PDA

View Full Version : Covariant phase space


Thomas Larsson
Oct20-04, 03:18 AM
<jabberwocky><div class="vbmenu_control"><a href="jabberwocky:;" onClick="newWindow=window.open('','usenetCode','toolbar=no, location=no,scrollbars=yes,resizable=yes,status=no ,width=650,height=400'); newWindow.document.write('<HTML><HEAD><TITLE>Usenet ASCII</TITLE></HEAD><BODY topmargin=0 leftmargin=0 BGCOLOR=#F1F1F1><table border=0 width=625><td bgcolor=midnightblue><font color=#F1F1F1>This Usenet message\'s original ASCII form: </font></td></tr><tr><td width=449><br><br><font face=courier><UL><PRE>\n\nAs a response to Urs in another forum, let me describe how phase space\ncan be described in a covariant way.\n\nA phase space has local coordinates (q, p). Time evolution is given\nby Hamilton\'s equations, dq/dt = {q, H}, dp/dt = {p, H}, where H is\nthe Hamiltonian. The solution to these equations is some curve\n(q(t), p(t)). Since the curve is completely specified by the pair\n(q, p) = (q(0), p(0)), there is a 1-1 correspondence between solutions\nto the equations of motion and points in phase space.\n\nWe can therefore identify phase space with the space of curves solving\nHamilton\'s equation. This is the covariant phase space. The Hamiltonian\ntakes a very simple form in this formalism; it is simply the generator of\nt-translations. However, we have of course not gained anything from a\npractical point of view, because describing these curves is equivalent to\nsolving Hamilton\'s equations.\n\nHowever, we are not mainly interested in phase space itself, but rather in\nthe space of functions over it, C(q,p). It is this function space which\ncorresponds to the Hilbert space after quantization. C(q,p) can be easily\ndescribed as the space of arbitrary functionals C[q(t),p(t)], modulo the\nideal generated by the equations of motion.\n\nActually, people usually ignore p(t) because it is fully specified by the\nEuler-Lagrange equations, symbolically written as E(t) = 0. These equation\ngenerate an ideal I in the space of functionals C[q(t)]. The ideal I\nconsistent of functionals of the form E(t) f[q(t)], f arbitrary, and we want\nto construct the space C[q(t)]/I. This space can be obtained as the zeroth\ncohomology group of a certain complex, called the Koszul-Tate complex in\nmath. Cohomology and complexes are scary, I know, but people who know about\nthe BRST complex will hopefully still follow.\n\nSo we introduce a fermionic antifield q*(t) and a KT differential d defined\nby\n\nd q(t) = 0, d q*(t) = E(t).\n\nThe enlarged function space C[q(t),q*(t)] can be decomposed into\nsubspace V_i of fixed antifield number. Since the KT differential obviously\nis nilpotent (E(t) depends on q(t) but not on q*(t)), it defines a complex\n\n0 &lt;-- V_0 &lt;-- V_1 &lt;-- V_2 &lt;-- ... .\n\nThe cohomology space are defined as usual as H_i = ker d/im d. It is a\nstandard proof that this complex yields a resolution of C[q(t)]/I, i.e.\nthat H_0 = C[q(t)]/I and H_i = 0 for all i &gt; 0.\n\nLet us now return to the full phase space by adding the momenta p(t) and\np*(t), satisfying\n\n[p(t), q(t\')] = delta(t-t\'), [p*(t), q*(t\')]_+ = delta(t-t\').\n\nThe KT differential acts on arbitary functional F(q,q*,p,p*) as dF = [Q, F],\nwhere\n\nQ = \\int dt E(t) p*(t).\n\nWe can again introduce an enlarged function space C[q(t),q*(t),p(t),p*(t)]\nwhich decomposes into subspaces of fixed antifield number. The complex now\nextends to infinity in both directions,\n\n... V_-2 &lt;-- V_-1 &lt;-- V_0 &lt;-- V_1 &lt;-- V_2 &lt;-- ... .\n\nAgain we identify the zeroth cohomology group H^0(Q) as the space of\nfunctions over the physical phase space.\n\nNow let\'s quantize! The idea is that if we quantize each space V_i, then the\ncohomology group H^0(Q) is the Hilbert space (or pre-Hilbert spaces, since\nwe don\'t talk about an inner product) of the quantum theory with Euler-\nLagrange equation E(t) = 0.\n\nSo, what about diffeomorphisms? Assume that the dynamics is diff invariant,\nas in GR. Then the unrestricted spaces V_i are spaces of tensor fields,\nand we know how to quantize them, making all diffeomorphisms into\nwell-defined operators. This is the stuff about passing to the Taylor data\nand normal ordering. Moreover, the KT operator Q is diff invariant, even\nafter quantization. It is important to notice here that Q is already\nnormal ordered, because E(t) and p*(t) commute. Therefore, it always\nremains nilpotent after quantization, and the cohomology groups will\ncarry representations of the anomalous diffeomorphism algebra.\n\nWe now see way the physical 4-diff symmetry becomes the Dirac algebra in\nconventional canonical quantization. We label phase space by points (q,p) =\n(q(0),p(0)). The result of some 4-diffs, e.g. time translations, lies\noutside points of this form. We must therefore add a compensating\ntransformation to get back to a point of the form (q\'(0),p\'(0)). But this\nis only an artefact of our choice of coordinatization and nothing physical.\nPhysically, the Dirac algebra and the 4-diff algebra are equivalent.\n</UL></PRE></font></td></tr></table></BODY><HTML>');"> <IMG SRC=/images/buttons/ip.gif BORDER=0 ALIGN=CENTER ALT="View this Usenet post in original ASCII form">&nbsp;&nbsp;View this Usenet post in original ASCII form </a></div><P></jabberwocky>As a response to Urs in another forum, let me describe how phase space
can be described in a covariant way.

A phase space has local coordinates (q, p). Time evolution is given
by Hamilton's equations, dq/dt = {q, H}, dp/dt = {p, H}, where H is
the Hamiltonian. The solution to these equations is some curve
(q(t), p(t)). Since the curve is completely specified by the pair
(q, p) = (q(0), p(0)), there is a 1-1 correspondence between solutions
to the equations of motion and points in phase space.

We can therefore identify phase space with the space of curves solving
Hamilton's equation. This is the covariant phase space. The Hamiltonian
takes a very simple form in this formalism; it is simply the generator of
t-translations. However, we have of course not gained anything from a
practical point of view, because describing these curves is equivalent to
solving Hamilton's equations.

However, we are not mainly interested in phase space itself, but rather in
the space of functions over it, C(q,p). It is this function space which
corresponds to the Hilbert space after quantization. C(q,p) can be easily
described as the space of arbitrary functionals C[q(t),p(t)], modulo the
ideal generated by the equations of motion.

Actually, people usually ignore p(t) because it is fully specified by the
Euler-Lagrange equations, symbolically written as E(t) = . These equation
generate an ideal I in the space of functionals C[q(t)]. The ideal I
consistent of functionals of the form E(t) f[q(t)], f arbitrary, and we want
to construct the space C[q(t)]/I. This space can be obtained as the zeroth
cohomology group of a certain complex, called the Koszul-Tate complex in
math. Cohomology and complexes are scary, I know, but people who know about
the BRST complex will hopefully still follow.

So we introduce a fermionic antifield q*(t) and a KT differential d defined
by

d q(t) = 0, d q*(t) = E(t).

The enlarged function space C[q(t),q*(t)] can be decomposed into
subspace V_i of fixed antifield number. Since the KT differential obviously
is nilpotent (E(t) depends on q(t) but not on q*(t)), it defines a complex

<-- V_0 <-- V_1 <-- V_2 <-- ... .

The cohomology space are defined as usual as H_i = ker d/im d. It is a
standard proof that this complex yields a resolution of C[q(t)]/I, i.e.
that H_0 = C[q(t)]/I and H_i = for all i > .

Let us now return to the full phase space by adding the momenta p(t) and
p*(t), satisfying

[p(t), q(t')] = \delta(t-t'), [p*(t), q*(t')]_+ = \delta(t-t').

The KT differential acts on arbitary functional F(q,q*,p,p*) as dF = [Q, F],
where

Q = \int dt E(t) p*(t).

We can again introduce an enlarged function space C[q(t),q*(t),p(t),p*(t)]
which decomposes into subspaces of fixed antifield number. The complex now
extends to infinity in both directions,

... V_-2 <-- V_-1 <-- V_0 <-- V_1 <-- V_2 <-- ... .

Again we identify the zeroth cohomology group H^0(Q) as the space of
functions over the physical phase space.

Now let's quantize! The idea is that if we quantize each space V_i, then the
cohomology group H^0(Q) is the Hilbert space (or pre-Hilbert spaces, since
we don't talk about an inner product) of the quantum theory with Euler-
Lagrange equation E(t) = .

So, what about diffeomorphisms? Assume that the dynamics is diff invariant,
as in GR. Then the unrestricted spaces V_i are spaces of tensor fields,
and we know how to quantize them, making all diffeomorphisms into
well-defined operators. This is the stuff about passing to the Taylor data
and normal ordering. Moreover, the KT operator Q is diff invariant, even
after quantization. It is important to notice here that Q is already
normal ordered, because E(t) and p*(t) commute. Therefore, it always
remains nilpotent after quantization, and the cohomology groups will
carry representations of the anomalous diffeomorphism algebra.

We now see way the physical 4-diff symmetry becomes the Dirac algebra in
conventional canonical quantization. We label phase space by points (q,p) =
(q(0),p(0)). The result of some 4-diffs, e.g. time translations, lies
outside points of this form. We must therefore add a compensating
transformation to get back to a point of the form (q'(0),p'(0)). But this
is only an artefact of our choice of coordinatization and nothing physical.
Physically, the Dirac algebra and the 4-diff algebra are equivalent.

Urs Schreiber
Oct20-04, 10:48 AM
<jabberwocky><div class="vbmenu_control"><a href="jabberwocky:;" onClick="newWindow=window.open('','usenetCode','toolbar=no, location=no,scrollbars=yes,resizable=yes,status=no ,width=650,height=400'); newWindow.document.write('<HTML><HEAD><TITLE>Usenet ASCII</TITLE></HEAD><BODY topmargin=0 leftmargin=0 BGCOLOR=#F1F1F1><table border=0 width=625><td bgcolor=midnightblue><font color=#F1F1F1>This Usenet message\'s original ASCII form: </font></td></tr><tr><td width=449><br><br><font face=courier><UL><PRE>"Thomas Larsson" &lt;thomas_larsson_01@hotmail.com&gt; schrieb im Newsbeitrag\nnews:24a23f36.0410192049.4fa0703e@pos ting.google.com...\n\n&gt; The KT differential acts on arbitary functional F(q,q*,p,p*) as dF = [Q,\n&gt; F],\n&gt; where\n&gt;\n&gt; Q = \\int dt E(t) p*(t).\n\n&gt; So, what about diffeomorphisms? Assume that the dynamics is diff\n&gt; invariant,\n&gt; as in GR. Then the unrestricted spaces V_i are spaces of tensor fields,\n\n\nFor GR the above discussion must be generalized to include more than one\nconstraint.\n\n\n&gt; and we know how to quantize them, making all diffeomorphisms into\n&gt; well-defined operators. This is the stuff about passing to the Taylor data\n&gt; and normal ordering. Moreover, the KT operator Q is diff invariant, even\n&gt; after quantization. It is important to notice here that Q is already\n&gt; normal ordered, because E(t) and p*(t) commute.\n\n\nThe ordering inside of E is irrelevant only as long as E is the only\nconstraint there is. As soon as there are more constraints ordering plays a\nrole. In particular, as soon as you use the ADM constraints of GR with\nordinary weakly continuous reps of p and q no ordering can be found which\nmakes the commutators of the constraints well defined.\n\n\n&gt; Therefore, it always remains nilpotent after quantization,\n\n\nMaybe I am not getting your point, but this is not true already for 2d\ngravity, as you know.\n\n</UL></PRE></font></td></tr></table></BODY><HTML>');"> <IMG SRC=/images/buttons/ip.gif BORDER=0 ALIGN=CENTER ALT="View this Usenet post in original ASCII form">&nbsp;&nbsp;View this Usenet post in original ASCII form </a></div><P></jabberwocky>"Thomas Larsson" <thomas_larsson_01@hotmail.com> schrieb im Newsbeitrag
news:24a23f36.0410192049.4fa0703e@posting.google.c om...

> The KT differential acts on arbitary functional F(q,q*,p,p*) as dF = [Q,
> F],
> where
>
> Q = \int dt E(t) p*(t).

> So, what about diffeomorphisms? Assume that the dynamics is diff
> invariant,
> as in GR. Then the unrestricted spaces V_i are spaces of tensor fields,


For GR the above discussion must be generalized to include more than one
constraint.


> and we know how to quantize them, making all diffeomorphisms into
> well-defined operators. This is the stuff about passing to the Taylor data
> and normal ordering. Moreover, the KT operator Q is diff invariant, even
> after quantization. It is important to notice here that Q is already
> normal ordered, because E(t) and p*(t) commute.


The ordering inside of E is irrelevant only as long as E is the only
constraint there is. As soon as there are more constraints ordering plays a
role. In particular, as soon as you use the ADM constraints of GR with
ordinary weakly continuous reps of p and q no ordering can be found which
makes the commutators of the constraints well defined.


> Therefore, it always remains nilpotent after quantization,


Maybe I am not getting your point, but this is not true already for 2d
gravity, as you know.

Thomas Larsson
Oct21-04, 02:47 AM
<jabberwocky><div class="vbmenu_control"><a href="jabberwocky:;" onClick="newWindow=window.open('','usenetCode','toolbar=no, location=no,scrollbars=yes,resizable=yes,status=no ,width=650,height=400'); newWindow.document.write('<HTML><HEAD><TITLE>Usenet ASCII</TITLE></HEAD><BODY topmargin=0 leftmargin=0 BGCOLOR=#F1F1F1><table border=0 width=625><td bgcolor=midnightblue><font color=#F1F1F1>This Usenet message\'s original ASCII form: </font></td></tr><tr><td width=449><br><br><font face=courier><UL><PRE>\n\n"Urs Schreiber" &lt;Urs.Schreiber@uni-essen.de&gt; wrote in message news:&lt;cl61c7\\$kit\\$1@lfa222122.richmond.edu&gt;...\ n\n&gt; For GR the above discussion must be generalized to include more than one\n&gt; constraint.\n\nE is not a constraint, it is the dynamics, i.e. the Euler-Lagrange\nequation.\n\nBut I over-simplified a bit; in the presence of gauge symmetries, you\nmust add a second-order antifield for the Noether identities.\nNevertheless, the Koszul-Tate differential always consists of pieces\nthat commute. If you want to mod out constraints, you must also add a\nBRST piece. That does not consist of commuting pieces, and cannot be\nwell-defined, AFAIU. But we don\'t want to mod out infinite spacetime\nsymmetries anyway, since that would clash with locality.\n\nWhat I describe is very similar to Batalin-Vilkovisky approach, see\ne.g. chapter 17 in\n\nHenneaux, M. and Teitelboim, C.:\nQuantization of gauge systems\nPrinceton Univ. Press (1992)\n\nOr look at Jim Stasheff\'s secret paper,\nhttp://www.arxiv.org/abs/hep-th/9712157.\nThat\'s where I learned about the BV cohomology.\nOr search the arxiv for interesting papers. Good keywords could be\n"anti-fields", "BV" or "Henneaux".\n\nThere are two differences, though.\n\n1. The BV formalism is aimed at a path-integral integral formalism.\nSince the entire dynamics is encoded in q(t), they skip p(t) and call\nthe space of q-solutions "the covariant phase space". This is really\na misnomer, because an honest phase space needs not only dynamics but\nalso a proper Poisson bracket. Since I do Hamiltonian quantization,\nI need the full space of (q,p)-solutions.\n\n2. The full BV-BRST differential has both a KT part, which depends on\nboth antifields and implements the dynamics and the constraints, and a\nlongitudinal part, which depends on ghosts and mod out gauge\nsymmetries. In the presence of anomalies, the longitudinal part must be\nomitted. But this is good because the KT part is already normal ordered,\nso we don\'t have to deal with the nasty BRST part.\n\n&gt;\n&gt; The ordering inside of E is irrelevant only as long as E is the only\n&gt; constraint there is.\n\nNo, no, no! The Euler-Lagrange operator E depends only on the fields and\nnot on their momenta. I don\'t identify velocities and momenta. In BRST,\nthe idea is to quantize first and impose constraints later. In BV, KT and\nBRST fit together into the BV differential, so dynamics and constraints\nare on the same footing. So the idea is to quantize first and impose\ndynamics later.\n\nThis illustrates a very useful idea in mathematics. If you have some\ncomplicated space such as functions over the covariant phase space, replace\nit by a sequence of simpler spaces connected by a nilpotent operator.\n\n&gt; As soon as there are more constraints ordering plays a\n&gt; role. In particular, as soon as you use the ADM constraints of GR with\n&gt; ordinary weakly continuous reps of p and q no ordering can be found which\n&gt; makes the commutators of the constraints well defined.\n&gt;\n\nIn a conventional Hamiltonian theory such as ADM, the phase space\ncoordinates are (q(0), p(0)). A 3-diff preserves the timeslice, so\nit acts like\n\nsymmetry\n(q(0), p(0)) ----------&gt; (q\'(0), p\'(0))\n\nAll other 4-diffs move us out of the chosen timeslice, so we must\ncompensate for that, symbolically\n\nsymmetry compensation\n(q(0), p(0)) ----------&gt; (q\'(t), p\'(t)) ------------&gt; (q\'(0), p\'(0))\n\nThe combination of 4-diffs+compensations generate the Dirac algebra.\nHowever, if we canonically quantize gravity in the covariant phase\nspace, we don\'t need to compensate, so the symmetry is the untainted\n4-diff algebra. This is still Hamiltonian quantization, but not ADM.\n\nYou keep repeating that no normal ordering can be found. If you work\nwith the fields themselves, that is true; I was fully aware of that in\nearly 1988. To avoid that, we must do a Taylor expansion, truncate at\nsome finite order p, and express everything in terms of Taylor data.\nBecause we have truncated without sacrificing general covariance, we\ncan now construct a well-defined, normal-ordered action of diffeomorphisms.\n\nThe simplest example. The diff algebra is generated by vector fields\nX = X^u(x) d/dx^u. Our canonical variables are two operator functions\nq^u(t) and p_u(t), satisfying\n\n[p_v(t), q^u(t\')] = \\delta^u_v \\delta(t - t\').\n\nThe diffeomorphism generators are now\n\nL_X = \\int dt : X^u(q(t)) p_u(t) :\n\nwhere normal ordering moves negative frequency modes to the right.\nIt is easy to verify that these operators satisfy the Virasoro algebra\nin 4D and that they are completely well-defined. A similar strategy\nworks if you add Taylor coefficients up to order p, and their canonical\nmomenta.\n\nSo your claim that no normal ordering can be found is only true if you\nstick to the fields themselves. By starting from the regularized\nfields, i.e. the truncated Taylor series, a well-defined normal\nordering can be found. This leaves us with the problem of ultimately\nremoving the regulator, i.e. to pass to infinite Taylor series (which I\nidentify with the fields). With natural dynamics, this condition\nselects four spacetime dimensions, which I think is rather remarkable.\n\nI hope you understand my point now.\n</UL></PRE></font></td></tr></table></BODY><HTML>');"> <IMG SRC=/images/buttons/ip.gif BORDER=0 ALIGN=CENTER ALT="View this Usenet post in original ASCII form">&nbsp;&nbsp;View this Usenet post in original ASCII form </a></div><P></jabberwocky>"Urs Schreiber" <Urs.Schreiber@uni-essen.de> wrote in message news:<cl61c7$kit$1@lfa222122.richmond.edu>...

> For GR the above discussion must be generalized to include more than one
> constraint.

E is not a constraint, it is the dynamics, i.e. the Euler-Lagrange
equation.

But I over-simplified a bit; in the presence of gauge symmetries, you
must add a second-order antifield for the Noether identities.
Nevertheless, the Koszul-Tate differential always consists of pieces
that commute. If you want to mod out constraints, you must also add a
BRST piece. That does not consist of commuting pieces, and cannot be
well-defined, AFAIU. But we don't want to mod out infinite spacetime
symmetries anyway, since that would clash with locality.

What I describe is very similar to Batalin-Vilkovisky approach, see
e.g. chapter 17 in

Henneaux, M. and Teitelboim, C.:
Quantization of gauge systems
Princeton Univ. Press (1992)

Or look at Jim Stasheff's secret paper,
http://www.arxiv.org/abs/http://www.arxiv.org/abs/hep-th/9712157.
That's where I learned about the BV cohomology.
Or search the arxiv for interesting papers. Good keywords could be
"anti-fields", "BV" or "Henneaux".

There are two differences, though.

1. The BV formalism is aimed at a path-integral integral formalism.
Since the entire dynamics is encoded in q(t), they skip p(t) and call
the space of q-solutions "the covariant phase space". This is really
a misnomer, because an honest phase space needs not only dynamics but
also a proper Poisson bracket. Since I do Hamiltonian quantization,
I need the full space of (q,p)-solutions.

2. The full BV-BRST differential has both a KT part, which depends on
both antifields and implements the dynamics and the constraints, and a
longitudinal part, which depends on ghosts and mod out gauge
symmetries. In the presence of anomalies, the longitudinal part must be
omitted. But this is good because the KT part is already normal ordered,
so we don't have to deal with the nasty BRST part.

>
> The ordering inside of E is irrelevant only as long as E is the only
> constraint there is.

No, no, no! The Euler-Lagrange operator E depends only on the fields and
not on their momenta. I don't identify velocities and momenta. In BRST,
the idea is to quantize first and impose constraints later. In BV, KT and
BRST fit together into the BV differential, so dynamics and constraints
are on the same footing. So the idea is to quantize first and impose
dynamics later.

This illustrates a very useful idea in mathematics. If you have some
complicated space such as functions over the covariant phase space, replace
it by a sequence of simpler spaces connected by a nilpotent operator.

> As soon as there are more constraints ordering plays a
> role. In particular, as soon as you use the ADM constraints of GR with
> ordinary weakly continuous reps of p and q no ordering can be found which
> makes the commutators of the constraints well defined.
>

In a conventional Hamiltonian theory such as ADM, the phase space
coordinates are (q(0), p(0)). A 3-diff preserves the timeslice, so
it acts like

symmetry
(q(0), p(0)) ----------> (q'(0), p'(0))

All other 4-diffs move us out of the chosen timeslice, so we must
compensate for that, symbolically

symmetry compensation
(q(0), p(0)) ----------> (q'(t), p'(t)) ------------> (q'(0), p'(0))

The combination of 4-diffs+compensations generate the Dirac algebra.
However, if we canonically quantize gravity in the covariant phase
space, we don't need to compensate, so the symmetry is the untainted
4-diff algebra. This is still Hamiltonian quantization, but not ADM.

You keep repeating that no normal ordering can be found. If you work
with the fields themselves, that is true; I was fully aware of that in
early 1988. To avoid that, we must do a Taylor expansion, truncate at
some finite order p, and express everything in terms of Taylor data.
Because we have truncated without sacrificing general covariance, we
can now construct a well-defined, normal-ordered action of diffeomorphisms.

The simplest example. The diff algebra is generated by vector fields
X = X^u(x) d/dx^u. Our canonical variables are two operator functions
q^u(t) and p_u(t), satisfying

[p_v(t), q^u(t')] = \delta^u_v \delta(t - t')[/itex].

The diffeomorphism generators are now

[itex]L_X = \int dt : X^u(q(t)) p_u(t) :

where normal ordering moves negative frequency modes to the right.
It is easy to verify that these operators satisfy the Virasoro algebra
in 4D and that they are completely well-defined. A similar strategy
works if you add Taylor coefficients up to order p, and their canonical
momenta.

So your claim that no normal ordering can be found is only true if you
stick to the fields themselves. By starting from the regularized
fields, i.e. the truncated Taylor series, a well-defined normal
ordering can be found. This leaves us with the problem of ultimately
removing the regulator, i.e. to pass to infinite Taylor series (which I
identify with the fields). With natural dynamics, this condition
selects four spacetime dimensions, which I think is rather remarkable.

I hope you understand my point now.