Presentation is loading. Please wait.

Presentation is loading. Please wait.

Update on Modeling Activities Gun Simulation Optimization Using Genetic Algorithms.

Similar presentations


Presentation on theme: "Update on Modeling Activities Gun Simulation Optimization Using Genetic Algorithms."— Presentation transcript:

1 Update on Modeling Activities Gun Simulation Optimization Using Genetic Algorithms

2 Gun Simulation EGun (EGN2) 2k or 4k usd (?) Keerthi is interested in upgrading his IGUN. Arcane version: 575 usd from ESTSC. Fortran (on DOS?) Aether/Xenos (inc. Omnitrak) Suite of tools (Mesh generation, field, particle, ……) Got trial copy and set up initial environment on PC Suite tried ~10k usd Does more than e-gun; Some quirks GPT from Pulsar.nl – from several sources User friendly Time domain capability? 2800 usd per license (non-profit, 50% discount for additional) No reply to inquiries – Quality of support?

3 Gun Simulation Some Tests with Xenos

4 Optimizing in Global Parameter Space Is it worth doing (e.g., genetic algorithm a la Cornell ERL/Bazarov)? Real gain in quality of solution (Cornell example)? Laser spot reduced 50%;  x reduced 6-fold over proposed design (  z unchanged ) No pronounced single responsible factor Global optimum not easily accessible by local methods May be a good time to explore global parameters (options not too frozen) Evidence of benefit in Cornell case Started from existing design parameters Solution has smooth dependence over wide range (no structure) Should not compare options optimized to different degrees Example: Thermal energy effect on emittance vs space charge effect Advantages of the method itself Collection of viable solutions, not homing in on one solution (genetic or not) Insight into interplay between multiple objectives More rigorous handling of hard inequality constraints Convince everyone that parameter space has been sufficiently explored Infrastructure may be useful later

5 Optimizing Cornell ERL Injector DC photo gun + solenoids + buncher + SRF cavities Varying 22 parameters (laser profile, gun voltage, solenoid, RF phase, position ……) Examine different bunch charge scenarios under multiple constraints RF field calculated by SuperLANS Interplay between multiple objectives (  x,  z,  p …) in optimized configurations Genetic algorithm (PISA) driving Astra running on parallel cluster Pareto front constraining multiple objectives Single merit function with one combination of objectives Single merit function with another combination Single merit function ends up picking different points on the Pareto front. Pareto fronts

6 Sign of emittance compensation at 15 MeV Did the optimization code find it? In any case, it reduced emittance by a factor of 6, with no compromise to bunch length, over already “optimized” design.

7 Other points worth noting and with implications Parmela used to cross check optimized solutions. Agreement within factor of 2 in emittance. Bazarov’s further comments: Parmela is more unstable than Astra, and GPT is replacing Astra (?) there. Guidance with physical insight is needed. Computer-savvy is useful. What’s happening now Obtained Bazarov’s codes and sample files Gabriel tapped (compile code, test run samples) C++ installed on the Linux HLA server; Need to install Astra. What’s next Evaluate the case (resource, learning curve, time frame, payoff) Where/How to run it? HLA server ? Cluster upstairs:Off limits WestGrid?May need boost in priority License issues? Computation demand should be much less than Cornell case.

8 Latest on GPT Time domain tracking?Yes Accepted inputs & computation included Standard parameterized components (solenoid, RF, …) Field maps or analytical field expressions User defined components (?) Space charge force included (no wake field) Simple image charge force GUI mode or executables capable of massive parameter scan Utility modules for handling parameter scan and analyzing outcome Other License tied to process & user, not number of CPUs. No trial versions Note GPT does not do external EM field calculation itself.


Download ppt "Update on Modeling Activities Gun Simulation Optimization Using Genetic Algorithms."

Similar presentations


Ads by Google