D. L. Bruhwiler, 1 G. I. Bell, 1 A. Sobol, 1 P. Messmer, 1 J. Qiang, 2 R. Ryne, 2 W. Mori, 3 A....
-
Upload
pearl-cross -
Category
Documents
-
view
219 -
download
0
Transcript of D. L. Bruhwiler, 1 G. I. Bell, 1 A. Sobol, 1 P. Messmer, 1 J. Qiang, 2 R. Ryne, 2 W. Mori, 3 A....
D. L. Bruhwiler,1 G. I. Bell,1 A. Sobol,1
P. Messmer,1 J. Qiang,2 R. Ryne,2
W. Mori,3 A. Fedotov,4 I. Ben-Zvi,4
V. Litvinenko,4 S. Derbenev,5
R. Li,5 Y. Zhang,5 L. Merminga5
COMPASS – Community Petascale Project for Accelerator Science and
Simulation
Parallel Electron CoolingSimulations for SciDAC
1. Tech-X Corporation2. Lawrence Berkeley National Lab3. Univ. of California at Los Angeles4. Brookhaven National Lab5. Thomas Jefferson National Lab
COMPASS Kick-Off Meeting p. 2SciDAC – September 17, 2007 – Fermilab
Physics Motivation for Cooling Ion Beams
• A future polarized e-/ion collider is required to better probe the hadronic structure of matter– goal of the international nuclear physics community– ion luminosity of order 1033-1035 cm-2 s-1 is required
• orders of magnitude larger than present state-of-the-art– DOE Office of Nuclear Physics has two long-term concepts
• eRHIC (add e- linac+ring to the RHIC complex)• ELIC (add e- ring & ion linac+ring to Jefferson Lab)
• Higher luminosity requires cooling of ion beams– “electron cooling” is included in all present designs
• RHIC II luminosity upgrade requires e- cooling– near-term priority of the Office of Nuclear Physics– stochastic cooling is being implemented and may be good enough
• recent success of M. Blaskiewicz, M. Brennan & many collaborators• won’t work at higher intensities of eRHIC or ELIC
• Low-energy operation of RHIC is planned in 2010– recent idea to explore new physics– adequate statistics may require cooling for high luminosity
• Completely new idea: “coherent electron cooling”– combination of e- cooling and stochastic cooling– if it works, could provide 10x increase of friction or more
COMPASS Kick-Off Meeting p. 3SciDAC – September 17, 2007 – Fermilab
Courtesy of A. Fedotov et al., COOL ’07 Presentation (Sep. 10, 2007)
COMPASS Kick-Off Meeting p. 4SciDAC – September 17, 2007 – Fermilab
Electron Cooling for RHIC II
COMPASS Kick-Off Meeting p. 5SciDAC – September 17, 2007 – Fermilab
tp
Image courtesy of V. Litvinenko, BNL
Merminga, EIC2007, Apr 6-7 2007
ELIC LayoutELIC Layout
30-225 GeV protons15-100 GeV/n ions
3-9 GeV electrons3-9 GeV positronsGreen-field design of ion
complex directly aimed at full exploitation of science program.
COMPASS Kick-Off Meeting p. 7SciDAC – September 17, 2007 – Fermilab
Dynamical Friction/Diffusion is Long in the Tooth
• Case of isotropic plasma, with no external fields, was first explained 65 years ago– S. Chandrasekhar, Principles of Stellar Dynamics
(U. Chicago Press, 1942).– B.A. Trubnikov, Rev. Plasma Physics 1 (1965), p. 105.– NRL Plasma Formulary, ed. J.D. Huba (2000).
– Physics can be understood in two different ways• Binary collisions (integrate over ensemble of e-/ion collisions)• Dielectric plasma response (ion scatters off of plasma waves)
e-
e-
e-
e-
COMPASS Kick-Off Meeting p. 8SciDAC – September 17, 2007 – Fermilab
Idea for Electron Cooling is 40 Years Old
• Budker developed the concept in 1967– G.I. Budker, At. Energ. 22 (1967), p. 346.
• Many low-energy electron cooling systems:– continuous electron beam is generated– electrons are nonrelativistic & very cold– electrons are magnetized with a strong solenoid field
• suppresses transverse temperature & increases friction
• Fermilab has shown cooling of relativistic p-bar’s– S. Nagaitsev et al., PRL 96, 044801 (2006).
– 4.3 MeV e-’s (≈8) from a customized DC source– electrons are unmagnetized (solenoid for focusing)
• RHIC II, eRHIC, ELIC need “high-energy” cooler– 100 GeV/n -> ≈108 -> 54 MeV bunched electrons– Cooling is less efficient; new parameter regime
COMPASS Kick-Off Meeting p. 9SciDAC – September 17, 2007 – Fermilab
tp
RHIC II Electron Cooling System
Image taken from “RHIC II White Paper,” http://www.bnl.gov/cad/ecooling/docs/PDF/RHIC%20II_White_Paper.pdf
Merminga, EIC2007, Apr 6-7 2007
ELIC Circulator CoolerELIC Circulator Cooler
ion bunch
electron bunch
Electron circulator ring
Cooling section
solenoid
kickerkicker
SRF Linac
dump
electron injector
energy recovery
COMPASS Kick-Off Meeting p. 11SciDAC – September 17, 2007 – Fermilab
VORPAL supports use of BETACOOL
• BETACOOL code is used to model many turns• A.O. Sidorin et al., Nucl. Instrum. Methods A 558, 325 (2006).• A.V. Fedotov, I Ben-Zvi, D.L. Bruhwiler, V.N. Litvinenko, A.O. Sidorin,
New J. Physics 8, 283 (2006).
– a variety of electron cooling algorithms are available• in particular, models for the dynamical friction force
– many mechanisms for emittance growth are included
• VORPAL is used to study microphysics of friction– to increase understanding & make BETACOOL more effective
COMPASS Kick-Off Meeting p. 12SciDAC – September 17, 2007 – Fermilab
Numerical Approaches for Electron Cooling Simulations
• Langevin approach to solve Fokker-Planck equation– uses Rosenbluth potential (or Landau integral)– proof-of-principle demonstrated; to be extended to e-/ion interactions
• Fast multipole method (FMM) and tree-based algorithms– requires constant time step; inefficient for MD with a few close collisions– being reconsidered as part of the Langevin approach
• 4th-order predictor-corrector “Hermite” algorithm– taken from astrophysical dynamics community– generalized to include solenoid field– used successfully in molecular dynamics (MD) approach with a few ions– didn’t parallelize well, so we used a task farming approach
• astrophysicists use special “Grape” hardware to parallelize• Semi-analytic binary collision model
– also MD approach; very close connection to “Hermite” algorithm above– accurately models arbitrarily strong Coulomb collisions– arbitrary external fields included via 2nd-order operator splitting– scales well up to ~128 processors; must be generalized for petascale
• Electrostatic particle-in-cell (PIC)– cannot directly capture close Coulomb collisions– will rely on PETSc/Aztec00 for effective use of petascale hardware– could be combined effectively with “binary collision” model
COMPASS Kick-Off Meeting p. 13SciDAC – September 17, 2007 – Fermilab
Self-Consistent Langevin Solution of the Fokker-Planck/Landau Equation
COMPASS Kick-Off Meeting p. 14SciDAC – September 17, 2007 – Fermilab
A Test Example Showing Temperature Exchange in a 2-Species System
COMPASS Kick-Off Meeting p. 15SciDAC – September 17, 2007 – Fermilab
The Parallel VORPAL Framework is used to Simulate the Microphysics of Electron Cooling
• Electromagnetic PIC for laser-plasma– Nieter & Cary, J. Comp. Phys. (2004).
• Electrostatic PIC for beams & plasma– Messmer & Bruhwiler, Comp. Phys. Comm. (2004)
• Algorithms for simulating electron cooling physics– Fedotov, Bruhwiler, Sidorin, Abell, Ben-Zvi, Busby, Cary & Litvinenko, Phys. Rev.
ST/AB (2006).– Fedotov, Ben-Zvi, Bruhwiler, Litvinenko & Sidorin, New J. Phys. (2006).– Bell, Bruhwiler, Fedotov, Sobol, Busby, Stoltz, Abell, Messmer, Ben-Zvi &
Litvinenko, “Simulating the dynamical friction force on ions due to a briefly co-propagating electron beam,” J. Comp. Phys., in preparation.
• SRF Cavities, Electron guns, Dielectric structures (PBG)…– Nieter et al., J. Comp. Phys., in preparation.– Dimitrov, Bruhwiler, Smithe, Messmer, Cary, Kayran & Ben-Zvi, Proc. ICFA Beam
Dynamics Workshop on Energy Recovery Linacs (2007), in press.– Werner & Cary, J. Comp. Phys., in preparation.
• Large software development team (Tech-X & CU)– C++/MPI, object-oriented, template techniques, multi-physics– parallel or serial; cross-platform (Linux, AIX, OS X, Windows)
• Actively used throughout the beam & plasma communities– BNL, JLab, Fermilab, LBL, ANL, some universities, also outside the USA– commercial customers
• Development and use has been supported by several agencies since 2000– US DOE Office of Science (HEP, NP, FES & ASCR)– NSF (original grant); DOD (AFOSR, OSD)
COMPASS Kick-Off Meeting p. 16SciDAC – September 17, 2007 – Fermilab
Molecular dynamics approach – model each e-
Diffusive spreading of ion trajectories obscures any velocity drag due to dynamical friction.
For many millions of turns, friction forces will dominate diffusion.
COMPASS Kick-Off Meeting p. 17SciDAC – September 17, 2007 – Fermilab
Diffusive dynamics can obscure friction/drag
• Numerical trick of e-/e+ pairs can suppress diffusion– idea came from Alexey Burov– simulate with e-/e+ pairs that have identical initial conditions
• sign of external fields must be flipped for the positrons
– friction force, independent of sign of charge, is unchanged– diffusive kicks are approximately cancelled
• also use ~1,000 trajectories for each electron– RMS is reduced by Ntraj
1/2 from that of the original distribution
• from the Central Limit Theorem
• Use of many trajectories sometimes changes results !!– always true for field free case:
– in presence of external fields• some other limitation on min may hide this problem
• also,higher effective e- velocity can decrease res
trajreserel
trajcsimres N
nv
NN
,
COMPASS Kick-Off Meeting p. 18SciDAC – September 17, 2007 – Fermilab
CLT is used to pull <F> and error bars from binned data
COMPASS Kick-Off Meeting p. 19SciDAC – September 17, 2007 – Fermilab
Diffusive dynamics of ions can be correctly modeled
COMPASS Kick-Off Meeting p. 20SciDAC – September 17, 2007 – Fermilab
Perturbative calculations lead to Coulomb log
• Assume each e- trajectory is infinite and straight– integrate Coulomb force along trajectory to obtain v~-1
• integrating over all angles leads to zero
– by symmetry, v|| = 0 for each trajectory
• however, energy conservation requires v|| ≈ - v2 / 2v ~-2
• Approximation is very good for large – assumption of infinite trajectories not valid for finite
– choose physically reasonable cutoff max
• Approximation breaks down for small – choose cutoff min, for which v = vrel 90 deg scattering
e-
v=vi-ve vv||
COMPASS Kick-Off Meeting p. 21SciDAC – September 17, 2007 – Fermilab
Dynamical friction & finite-time effects on max
• For pe >> 2 and vi << e
– electron cloud screens ion charge; max = D = e/pe
• In opposite limit, pe < 2 (RHIC II param’s)– no screening of ion charge; choose max ~ max(vi,e)
– or calculate Coulomb log with finite-length trajectories• completely removes logarithmic singularity at large
for
for min ln d
COMPASS Kick-Off Meeting p. 22SciDAC – September 17, 2007 – Fermilab
Finite-time effects limit # of collisions for small
• Poisson statistics predict likelihood of Nc collisions– for all impact parameters less than
ne
= Nc
COMPASS Kick-Off Meeting p. 23SciDAC – September 17, 2007 – Fermilab
Finite-time effects on min lead to concept of res
• Friction force integrals assume, for all – there are plenty of trajectories to sample 4 sr
• so perpendicular kicks average out• so longitudinal kicks accumulate correctly
– and sufficient trajectories to sample e- velocities
• How many collisions are needed?– good agreement with simulations for Nc = 120
– for RHIC II parameters shown above• min ≈ 2.2 e-7 m res ≈ 1.8 e-5 m
• for max ~ vi ≈ 7.5 e-4 m >20% smaller Coulomb log
COMPASS Kick-Off Meeting p. 24SciDAC – September 17, 2007 – Fermilab
Physical Parameters from 2006 RHIC II Design
COMPASS Kick-Off Meeting p. 25SciDAC – September 17, 2007 – Fermilab
Failure to sample small reduces friction force
yields choice Nc ≈ 120
COMPASS Kick-Off Meeting p. 26SciDAC – September 17, 2007 – Fermilab
Numerical effects of domain size are understood
COMPASS Kick-Off Meeting p. 27SciDAC – September 17, 2007 – Fermilab
Unmagnetized high-energy cooling w/ undulator
• Purpose of the helical undulator magnet– provides focusing for electrons– suppresses e-/ion recombination
• Modest fields (~10 Gauss) effectively reduce recombination via ‘wiggle’ motion of electrons:
• What’s the effect of ‘wiggle’ motion on cooling?– increases minimum impact parameter of Coulomb log
][][104.1~ 232
GBmxvk ww
beamw
gyroosc
resosc ,2max min
COMPASS Kick-Off Meeting p. 28SciDAC – September 17, 2007 – Fermilab
VORPAL simulations verify effect of undulator
• Coherent e- wiggle motion – increases the effective minimum impact parameter– dynamical friction force is only reduced logarithmically
COMPASS Kick-Off Meeting p. 29SciDAC – September 17, 2007 – Fermilab
Simple model of single-wavelength error fields
zBzyxB erry 2sin,,
2,, max err
z
y BzyxBdzM
TmMm/sxMm
ev
eerr
65 10for 102
m 80 cm 10 MBerr 2 TmxMTm 56 10210
ctcBtyxE errx 2sin,,' Lorentz transform to beam frame:
e- oscillations:
mTmMmxvc
xerr 1 ,10for 109 2
67err
COMPASS Kick-Off Meeting p. 30SciDAC – September 17, 2007 – Fermilab
Long wavelengths increase effective e- temp
COMPASS Kick-Off Meeting p. 31SciDAC – September 17, 2007 – Fermilab
Short wavelengths increase effective min
COMPASS Kick-Off Meeting p. 32SciDAC – September 17, 2007 – Fermilab
Key Questions for Electron Cooling Simulations
• Electron cooling is key to future NP accelerators– never demonstrated for high-energy ions– cooling effectiveness may be “just enough”– simulations must help to improve design & reduce risk
• How best to suppress e-/ion recombination?– strong solenoid (used in all low-energy coolers)– helical undulator magnet (recent alternative concept)
• Quantify all phenomena that weaken cooling– e- wiggle motion in undulator– wide variety of magnetic field errors– high density of ions, complicated e- distributions, …
• New ideas must be supported– concept of “coherent electron cooling”– low-energy cooling for RHIC and/or RHIC II
COMPASS Kick-Off Meeting p. 33SciDAC – September 17, 2007 – Fermilab
Courtesy of V. Litvinenko & Y. Derbenev (FEL 2007)
COMPASS Kick-Off Meeting p. 34SciDAC – September 17, 2007 – Fermilab
Courtesy of V. Litvinenko & Y. Derbenev (FEL 2007)
COMPASS Kick-Off Meeting p. 35SciDAC – September 17, 2007 – Fermilab
Courtesy of A. Fedotov et al., COOL ’07 Presentation (Sep. 14, 2007)
COMPASS Kick-Off Meeting p. 36SciDAC – September 17, 2007 – Fermilab
Courtesy of A. Fedotov et al., COOL ’07 Presentation (Sep. 14, 2007)
COMPASS Kick-Off Meeting p. 37SciDAC – September 17, 2007 – Fermilab
Parallel Scaling for Electron Cooling Simulations
Execution times for Trillinos-based Poisson solve of a 3D Gaussian beam, for a 1026×65×65 grid (solid) and 4104×65×65 grid (dotted), using AMG precon-ditioned CGS (diamond) or Gauss-Seidel preconditioned CGS (stars).
To date, sim’s have been confined to ~256 processors or less.
COMPASS Kick-Off Meeting p. 38SciDAC – September 17, 2007 – Fermilab
Challenges for Petascale Cooling simulations
• Petascale path for MD involves move to PIC– O(Nion*Ne) scaling of pure MD limits problem size– hybrid electrostatic PIC/MD model will be considered
• PIC efficiently captures distant interactions– pure PIC will be useful, if min is bounded from below
• e.g. L, xerr, res
• Input needed from CETs– use of PETSc or Trilinos to rapidly solve Poisson eq.
• path to petascale for PIC and/or hybrid PIC/MD• VORPAL uses Trilinos for electrostatic PIC now
– plan to implement PETSc as an option– assistance in scaling and speed for other algorithms
• particle push, charge deposition, etc.– access to additional computing resources?
• will request 10x-50x larger allocation from NERSC this year– visualizing details of the electron wake
• one solution is IDL + AVS (used successfully in the past)• will try to benefit from ongoing work with VORPAL & VisIT
COMPASS Kick-Off Meeting p. 39SciDAC – September 17, 2007 – Fermilab
Future Plans for Langevin Approach
• Extend the Langevin approach to electrons and ions• Include effects of magnetic fields in the calculation of
Fokker-Planck collisional operator• Explore an efficient FMM-based technique to compute
the F-P collisional operator• Implement improved time integration schemes for
electrons and ion dynamics• Integrate the Fokker-Planck solver with our other
modules for multi-physics modeling
COMPASS Kick-Off Meeting p. 40SciDAC – September 17, 2007 – Fermilab
Acknowledgements
We thank O. Boine-Frankenheim, A. Burov, A. Jain, S. Nagaitsev, A. Sidorin, G. Zwicknagel & members of the Physics group of the RHIC Electron Cooling Project for many useful discussions.
We acknowledge assistance from the VORPAL team: J. Carlsson, J.R. Cary, B. Cowan, D.A. Dimitrov, A. Hakim, P. Messmer, P. Mullowney, C. Nieter, K. Paul, S.W. Sides, N.D. Sizemore, D.N. Smithe, P.H. Stoltz, S.A. Veitzer, D.J. Wade-Stein, G.R. Werner & N. Xiang.
Work at LBNL was supported by SciDAC-1.
Work at BNL and JLab was supported by the U.S. DOE Office of Science, Office of Nuclear Physics.
Work at Tech-X Corp. was supported by the U.S. DOE Office of Science, Office of Nuclear Physics under grants DE-FG03-01ER83313 and DE-FG02-04ER84094. We used computational resources of NERSC, BNL and Tech-X Corp.