Electromagnetic field simulations for accelerator optimization

Alexej Grudiev CERN, BE-RF

1st oPAC Workshop: Grand Challenges in Accelerator Optimization, 26-27 June 2013, CERN

Outline

• A review of tools for electromagnetic field simulations used at CERN will be given. • A number of examples of its usage will be presented covering several areas of

accelerator component design and optimization both RF and non-RF equipment: • accelerating cavities, • collimation devices, • etc.

• A brief review of CLIC main linac RF frequency and accelerating gradient optimization will be given as an example of incorporating electromagnetic field simulation into a global optimization process including both

• constraints coming from the beam dynamics simulations and • the empirical RF constraints related to the high gradient linac operation.

CST

HFSS

ACE3P

GdfidL

Packages for computer simulations of electromagnetic EM fields and more

CST Studio Suite

CST STUDIO SUITE:- CST MWS- CST DS- CST EMS- CST PS- CST MPS- CST PCBS- CST CS- CST MICROSTRIPES- Antenna Magus

CST: All you need in one package•Powerful and user-friendly Input:•Probably the best time domain (TD) solver for wakefields or beam coupling impedance calculations (MAFIA)

• Beta < 1• Finite Conductivity walls

•Once geometry input is done it can be used both for TD and FD simulations •Moreover using Design Studio (DS) it can be combined with the other studios for multiphysics and integrated electronics simulation, but this is relatively fresh fields of expertise for CST •Accelerator physics oriented post processor, especially in MWS and PS•Enormous progress over the last few years compared to the competitors.

Courtesy of Igor Syratchev

An example of what can be solved easily on a standard PC

CST (examples)Two examples of what can be solved on bigger PC: 128 GB of RAM and 24 CPUs

CLIC accelerating structure from Cu with HOM damping loads from SiC(frequency dependent lossy material)

Giovanni De Michele

CST (examples)Transverse wake at offset of 0.5 mm

Transverse beam couping impedance at offset of 0.5 mm

Zx [Ω]

f [GHz]

s [mm]

Wx [V/pC]

CST (examples)LHC TDI 5m long with ferrite

Benoit Salvant

CST MWS: Example. S-parameters in CLIC Crab cavity

Mesh view

Praveen Ambattu

CST: Shortcomings1. Cartesian mesh: Especially in FD can results to less

accurate calculations of frequency, Q-factor, surface fields compared to tetrahedral mesh (HFSS, ACE3P). => if possible use tetrahedral mesh which became available recently and essentially gives the same results as FEM codes (HFSS, ACE3P).

2. Boundary conditions can be set only in Cartesian planes3. No Field Calculator (HFSS)4. ...

HFSS: Still an excellent tool for FDHigh-Performance Electronic Design

Ansoft DesignerANSYS HFSSANSYS Q3D ExtractorANSYS SIwaveANSYS TPA

Electromechanical DesignANSYS MultiphysicsANSYS MaxwellANSYS SimplorerANSYS PExprtANSYS RMxprt

Product optionsAnsoftLinks for ECADAnsoftLinks for MCADANSYS Distributed SolveANSYS Full-Wave SPICEANSYS OptimetricsANSYS ParICs

•HFSS was and I think still is superior tool for FD simulations both S-pars and eigenmode, though CST shows significant progress in the recent years•Automatic generation and refinement of tetrahedral mesh•Most complete list of boundary conditions which can be applied on any surface•Ansoft Designer allows to co-simulate the pick-up (antenna), cables plus electronics and together with versatile Optimetrics optimise the design of the whole device •Recently HFSS became an integral part of ANSYS – reference tool for thermo-mechanical simulations -> multiphysics•REcently time-dependent solver has been released

HFSS (examples, eigenmode)

LHC TDI 5m long beam dump: One of the most dangerous eigenmodes at 1.227 GHz, Q = 873, Tetrahedral mesh with mixed order (0th , 1st , 2nd) elements: Ntetr = 1404891 Solution obtained on a workstation with 128 GB of RAM,

0 20 40 60 80 100 120 140 160 180 2000

0.2

0.4

0.6

0.8

Time [ns]

Tim

e R

es

po

ns

e [

V W

-1

/2/s

]

0 20 40 60 80 100 120 140 160 180 2000

20

40

60

80

Wa

ke

Po

ten

tia

l [V

/ p

C]

11.5 11.6 11.7 11.8 11.9 12 12.1 12.2 12.3 12.4 12.50

1

2

3

4

Frequency [GHz]

Fre

qu

en

cy

Re

sp

on

se

[k

V W

-1

/2]

11.5 11.6 11.7 11.8 11.9 12 12.1 12.2 12.3 12.4 12.50

1

2

3

4

Be

am

Im

pe

da

nc

e [

V /

A

]

HFSS (example, S-parameters)Port excitation Incident plane wave

excitation

Inverse FFT

O. KononenkoO. Kononenko

HFSS example

HFSS: shortcomings

1. No possibility to simulate particles2. Automatic mesh is not always perfect, but it

has improved after adoption by ANSYS3. TD and multiphysics are only recently

implemented, but thermo-mechanics from ANSYS is a reference by itself

4. ...

GdfidL: Parallel and easy to use toolbruns@gdfidl.de The GdfidL Electromagnetic Field simulator GdfidL computes electromagnetic fields in 3D-structures using parallel or scalar computers. GdfidL computes •Time dependent fields in lossfree or lossy structures. The fields may be excited by

• port modes, • relativistic line charges.

•Resonant fields in lossfree or lossy structures. •The postprocessor computes from these results eg. Scattering parameters, wake potentials, Q-values and shunt impedances. Features •GdfidL computes only in the field carrying parts of the computational volume. •GdfidL uses generalised diagonal fillings to approximate the material distribution. This reduces eg. the frequency error by about a factor of ten. •For eigenvalue computations, GdfidL allows periodic boundary conditions in all three cartesian directions simultaneously. •GdfidL runs on parallel and serial computers. GdfidL also runs on clusters of workstations. Availability •GdfidL only runs on UNIX-like operating systems.

GdfidL (example)

CLIC accelerating structure from Cu with HOM damping loads from SiC(frequency dependent properties)

GdfidL: shortcomings

1. Available only under UNIX-like systems2. Geometry input is limited. Often other 3D

input tools have to be used.3. ...

ACE3P

ACE3P: example

Arno Candel et. al., SLAC-PUB-14439

CLIC two-beam module rf circuit

PETS

AS

AS

waveguide

ACE3P: shortcomings

• Very complex package to use. It is not user-friendly at all and requires a lots of time to invest before it can be used efficiently

• It is not a commercial product -> no manual reference, limited tech support. No it is an open source.

• ...

Summary for the EM simulation toolsCST1.

2.

3.

GdfidL ANSYS HFSS

ACE3P

Larger objects in TD Better FD calculations,3D EM + circuit co-simulation,

RF + thermal + structural

Accurate solution for very larger objects in TD and FD

C L I CC L I C

Alexej Grudiev, CLIC main linac structure optimization.

CLIC main linac accelerating structure optimization.

Brief Review of what was done back in 2007

C L I CC L I C

Alexej Grudiev, CLIC main linac structure optimization.

General layout of CLIC at 3 TeV

More on CLIC : http://clic-study.org/

C L I CC L I C

Alexej Grudiev, CLIC main linac structure optimization.

Optimization procedure

Bunch population

Structureparameters

Cell parameters

Bunch separation

BD

<Ea>, f, ∆φ, <a>, da, d1, d2

N

Ns

Q, R/Q, vg, Es/Ea, Hs/Ea Q1, A1, f1

BD

η, Pin, Esmax, ∆Tmax

Ls, Nb

rfconstraints

Cost function minimization

YES

NO

C L I CC L I C

Alexej Grudiev, CLIC main linac structure optimization.

Optimization parameter space

All structure parameters are variable:

<Eacc> = 90 – 150 MV/m,

f = 10 – 30 GHz,

Δφ = 120o, 150o, <a>/λ= 0.09 - 0.21, Δa/<a> = 0.01 – 0.6, d1/λ= 0.025 - 0.1, d2 > d1

Ls = 100 – 1000 mm.

N structures:71422460614--------------68.866.560

C L I CC L I C

Alexej Grudiev, CLIC main linac structure optimization.

Structure parameter calculation

Fundamental mode:

2

1'PI

Q

R

vP

Qvdz

dP

gg

Dipole mode:

cells

i

iN

ii

Q

t

it teAW1

12

1 )sin(' 1

1

Ns

η, Pin, Esmax, ∆Tmax

P(z)

NI

From 3D simulation (few hours) to an adequate model (few seconds)

C L I CC L I C

Alexej Grudiev, CLIC main linac structure optimization.

Cell parameter calculation

11,da

22 ,da

a/λ0.7 1.5

2.30.1

0.25

0.4

Single cell parameter interpolation

d/λ

WDS 2 cells

Q, R/Q, vg, Es/Ea, Hs/Ea

C L I CC L I C

Alexej Grudiev, CLIC main linac structure optimization.

Optimization constraints

Beam dynamics (BD) constraints based on the simulation of the main linac, BDS and beam-beam collision at the IP:

• N – bunch population depends on <a>/λ, Δa/<a>, f and <Ea> because of short-range wakes

• Ns – bunch separation depends on the long-range dipole wake and is determined by the condition:

Wt,2 · N / Ea < 10 V/pC/mm/m · 4x109 / 150 MV/m

D. Schulte

RF breakdown and pulsed surface heating (rf) constraints:• ΔTmax(Hsurf

max, tp) < 56 K

• Esurfmax < 380 MV/m

• Pintp1/3/Cin = 18 MW·ns1/3/mm @ X-band

C L I CC L I C

Alexej Grudiev, CLIC main linac structure optimization.

Optimizing Figure of Merit

Luminosity per linac input power:

N

L

EefNNEe

fNL

P

L b

cmrepbc

repbb

l

1

Collision energy is constant

Figure of Merit (FoM = ηLbx/N)

C L I CC L I C

Alexej Grudiev, CLIC main linac structure optimization.

Parametric Cost Model

Total cost = Investment cost + Electricity cost for 10 years

Ct = Ci + Ce

Ci = Excel{fr; Ep; tp; Ea ; Ls ; f ; Δφ} Repetition frequency; Pulse energy; Pulse length; Accelerating gradient; Structure length (couplers included); Operating frequency; rf phase advance per cell

Ce = (0.1011+7.1484/FoM)/12 [a.u.]

Figure of Merit (ηL/N) in a.u. (the same as before) [a.u.]=[1e34/bx/m2•%/1e9]

Hans Braun, 2006

C L I CC L I C

Alexej Grudiev, CLIC main linac structure optimization.

CLIC performance and cost versus gradient

• Performance increases with lower accelerating gradient (mainly due to higher efficiency)• Flat cost variation in 100 to 130 MV/m with a minimum around 120 MV/m

Ecms = 3 TeV L(1%) = 2.0 1034 cm-2s-1

Previous PreviousNew New Optimum

Performance Cost

C L I CC L I C

Alexej Grudiev, CLIC main linac structure optimization.

CLIC performance and cost versus frequency

Ecms = 3 TeV L(1%) = 2.0 1034 cm-2s-1

• Maximum Performance around 14 GHz • Flat cost variation in 12 to 16 GHz frequency range with a minimum around 14 GHz

New NewPrevious PreviousOptimumOptimum

Performance Cost