[MAP] Use of x,y,t as Hamiltonian coordinates.
Kirk T McDonald
kirkmcd at Princeton.EDU
Fri Mar 11 19:29:17 EST 2011
Folks,
This email greatly clarifies things for me, thanks to help from Alex Dragt.
One aspect of this clarification is that we may be making a formal error in the way we calculate (longitudinal) emittance when fields are present.
--Kirk
This email takes up the related theme that we prefer to use (x,y,t), rather than (x,y,z) as the q coordinates, with z rather than t as the independent variable in the Hamiltonian.
Thanks to Alex for pointing me to sec.1.6 of his impressive e-book!
http://www.physics.umd.edu/dsat/
What I get from this is that the canonical momentum associated with coordinate t is
p_t = – H, where H = the Hamiltonian.
For a charged particle in an external electromagnetic field, the Hamiltonian H includes both the scalar potential V and the vector potential A.
[Dragt gives the Hamiltonian using coordinates (x,y,z) in his eq. (1.5.130) on .pdf page 142 = text page 88, and the Hamiltonian using coordinates (x,y,t) in his eq. (1.6.16) on .pdf page 153 = text page 99.]
The Hamiltonian is NOT the mechanical energy when nonzero electromagnetic fields are present.
So, when calculating “emittances” in these coordinates, we should NOT be using the energy E, but the Hamiltonian H as the “momentum” associated with coordinate t.
This is contrary to what I have been told we are doing in, say, the programs ICOOL/ECALC, although perhaps I have a misunderstanding about this.
I have the impression that we are using E rather than H, but that we do use canonical momenta in the x and y coordinates when calculating emittances.
This might be the source of some numerical troubles that may exist in recent calculations.
I presume that in the Hamiltonian formalism with coordinates (x,y,t) and canonical momenta
(p_mech,x + q A_x / c,p_mech,y + q A_y /c,-H)
that the density in this space space of a set of noninteracting particles is invariant with respect to the independent variable z, according to Liouville’s theorem.
Next, I review Swann’s argument to see what happens if we simply ignore the scalar and vector potential. In this case we would associate the mechanical energy E as the (noncanonical) “momentum” of coordinate t. So, the goal is to keep using coordinates (x,y,t) but to switch to “momenta” (p_mech,x,p_mech,y,E).
The off-diagonal elements in the Jacobian (Swann, p. 226) again don’t matter.
However, we have a problem in that the (E,H) diagonal matrix element is
partial H / partial E, which is not equal to 1 when fields are present.
This means that we cannot simply ignore the vector (and scalar) potentials when calculating phase volumes (and presumably also emittance) when using coordinates (x,y,t) rather than (x,y,z).
I believe that a gauge transformation remains a canonical transformation when using the coordinates (x,y,t), so phase volume (and presumably also emittance) is gauge invariant.
Since we are strongly committed to the use of coordinates (x,y,t), it seems that we will have to continue including the vector (and scalar!) potentials in our emittance calculations.
We may presently be making an error in using E_mech, rather than H as the canonical momentum associated with coordinate t.
From: alex dragt
Sent: Friday, March 11, 2011 4:24 PM
To: Sergei Nagaitsev ; Don Summers ; Yuri Alexahin
Cc: MAP List ; alex dragt ; Kirk T McDonald ; srmane001 at gmail.com
Subject: [MAP] Liouville's theorem, kinematic invariants, and dynamic invariants
Dear All,
Confusion reigns: As described in the papers
A. Dragt, F. Neri, et al., "LIE ALGEBRAIC TREATMENT OF LINEAR AND NONLINEAR BEAM DYNAMICS",
Annual Review of Nuclear and Particle Science 38, p. 455 (1988).
A. Dragt, R. Gluckstern, et al., "THEORY OF EMITTANCE INVARIANTS", Lecture Notes in
Physics 343: Proceedings of the Joint US-CERN Capri School on Accelerator Physics,
Springer Verlag (1989).
A. Dragt, F. Neri, et al., "GENERAL MOMENT INVARIANTS FOR LINEAR HAMILTONIAN SYSTEMS",
Physical Review A, 45, p. 2572 (1992).
there are two kinds of invariants, which I call "kinematic" and "dynamic". It is important not to confound them.
The Courant-Snyder invariants, generally treated in the linear approximation as described by Sergei below, are examples of dynamic invariants. Their use requires a knowledge of the equations of motion or, equivalently, a one-period or one-turn map. (They generally apply to periodic systems for which the linear part of the map has all its eigenvalues on the unit circle and distinct, but in principle could also be constructed when the eigenvalues are off the unit circle.) Contrary to the reservations that appear to be expressed by Sergei, they are extendable to the nonlinear case using normal form methods. See, for example, the first paper above and Sections 8.10 and 8.11 of the MaryLie manual available at the Web site
http://www.physics.umd.edu/dsat/
There are also papers in the literature that extend them, at least in the linear case, to non-periodic systems.
Examples of kinematic invariants include what I call "moment invariants" and "eigen emitances". They involve moments of particle distributions and, like Liouville's theorem, make no specific use of the equations of motion save for the symplectic condition. It is these invariants that are the current focus of interest with regard to questions of "emittance partitioning". See the draft paper
arXiv:1010.1558v2 [physics.acc-ph]
Their computation and use are described in the papers above and Section 8.37 of the MaryLie manual.
Moment invariants are currently known only for the case of linear maps, but there is some evidence that they should also exist in the nonlinear case. See Chapter 26 (only partially complete) and Section 6.8.2 of "Lie Methods ...", also available at the Web site above. In the past I have tried to construct moment invariants for nonlinear maps by finding Casimir operators for the full Lie algebra of all symplectic maps. The problem is difficult because the usual method for constructing Casimir operators requires that the Killing metric be invertible. See Section 21.11 of "Lie Methods ...". This metric is not invertible in the nonlinear case, and therefore the standard machinery for constructing Casimirs fails. With considerable effort I found an alternate approach with some promising results, but was eventually overwhelmed by algebraic complexity. Moment invariants for the nonlinear case, if they exist at all, will require many pages even to write them down.
I hope these remarks provide some clarification.
Best,
Alex
On Mar 10, 2011, at 7:32 PM, Sergei Nagaitsev wrote:
Dear Kirk,
let me add my two cents to this discussion:
1. For the beam emittance to be a useful quantity, it needs to be conserved as the beam propagates along the beamline (consider non-interacting particles) through various external electromagnetic fields. To this end, the definition of emittance that reflects such a property is "an ensemble-averaged action". The particle motion is assumed to be integrable, i.e. there exists 3 functionally-independent constants of the motion in involution with the Hamiltonian. Particle actions, expressed through such constants of motion, are also constants of motion. Thus, the average actions (emittances) are conserved. In a linear-focusing transport channel with linear rf focusing, such constants of motion exist and are called Courant-Snyder invariants (2 transverse and 1 longitudinal). In a nonlinear (and generally time-dependent) focusing channel such constants of motion might not exist, therefore, the emittance (as a conserved quantity) is not defined. We may use an approximate expression for the emittance by using linear-only focusing to define actions and then by treating non-linearities as a perturbation. This leads to an effective emittance growth if a beamline has nonlinear elements even though the Liouville's theorem states that the phase-space density is conserved. In some cases this emittance growth is not "real" (irreversible) but just a reflection that we are using an incorrect action definition. Finally, my definition of the emittance (as the average action) is identical to the definition through eigen-values of a sigma matrix only in a case of a linear focusing channel. However, where possible (like in case of a bunch occupying a large portion of an rf bucket) we should use exact actions, not approximate.
2. When averaging particle actions over the distribution function at a given time t it is useful to remember that a time slice t=const in one frame is not t'=const in another frame because of Lorentz transformations.
Sergei
----- Original Message -----
From: Kirk T McDonald <kirkmcd at Princeton.EDU>
Date: Thursday, March 10, 2011 8:49 pm
Subject: Re: [MAP] Liouville's theorem and electromagnetic fields
To: alex dragt <dragtnb at comcast.net>
Cc: MAP List <map-l at lists.bnl.gov>
Folks,
I have added Alex' paper to DocDB 560. See Appendix A.
It is gratifying to see that the fact that Liouville's theorem holds
for
both mechanical and canonical phase space is "well known to those who
know".
The challenge now is to learn how best to use the "freedom" offered to
us by
this apparently nonintuitive result.
--Kirk
--------------------------------------------------
--------------------------------------------------------------------------------
_______________________________________________
MAP-l mailing list
MAP-l at lists.bnl.gov
https://lists.bnl.gov/mailman/listinfo/map-l
-------------- next part --------------
An HTML attachment was scrubbed...
URL: https://lists.bnl.gov/mailman/private/map-l/attachments/20110311/207bda08/attachment-0001.html
More information about the MAP-l
mailing list