414: Then $\omega=-d\theta = \pmatrix{0 & -1 \cr 1 & 0}$ 
415: is an exact symplectic form satisfying
416: \lbeq{e.omegap}
417: (YX\omega)(q,p):=X^q(q,p)\cdot Y^p(q,p)-Y^q(q,p)\cdot X^p(q,p)
418: \eeq
419: for arbitrary vector fields $X,Y$. Its inverse 
420: satisfies
421: \lbeq{e.omegaq}
422: X=\zeta\omega^{-1} \iff X^q=\zeta_p,~~X^p=-\zeta_q.
423: \eeq
424: for arbitrary linear forms $\zeta$. With the Lie product
425: \lbeq{e.canpoisson}
426: f\lp g := df \omega^{-1} dg 
427: = \partial_p f\cdot \partial_q g - \partial_p g \cdot  \partial_q f,
428: \eeq
429: the algebra $\Ez=C^\infty(\Mz )$ of scalar fields on phase space $\Mz $ 
430: is a Poisson algebra. $\lp$, $\omega$, and $\theta$ are called the 
431: \bfi{canonical Lie product}, the \bfi{canonical symplectic form},
432: and the \bfi{canonical linear form}\footnote{
433: In the notation using components and the Einstein 
434: summation convention, we have $\theta = p_j dq^j$
435: and $\omega = dq^j \wedge dp_j$. Here the linear forms $dq^j$ and
436: $dp_j$, given by $Xdq^j:=(X^q)^j$ and $Xdp_j:=(X^p)_j$, are the
437: gradients of the functions $q^j$ and $p_j$ mapping a general state
438: $(q,p)$ to the indicated components.
439: } % end footnote
440: on phase space $\Mz $.
441: \end{thm}
442: \bepf
443: $\omega$ is an exact 2-form since $\omega=d(-\theta)$. To prove 
444: \gzit{e.omegap}, \at{check the following} 
445: we use  \gzit{e.dtheta} to work out 
446: $(YX\omega)(q,p)=Yd(X\theta)-Xd(Y\theta)+(X\lp Y)\theta
447: =Yd(X^qp)-Xd(Y^qp)+(X\lp Y)^qp
448: =Y^q\partial_q(X^qp)+Y^p\partial_p(X^qp)
449: -X^q\partial_q(Y^qp)-X^p\partial_p(Y^qp)+(X\lp Y)^qp$. 
450: Using $(X\lp Y)^q=X\partial Y^q-Y \partial X^q 
451: =X_q\partial_qY^q+X_p\partial_pY^q-Y_q\partial_q X^q-Y_p \partial_pX^q$,
452: which follows from \gzit{e.ldc}, 
453: the product rule, and $\partial_p p =1$, everything cancels except 
454: for $Y^pX^q-X^pY^q$. This proves \gzit{e.omegap}.
455: %
456: \MMt{
457: Write $X\theta$ as $X^i\theta_i = X^{q^{\mu}}p_\mu$.
458: \[
459: \bary{rcl}
460: Y^jX^i\omega_{ij}
461:  &=& Y^jX^i d_{[j}\theta_{i]}
462:      ~=~ Y^jX^i d_j\theta_i - Y^jX^i d_i\theta_j \\
463:  &=& Y^j X^{q^\mu} d_j(p_\mu) - Y^{q^\mu} X^i d_i(p_{\mu})
464:  	~=~  Y^{p_{\mu}}X^{q^{\mu}} - Y^{q^{\mu}}X^{p_{\mu}}
465: \eary
466: \]
467: }
468: By comparing \gzit{e.zeta} with \gzit{e.omegap}, we see that
469: \lbeq{e.zetaX}
470: \zeta=X\omega \iff \zeta_q=-X^p,~~\zeta_p=X^q.
471: \eeq
472: Thus $\omega$ maps $X=(X^q,X^p)$ to $(-X^p,X^q)$, corresponding to 
473: right multiplication by the matrix 
474: $\pmatrix{0 & -1 \cr 1 & 0}$.
475: Since the equations \gzit{e.zetaX} are uniquely solvable for $X$
476: by \gzit{e.omegaq}, we conclude that $\omega$ is nondegenerate, 
477: hence symplectic.
478: 
479: Since every exact 2-form is closed, Theorem \ref{t2.2} applies and 
480: gives the final assertion.
481: \epf
482: 
483: Using the chain rule, the dynamics \gzit{e.hamc} is easily seen to be 
484: equivalent to the \bfi{Hamilton equations} in \bfi{general form}, 
485: \lbeq{e.heg}
486: \dot f = \frac{df}{dt} = H \lp f;
487: \eeq
488: cf. Chapter \ref{s.cao}.
489: 
490: 
491: 
492: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
493: \section{Constrained Hamiltonian dynamics}\label{s.hamcon}
494: 
495: In the constrained case, additional parameters (e.g., Lagrange 
496: multipliers) are needed to describe the possible states of the system.
497: Therefore we take $\Ez=C^\infty(\Mz \times \Uz)$ and 
498: $\Wz=C^\infty(\Mz \times \Uz,\Cz\Fz\times\Cz\Fz^*\times\Cz\Uz)$, 
499: the spaces of scalar fields and vector fields, respectively, on an 
500: \bfi{augmented cotangent 
501: bundle} $\Mz \times \Uz$ of $\Mz _c$, where, as before, the phase space 
502: is $\Mz =\Mz _c\times \Fz$, and $\Uz$ is a convenient vector space.
503: 
504: A classical, conservative, \bfi{constrained mechanical system} 
505: is again defined by a \bfi{Hamiltonian} $H\in\Ez$. 
506: The point $x =(q,p,u)\in \Mz \times \Uz$ is called the \bfi{state} 
507: with \bfi{position} $q\in \Mz _c$, \bfi{momentum} $p\in\Fz$,
508: and \bfi{parameter} $u\in \Uz$; however, due to the constraints 
509: dervied below from $H$, not all points in $\Mz \times \Uz$
510: are physical. As we shall see, the accessible phase space may also be 
511: smaller than $\Mz $.
512: The \bfi{energy} of the system in the state $(q,p,u)$ is 
513: the value $H(q,p,u)$ of the Hamiltonian at $(q,p,u)$. 
514: 
515: The state of the system again varies with time $t\in [\ul t,\ol t]$.
516: The time dependence is modeled by a \bfi{trajectory}, a state-valued, 
517: arbitrarily often differentiable function of time, now defining 
518: position $q(t)$, momentum $p(t)$, and parameter $u(t)$ at time $t$.
519: These are constrained by the \bfi{extended Hamiltonian equations},
520: \lbeq{e.hamcu}
521:   \dot q = \frac{dq}{dt} = \partial_p H,~~~ 
522:   \dot p = \frac{dp}{dt} = -\partial_q H,~~~
523:   0 = \partial_u H.
524: \eeq
525: Here $\partial_u = \partial/\partial u$ denotes the gradient operator 
526: with respect to the parameter $u$. Thus, in place of a system of 
527: ordinary differential equations in the unconstrained case we now have
528: a system of  
529: \bfi{differential-algebraic equations} (\bfi{DAE}) involving the 
530: \bfi{holonomic constraints} 
531: \lbeq{e.constr}
532: 0 = \partial_u H(q,p,u).
533: \eeq
534: Again the extended Hamiltonian equations automatically imply
535: the \bfi{conservation of energy}:
536: $\frac{d}{dt} H(q,p,u)
537: =\partial_q H \cdot \dot q + \partial_p H \cdot \dot p
538: + \partial_u H \cdot \dot u = 0$. 
539: 
540: The case where the symmetric \bfi{Hessian matrix}
541: \[
542: G := \partial_u^2H(q,p,u)
543: \]
544: is invertible is referred to as the \bfi{regular} case.
545: Then, by the implicit function theorem, \gzit{e.constr} can be solved
546: locally uniquely for $u=u(q,p)$, which implies that \gzit{e.hamcu}
547: may be viewed as an ordinary differential equation in $q$ and $p$ alone.
548: In the \bfi{singular} case where the Hessian $G$ is not invertible, 
549: the constraints imply restrictions on $p$. Thus, not the whole phase 
550: space is dynamically accessible, and the analysis of solvability of 
551: the DAE is more involved. The details depend on the so-called 
552: \bfi{index} of a DAE, index 1 corresponding to the regular case, 
553: index $>1$ to the singular case, and are beyound our treatment.
554: 
555: \begin{expl}\label{ex.gauge}
556: We consider the constrained Hamiltonian system with $\Fz=\Rz^3$ and 
557: $\Uz=\Rz$, defined by the Hamiltonian
558: \[
559: H(\q,\p,u):= \half \p^2 + V(\k\times\q)-(\k\cdot\p)u,
560: \]
561: where $V(\E)$ is a potential energy function.
562: The special case $V(\E):=\half \E^2$, describes the dynamics of a 
563: single Fourier mode with wave vector $\k$ of the free electromagnetic 
564: field. A straightforward calculation gives the dynamics
565: \[
566: \dot\q = \p-\k u,~~~\dot\p = \k \times \Nabla V(\k\times\q),~~~
567: 0=\k\cdot\p.
568: \]
569: Since $G=\partial_u^2H=0$, this is a singular case. Indeed, the 
570: dynamically relevant part of the phase space is characterized by the 
571: \bfi{transversality condition} $0=\k\cdot\p$, wheras the multiplier $u$
572: is completely undetermined by the dynamics. This implies that the 
573: dynamics of $q$ is determined only up to an arbitrary multiple of $\k$;
574: in other wordss, only $\k\times\q$ is determined at all times by the 
575: initial conditions.
576: 
577: Note that $H\lp \k\cdot\p = \k\cdot(H\lp \p)=\k\cdot\p=0$,
578: hence the constraint $0=\k\cdot\p$ is automatically satisfied at all
579: times if it is satisfied at some time. Thus, in the terminology of 
580: constrained mechanics, it is called a \bfi{first class constraint},
581: and gives rise to \bfi{gauge symmetries}. 
582: A \bfi{gauge transform} replaces $\q$ by $\q+\k s(\q)$ with an 
583: arbitrary scalar field $s(\q)$, and leaves everything of fdynamical 
584: interest invariant. The \bfi{gauge invariant} quantities are those
585: in the centralizer $C(\k\cdot\p)$ of the constraint. $f$ belongs to the
586: centralizer iff it Lie commutes with $\k\cdot\p$, which is the case 
587: iff $\k\cdot\partial_q f=0$, hence iff $f$ depends only on $p$ and 
588: $\k\times\q$. Thus, the centralizer consists of all smooth functions of 
589: \[
590: \B:=\k\times \q,~~~\E:=-\p,
591: \]
592: and the dynamics of the gauge invariant quantities is determined by
593: \lbeq{e.BE}
594: \dot \B = - \k \times \E,~~~\dot\E=\k \times \nabla V(\B),~~~
595: 0 = \k\cdot\E.
596: \eeq
597: \end{expl}
598: 
599: The extended Hamiltonian equations may also be derived from a 
600: variational principle. Now the \bfi{action} is defined on smooth 
601: paths in $\Mz\times \Uz$,
602: \lbeq{e.actionu}
603: I(q,p,u) := \int_{\ul t}^{\ol t} dt~ 
604: \Big(p(t)\cdot \dot q(t) - H(q(t),p(t),u(t))\Big).
605: \eeq
606: Variations of the arguments show as before that 
607: the path $(q,p,u)$ is a stationary point of the action if
608: and only if the extended Hamiltonian equations \gzit{e.hamcu} hold;
609: the constraint equations derive from
610: \[
611: I(q,p,u+\delta u) - I(q,p,u) \approx  \int_{\ul t}^{\ol t} dt~ 
612: \Big(-\partial_u H(q(t),p(t),u(t))\cdot\delta u\Big).
613: \]
614: A vector field $X\in\Wz$ is now a triple of functions 
615: \[
616: X=(X^q,X^p,X^u)\in C^\infty(\Mz ,\Cz\Fz)\times 
617: C^\infty(\Mz ,\Cz\Fz^*)\times C^\infty(\Mz ,\Cz\Uz); 
618: \]
619: its value at the state $(q,p,u)$ is 
620: $X(q,p,u)=(X^q(q,p,u),X^p(q,p,u),X^u(q,p,u))$. 
621: Associated with each vector field $X$ is the derivation $Xd$ defined by
622: \[
623: Xdf := X^q\cdot \partial_q f + X^p \cdot \partial_p f 
624: + X^u \cdot \partial_u f.
625: \]
626: It is again easy to check that this defines a commutative differential 
627: geometry.
628: In particular, a general linear form $\zeta$
629: is described by a triple of functions $(\zeta_q,\zeta_p,\zeta_u)\in 
630: C^\infty(\Mz ,\Cz\Fz^*)\times C^\infty(\Mz ,\Cz\Fz^*)
631: \times C^\infty(\Mz ,\Cz\Uz)$
632: such that
633: \lbeq{e.czeta}
634: X\zeta = X^q\cdot\zeta_q  + X^p \cdot \zeta_p  + X^u \cdot \zeta_u.
635: \eeq
636: In analogy to the unconstrained case, we define the linear form 
637: $\theta$ by
638: \[
639: (X\theta)(q,p,u):=X^q(q,p,u) \cdot p ~.
640: \]
641: Thus $\theta=(p,0,0)$ and a similar calculation as before gives
642: the exact 2-form
643: \[
644: \omega:=-d\theta = \pmatrix{0 & -1 & 0 \cr 1 & 0 & 0 \cr 0 & 0 & )},
645: \] 
646: and 
647: \[
648: (YX\omega)(q,p,u)
649: :=X^q(q,p,u)\cdot Y^p(q,p,u)-Y^q(q,p,u)\cdot X^p(q,p,u).
650: \]
651: Since no differentiation by the parameters $u$ is involved, 
652: the 2-form $\omega$ is now degenerate and hence no longer 
653: symplectic. As a result, $\Ez(\omega)$ is strictly smaller than $\Ez$;
654: a scalar field $f$ is found to be compatible with $\omega$ and hence in 
655: $\Ez(\omega)$ only if $\partial_u f = 0$, i.e., $f$ is independent of 
656: $u$. Thus $\Ez(\omega)=C^\infty(\Mz )$ is again the Poisson algebra of 
657: scalar fields on phase space, with Lie product \gzit{e.canpoisson}.
658: 
659: The Hamilton equations \gzit{e.heg} remain valid, too; note that
660: by the general theory, $ H \lp f\in\Ez(\omega)$, although $H$ depends
661: on $u$.
662: 
663: If the Hamiltonian $H(q,p,u)=H(q,p)$ is independent of $u$, everything
664: reduces to what we said about unconstrained Hamiltonian mechanics.
665: Constrained Hamiltonian mechanics with unconstrained Hamiltonian 
666: $H_0(q,p)$ and $u$-independent holonomic constraints $C(q,p)=0$
667: are obtained by introducing a vector $u$ of Lagrange multipliers for 
668: the constraints and defining $H(q,p,u)=H_0(q,p) +C(q,p) \cdot u$.
669: Note that $\partial_u H(q,p,u) =  C(q,p)$ simply recovers the 
670: holonomic constraints. Thus, we see that the components of $u$ which 
671: occur only linearly in $H$ behave as multipliers of $u$-independent 
672: holonomic constraints.
673: \at{But what happened with the constraint 
674: equation? check this on examples! Treat the regular case and the
675: particular case of a regular quadratic term. Discuss the gauge case.}
676: 
677: \bigskip
678: Note that there is another class of models for conservative Hamiltonian 
679: dynamics, defined by so-called \bfi{nonholonomic constraints}. 
680: \at{refs} There 
681: the constrained dynamics is not given by \gzit{e.hamcu} but instead by
682: \[
683:   \dot q = \frac{dq}{dt} = \partial_p H(q,p),~~~ 
684:   \dot p = \frac{dp}{dt} = -\partial_q H(q,p)+A(q) u,~~~
685:   0 = \partial_p H(q,p)\cdot A(q),
686: \]
687: where $A(q)$ maps a multiplier vector $u\in\Uz$ to an element from 
688: $\Fz$, and the Hamiltonian $H$ again defines the energy.
689: The energy is conserved since $\frac{d}{dt} H(q,p) 
690: = \partial_q H(q,p)\cdot \dot q+\partial_p H(q,p)\cdot \dot p
691: =\partial_p H(q,p)\cdot A(q) u =0$. However, now the dynamics can 
692: usually no longer be written in terms of a variational principle. 
693: Only the special \bfi{integrable} case where $A(q) = \partial_q C(q)$ 
694: corresponds to holonomic constraints of the form $C(q)=0$ and a
695: modified Hamiltonian $\widetilde H(q,p,u):=H(q,p)-C(q) \cdot u$,
696: which agrees on the space of trajectories with $H$.
697: The most general conservative Hamiltonian system may have both 
698: holonomic and nonholonomic constraints; the reader may wish to write
699: down the defining equations and generalize the above discussion 
700: accordingly. 
701: 
702: 
703: 
704: 
705: 
706: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
707: \section{Lagrangian mechanics}\label{s.lagr}
708: 
709: Frequently, and especially in relativistic field theory, a classical
710: system is definied in terms of the Lagrangian approach to mechanics.
711: We consider here the \bfi{autonomous} case only, where the Lagrangian 
712: is time-independent.
713: 
714: The basic object is now a \bfi{Lagrangian} $L\in\C^\infty(T\Mz_c)$,
715: a function of ponits in the tangent space $T\Mz_c$ of a configuration
716: manifold $\Mz_c$. As in the Hamiltonian case, we restrict our attention
717: to the case where $\Mz_c$ is a nonempty, open subset of a convenient
718: vector space $\Fz$ over $\Rz$. Then the tangent space is 
719: $T\Mz_c=\Mz_c\times\Fz$, points in $T\Mz_c$ are pairs $(q,v)$
720: consisting of a configuration point $q\in\Mz_c$ and a tangent vector 
721: $v\in\Fz$ at $q$, referred to as \bfi{velocity}, and the Lagrangian is 
722: a function with function values $L(q,v)$.
723: 
724: The Lagrangian approach to mechanics can be represented
725: in the framework of constrained Hamiltonian dynamics by taking
726: $\Uz=\Fz$, and $u=v$. Then the choice 
727: \lbeq{e.Lham}
728: H(q,p,v):=up-L(q,v)
729: \eeq
730: for the Hamiltonian gives unconstrained Lagrangian mechanics.
731: Constrained Lagrangian mechanics with holonomic constraints $C(q,v)=0$ 
732: is similarly obtained by taking $\Uz=\Fz\times\Uz_0$ and $u=(v,u_0)$ 
733: and $H(q,p,v,u_0)=pv - L(q,v) +u_0^TC(q,v)$, where $u_0$ is a 
734: Lagrange multiplier. However, in the following, we only discuss the
735: unconstrained Lagrangian case.
736: 
737: Applying the general machinery of Section \ref{s.hamcon} to
738: \gzit{e.Lham}, we find as dynamical equations the 
739: \bfi{Euler-Lagrange equations}
740: \lbeq{po.2}
741:    \dot q = v,~~~\dot p = \partial_q L(q,v),~~~p=\partial_v L(q,v);
742: \eeq
743: and the action \gzit{e.action} reduces on the submanifold defined by 
744: $\dot q = v$ to 
745: \lbeq{e.Laction}
746: I(q) := \int_{\ul t}^{\ol t} dt~  L(q(t),\dot q(t)).
747: \eeq
748: The Hamiltonian \gzit{e.Lham} is time invariant since
749: \[
750:    (p\cdot\dot q)^\pdot = \dot p\cdot\dot q+p\cdot \ddot{q} = 
751: L_q\cdot \dot q+ L\dot q\cdot \ddot{q} = L(q,\dot q)^\pdot = \dot L.
752: \]
753: It is easily verified directly that the condition for $I(q)$ to be 
754: stationary at the path $q$ gives again the Euler-Lagrange equations 
755: \gzit{po.2}; this is usually taken as the starting point of the
756: Lagrangian approach.
757: 
758: \begin{expl}
759: The Lagrangian $L(q,v)=  \half m v^2 - \half k q^2$ defines the 
760: harmonic oscillator, as can be seen by writing down the
761: Euler-Lagrange equations. Note that the action need not be bounded 
762: below, as can be seen from the path $q(t)=s(1-t^2)$ in
763:  $[\ul t,\ol t]=[-1,1]$, where $I(q)=(4m-\frac{8}{15}k)s^2$ diverges 
764: to $-\infty$ when $k>7.5m$ and $s\to\infty$. Thus, it is inappropriate
765: to refer to the stationary action principle as \idx{principle of least 
766: action}, as often done for historical reasons.
767: \at{Check that at the harmonic solution we only have a saddle point!}
768: \end{expl}
769: 
770: If we change a Lagrangian $L(q,v)$ to
771: \[
772: \widetilde L(q,v):=L(q,v)+ v \dot \partial_q \phi(q)
773: \]
774: for some smooth function $\phi$, the action $I(q)$ remains
775: unchanged apart from a boundary term arising through integration by 
776: parts.  As a result, the new equations of motions and the old ones
777: are equivalent. On the other hand, the momentum changes from $p$
778: to $\widetilde p = p +  \partial_q \phi(q)$. This does not affect the
779: equation of motion in the form \gzit{e.heg} since the transformation
780: from $p$ to $\tilde p$ is a canonical transformation leaving the 
781: Lie product invariant. Indeed, it is not difficult to see that the
782: more general substitution of $p'=p+\chi(q)$ for $p$ preserves the
783: Lie product iff $\partial_q\chi(q)$ is symmetric. Necessity follows 
784: since for constants $a,b\in \Fz$,
785: \[
786: a \cdot p' \lp b \cdot p' = a\cdot \chi(q) b -b \cdot \chi(q) a
787: \]
788: must vanish, and sufficiency can be established by a more involved 
789: computation.
790: 
791: \at{reconsider the Hamiltonian example;
792: mention the field theory case, Lagrangian density, and covariant 
793: actions.}
794: 
795: 
796: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
797: \at{{\bf *********** Ignore the remainder of this section ***********}
798: It treats the constraint Lie product, and is 
799: not yet adapted to the uniform notation used throughout the book.}
800: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
801: 
802: We may work directly in the tangent manifold \at{define $\Ez$ etc.}
803: and define the linear form
804: \[
805: \theta_L = p = \partial_{\dot q} L, 
806: \]
807: \lbeq{e.66a}
808:  \theta_L(X):=Xp,
809: \eeq
810: and the canonical 2-form
811: \[
812: \omega_L = -d\theta_L.
813: \]
814: Then $d\omega_L = -ddp = 0$ implies that $\omega_L$ is closed, 
815: and hence Theorem \ref{t2.2} applies.
816: If $\omega_L$ is non-degenerate, we can solve for $\dot q$ in terms
817: of $p,q$ and get the Hamiltonian picture in the traditional way. The
818: Poisson algebra becomes the standard Poisson algebra on the cotangent
819: bundle. If $\omega_L$ is degenerate, we cannot solve for $\dot q$
820: and compatibility restricts the space $\Ez(\omega_L)$ of quantities.
821: \[
822:    p = L_{\dot q} \in \Ez \otimes \Cz\Fz 
823: \]
824: is the \bfi{canonical momentum}.
825: 
826: \at{check notation; how are the Lie products related?
827: Is $f\in\Ez(\omega_l)$ iff it is a function of $q$ and 
828: $\partial_v L(q,v)$? If so, just make a remark that not every $p$ 
829: has the right form in the singular case.}
830: On $\Ez=C^\infty(T  \Mz  )$, any Lagrangian $L=L(q,\dot q)$ defines
831: a Lie product on $\Ez(\omega_L)$
832: which induces the Euler-Lagrange dynamics defined by the action 
833: $I=\int dt L$. 
834: 
835: We rearrange the canonical 2-form as
836: \[
837:    \omega_L = dq \wedge dp = dq \wedge (p_q dq+ p_{\dot q} d \dot q).
838: \]
839: The condition for $f\in \Ez(\omega_L)$ to be compatible with $\omega$ 
840: requires the 
841: existence of a Hamiltonian vector field $X_f$ with 
842: \lbeq{e.sing}
843:    \frac{\partial f}{\partial \dot q} = G dq (X_f),~~~
844:    \frac{\partial f}{\partial q} = -G d \dot q (X_f),
845: \eeq
846: with the symmetric \bfi{Hessian matrix}
847: \lbeq{po.6a}
848: G:=\partial_q^2 L = 
849: \partial_q \partial_v L=\frac{\partial p}{\partial  \dot q}
850: \eeq
851: 
852: %%%%%%%%%%%%%
853: {\sc Case 1.} In the {\em regular} case, i.e., if the Hessian matrix
854: is invertible, we can solve the constraint equation $p=\partial_v L$ 
855: at least locally for $v$, getting an equation $\dot q = v(q,p)$.
856: In this case, we find from \gzit{e.sing} that
857: \lbeq{po.7}
858:    f \lp g = \frac{\partial f}{\partial q} \cdot G^{-1} 
859:    \frac{\partial g}{\partial \dot q}  - \frac{\partial g}
860:    {\partial q} \cdot G^{-1} \frac{\partial f}{\partial \dot q},
861: \eeq
862: where $f$, $g$ are functions of $q$ and $\dot q$. 
863: Note that
864: \[
865:    L_v(q,v(q,p))=p,
866: \]
867: and 
868: \[
869:    H(q,p) = pv(q,p)-L(q,v(q,p))
870: \]
871: has derivatives 
872: \[
873:    H_p=v(q,p)+p v_p(q,p) - L_v(q,v(q,p)) v_p (q,p) = v(q,p) = \dot q,
874: \]
875: \[
876:    H_q=pv_q (q,p) - L_q(q,v(q,p))-L_v(q,v(q,p)) v_q(q,p)
877:       = -L_q(q, \dot q)  = - \dot p,
878: \]
879: so that
880: \[
881:    \frac{d}{dt} f(q,p) = f_p \dot p + f_q \dot q = -f_p H_q+f_q H_p
882:    = H \lp f,
883: \]
884: with the canonical Lie product on phase space.
885: 
886: \at{is this just a repeat?}
887: Since $H = (p| \dot q)-L$, we have for solutions $q$
888: \[
889: \frac{\partial H}{\partial \dot q} 
890:   = \Big( \frac{\partial p}{\partial\dot q}\Big|\dot q\Big)+p-L_{\dot q}
891:   =\frac {\partial p}{\partial \dot q} \dot q,
892: \]
893: \[
894:   \bary{ll}
895:     \D\frac{\partial H}{\partial q} & = \D\Big( \frac{\partial p}
896:     {\partial q} \Big | \dot q \Big) - \frac{\partial L}{\partial q} = 
897:     \Big( \frac{\partial p}{\partial q}\Big| \dot q \Big) - \dot p\\
898:     & = \D\Big( \frac{\partial p}{\partial q} \Big| \dot q\Big) - 
899:     \Big( \Big( \frac{\partial p}{\partial q} \Big| \dot q\Big) +
900:     \Big( \frac{\partial p}{\partial \dot q}\Big|\ddot{q}\Big) \Big) =
901:    - \Big( \frac{\partial p}{\partial \dot q} \Big| \ddot{q} \Big).
902:   \eary
903: \]
904: Hence \at{why?}
905: \[
906:    X_H = \Big( \frac{d}{dt} q \Big| \partial_q \Big) +
907:    \Big( \frac{d}{dt} \dot q \Big| \partial_{\dot q} \Big)
908: \]
909: and 
910: \[
911:    H \lp g = dg (X_H) = \Big( \frac{\partial g}{\partial q} \Big| 
912:    \dot q \Big) + \Big( \frac{\partial g}{\partial \dot q} \Big|
913:    \ddot{q} \Big) = (g(q, \dot q))^\pdot = \dot g,
914: \]
915: so that $H$ generates the dynamics.
916: 
917: \at{but the dynamics is already assumed in the derivation!?}
918: 
919: 
920: 
921: %%%%%%%%%%%%%
922: {\sc Case 2.} 
923: In the {\em singular} case, i.e., when the Hessian matrix \gzit{po.6a} 
924: is not invertible,  condition \gzit{e.obs} is nontrivial, 
925: not all $f(q,\dot q)$ are compatible with $\omega$ and hence in the 
926: Poisson algebra. Then 
927: \gzit{po.7} only holds for the generalized inverse and \gzit{e.sing}
928: requires that the partial derivatives are in the range of $G$.
929: \at{These should be mappings 
930: of $T \Mz $ which leave both $H$ from \gzit{e.Lham} and $\theta_L$ 
931: from \gzit{e.66a}
932: invariant. These form a group. Typical transformations should have 
933: the form $\ol q = Q(q), \ol v = Q'(q)v+w(q)$.
934: Clarify! Only first class constraints 
935: have undetermined multipliers and hence an associated 
936: gauge freedom -- see 
937: [R. Jackiw,
938: (Constrained) Quantization Without Tears,
939: arXiv:hep-th/9306075}
940: The Poisson manifold (or orbifold?) 
941: is the set of orbits of the gauge group;
942:  cf. M/R p. 325. Restrict $\Ez$ accordingly, as in the symplectic case:]
943: 
944: The resulting Lie product (cf. \gzit{e.liep}) is 
945: \lbeq{po.lie}
946:    f \lp g = d g (X_f) 
947:    = \frac{\partial g}{\partial q} dq (X_f)
948:    + \frac{\partial g}{\partial \dot q} d \dot q (X_f).
949: \eeq
950: 
951: Note that the standard treatment in terms of symplectic 
952: manifolds requires regularity. In the singular case, complicated 
953: additional assumptions and arguments are needed to bring theories
954: with gauge symmetries (which are always singular) into the 
955: framework of symplectic geometry. 
956: 
957: 
958: