Code indexing in gitaly is broken and leads to code not being visible to the user. We work on the issue with highest priority.

Skip to content
Snippets Groups Projects

optpilot week

Last edited by Jochem Snuverink

AWA Drive Linac G2 Optimisation PSI Ring Turnseparation Superconducting Optimisation

AWA Drive Linac

beamline_pic

Input files: optLinac.dataoptLinac.tmpl

10 design variables

  • Laser radius
  • FWHM of laser pulse
  • Solenoid strength
  • Gun phase
  • Linac cavity phase L1-L6

Emittance (emitx) and bunch length (rms_s) are recorded/optimized at 12.51 m.

optLinac-1

Reduced problem in order to have one forward solution in less than 2 minutes:

  • Cores= 128
  • Gens = 1000
  • Indiv = 127
  • Grid = 8x8x16
  • Np = 5000
  • T1 = 5e-13
  • T2 = 5e-12
  • Total time for optimisation: 1271 (s)

Generation 1

optLinac1_gen1 Generation 999

optLinac1_gen100 optLinac-2

  • Cores= 128
  • Gens = 1000
  • India = 127
  • Grid = 16x16x32
  • Np = 10000
  • T1 = 5e-13
  • T2 = 5e-12
  • Total time for optimisation: 4157 (s)

Generation 1

optLinac2_gen1 Generation 999

optLinac2_gen999

optLinac-3

  • Cores= 128
  • Gens = 100
  • India = 127
  • Grid = 16x16x32
  • Np = 50000
  • T1 = 5e-13
  • T2 = 5e-12
  • Total time for optimisation: 1995.01 (s)

Generation 1

optLinac3_gen1 Generation 100

optLinac3_gen100

G2 Optimisation

Gantry

7 design variables

  • 7 quadrupole field gradients (all gantry 2 quadrupoles)

To optimize are the beam size (and the transmission) at the gantry isocenter (52.165m). The simulation started with a pre defined beam and also included the last quadrupole triplet (untouched settings) before the gantry (31.5926 - 52.50m).

  • Cores= 64
  • Gens = 100
  • Initial population = 63
  • Final population = 125
  • NumParticles = 8000
  • Timestep = 1e-11s

Generation 1

g2-small-1 Generation 100

g2-small-100

Generation 100 with constraint

constrs: CONSTRAINTS = (abs(drmsx-drmsy)>0.001); g2-small-100-w-constr

Results in comparison with opal to the previous "by hand" tune.

Envelopes: envelopes_1_49.pdf

Transmission: transmission_1_49

Isocenter:

IsoMon_LeftTune_RightOptPilot_1 Left: Tune by hand Right: OptPilot tune

Including Transmission

Runtime each Simulation: 51s

  • 100 Generation: rmsy over rmsx @Isocenter: rmsx_rmsy_100gen

NumParticles over rmsx @Isocenter: rmsx_nP_100gen

  • 1000 Generation: rmsy over rmsx @Isocenter: rmsx_rmsy_1000gen

NumParticles over rmsx @Isocenter: rmsx_nP_1000gen

NumParticles over rmsx @Isocenter (zoom): rmsx_nP_1000gen_zoom

PSI Ring Turnseparation

The PSI Ring parameters are not well-known. The goal of this study is to find good simulation parameters that match the measurements. As objectives we use the radial turn locations. As optimisation parameters we use the injection angle, radius, energy and radial momentum.

The OPAL probe code has been extended to find the radial turn locations in the same way as in the measurement, see figure below.

RRI2 RRI2 probe simulation histogram (for 5000 particles) with found peak locations.

RRI2-0038Y17 RRI2 probe measurement with peak locations.

Test 1

First trial of optimizing three objectives:

  • difference between simulation and measurement of peak 1
  • difference between simulation and measurement of peak 2
  • difference between simulation and measurement of peak 3 up to 5

Settings

  • 1000 generations,
  • 32 cores
  • 500 particles
  • no space charge
  • 7 turns
IF ( _optpilot_ ) {
    // ---- OPTIMIZER SECTION -------
    //dphiinit: DVAR, VARIABLE="phiinit", LOWERBOUND="106", UPPERBOUND="114";
    //drinit: DVAR, VARIABLE="rinit", LOWERBOUND="2000", UPPERBOUND="2060";
    //dprinit: DVAR, VARIABLE="prinit", LOWERBOUND="-0.02", UPPERBOUND="-0.01";
    //dbenergy: DVAR, VARIABLE="benergy", LOWERBOUND="0.071", UPPERBOUND="0.072";
    //dvars: DVARS=(dphiinit,drinit,dprinit, dbenergy);
 
    //dpeak1:OBJECTIVE,EXPR="fabs(radialPeak("RRI2.peaks", 1) - radialPeak("RawData/RRI2-0038Y17.peaks", 1))";
    //dpeak2:OBJECTIVE,EXPR="fabs(radialPeak("RRI2.peaks", 2) - radialPeak("RawData/RRI2-0038Y17.peaks", 2))";
    //dpeak3_5:OBJECTIVE,EXPR="sumErrSqRadialPeak("RRI2.peaks", "RawData/RRI2-0038Y17.peaks", 3, 5)";
 
    //objs: OBJECTIVES=(dpeak1,dpeak2,dpeak3_5);
    //constrs: CONSTRAINTS = ();
    //opt: OPTIMIZE, OBJECTIVES = objs, DVARS = dvars, CONSTRAINTS = constrs;
 }

dpeak1

Sum of all differences (simulation - measurement) of peak 1 per generation.

dpeak2

Sum of all differences (simulation - measurement) of peak 2 per generation.

dpeak3_5

Sum of all differences (simulation - measurement) of objective 3 per generation.

We see a nice decreasing curve for peak 1. Also the curve for peak 2 is mainly decreasing. However, the third objective is also increasing over the generations.

Superconducting Optimisation

Simple setup one dipole. Shift initial distribution in x-y plane and search for minimal emittance. Run 1000 generations.

Generation 1

g3-small-1

Generation 100

g3-small-100

Generation 999

g3-small-999

Open questions and new runs

  • Possibility to run optPilot while keeping the input files and maps on afs -> Achim is checking
  • Read from a file: structure for optPilot
# Element PROB_2 x (mm),  y (mm),  z (mm),  px ( ),  py ( ),  pz ( ), id,  turn,  time (ns) 
PROB_2   0.618627   2325.63   150   0.848461   -0.062113   0   2   1   16.1428 
PROB_2   0.606715   2300.1   150   0.741857   -0.000575306   0   1   1   17.0473 
PROB_2   0.697536   2494.72   -832.791   0.534304   0.320705   -0.402196   0   1   18.6247
  • New run:
//dv0: DVAR, VARIABLE="BEAM_RINIT", LOWERBOUND="-900", UPPERBOUND="-1100";
//dv1: DVAR, VARIABLE="FIELD_UNITS", LOWERBOUND="8.0",  UPPERBOUND="11.0";

//dvars: DVARS=(dv0,dv1);

//oey:OBJECTIVE,EXPR="fabs(sameSDDSVariableAt("rms_x",7.05) - 0.005)";
//oes:OBJECTIVE,EXPR="fabs(sameSDDSVariableAt("rms_s",7.05) - 0.005)";