Denis Perret Gallix graceful acfa2007

Slide1: 

M. Werlen D. Perret-Gallix FJPPL IN2P3-CNRS/KEK Minami-Tateya Group

What is Grace ?: 

What is Grace ? Cross-section Automatic Computation System for: Tree level One-loop level SM and MSSM Generator of “event generators” Bases/Spring framework Used at LEP I, II and targeting LHC, ILC physics and astro-particle calculations Developed by the Minami-Tateya Group (based in KEK)

Slide3: 

User input Theory model file Diagram generator amplitude generator Library CHANEL, loop Kinematics library kinematics code generated code diagram description convergence information Make file etc. symbolic code REDUCE, Form etc. PS file Drawer BASES(MC integral) SPRING (EG manager) parameter file Diagrams (figure) Events Cross sections distributions TREE LOOP FORTRAN code .fin .mdl .rin © Minami-Tateya (KEK)

Slide4: 

Systematic R.C. to the Higgs production in ILC w/ GRACE Tree level e+e-→ννHH © Minami-Tateya (KEK)

Slide5: 

GRACE/SUSY-loop project Systematic calculation of R.C. to the two-body decay of charginos Checked with the non-linear gauge invariance © Minami-Tateya (KEK) hep/ph 0701200 A) Tanβ 10. μ 400. M1 100. M2 197. M3 610. MA0 425. ~ SPA1a’

R.C. to the three-body decay of charginosGRACE/SUSY-loop: 

R.C. to the three-body decay of charginos GRACE/SUSY-loop © Minami-Tateya (KEK) B) Tanβ 10. μ 400. M1 100. M2 157. M3 610. MA0 431. ~ SPA1a’

Chargino x-sections: 

Chargino x-sections

Main issues: 

Main issues Complex procedures, Many Interfaces to ext. programs “GraceFUL” Project (Grace For U to Love) CPU/Memory performances GRID, clusters, Supercomputers, Feynman@Home High Arithmetic accuracy (beyond Double Precision) HAPPY (High Arithmetic Precision Processing Yoke) 3 projects

GraceFUL: 

GraceFUL Front-end package to Grace for: Simple individual use No GRACE code knowledge required to build integ/spring code Cover all actions from process selection to parameter dependent cross-section and event generation Interface to beamstrahlung and parton shower/hadronization Gather all information on a single spreadsheet System wide massive production system Local or distributed, private or public computing system: Supercomputer, cluster, GRID, Feynman@Home For all Grace packages SM, MSSM Tree level, 1-loop Generic processes (i.e. e+-e-->lepton-lepton-H or pp->4 jets)

Perl driven: 

Perl driven Build and manage configuration and parameter files Direct edition of XML files through XML editor/viewer (Amaya) Build and manage a directory tree for: Codes Configurations (sets of parameters) Results Build the parameter dependent GRACE code The “integ/spring” binaries No input file. Parameters hardwired (GRACE policy) Automates the Interface to Pythia, PDF, Circe (Beamstrahlung) Output Spring-> Pythia records Prepares the supporting Pythia (upinit and upevent code) Run and manage the Jobs Local run on user PC submit jobs to Supercomp, the GRID or Feynman@Home(PBS/LSF) Analyze/display the output, keep track of the results Summary through SpreadSheet Interface to Root: plots and ntuples for analysis Store all codes and results in a directory tree  a database

Perl driven code modification: Templates and XML: 

Perl driven code modification: Templates and XML Templates: kinit.tmpl w= [% w %] Data: default <w> 250.D0 </w> kinit.f, setmas.f, gfinit.f …. i.e. in kinit.f: w = 250.D0 Data: User selected <w> 500.D0 </w> } Merge Data: configuration file Instantiation New kinit.f file Integ, spring Makefile

Directory Tree for Code: 

Directory Tree for Code

Directory Tree for Running sets: 

Directory Tree for Running sets

Parameters in XML format: 

Parameters in XML format Single value: <p1> x </p1> x can be a number (or a Fortran expression using Grace variables (for expert only) Range: <p1> xmin : xmax : step ; order </p1> Order: 0…4 (0 inner loop) List: <p1> x1, x2, ….., x3 ; order </p1> Currently at most 5 (Range +List) parameters Examples: <p1> 250.d0 </p1> Single value <p1> 250.:400.:10.;0 </p1> from 250. to 400. by step of 10. <p1> 250.,300.,1000.,2000.;1 </p1> list of values

Slide15: 

e+ e-  jet, jet

Sample XML default parameter file build by GraceFUL: 

Sample XML default parameter file build by GraceFUL … <agz> 2.49D0 </agz> <agh> 100.0D0 </agh> <agx> AGW </agx> <agy> AGZ </agy> … <jhs1> 0 </jhs1> <jhe1> lextrn - 1 </jhe1> <jhs2> 0 </jhs2> <jhe2> lextrn - 1 </jhe2> <jhs3> 0 </jhs3> <jhe3> lepexa - 1 </jhe3> <jhs4> 0 </jhs4> <jhe4> lepexa - 1 </jhe4> <jgluon > 1 </jgluon > <jhiggs > 1 </jhiggs > </option> <option> <p1> 250.d0 </p1> <p2> 250.d0 </p2> <coscut1> -1 </coscut1> <coscut2> 1 </coscut2> <icost> 0 </icost> <itmx1> 5 </itmx1> <itmx2> 5 </itmx2> <ncall> 5000 </ncall> <pi> acos(- 1.0d0 ) </pi> <pi2> pi * pi </pi2> <rad> pi / 180.0d0 </rad> <gevpb> 0.38937966d9 </gevpb> <alpha> 1.0d0/128.07d0 </alpha> <alpha0> 1.0d0/137.0359895d0 </alpha0> <alphas> 0.12d0 </alphas> <amw> 80.22D0 </amw> <amz> 91.187D0 </amz> <ama> 0.0D0 </ama>

Slide17: 

W: 500 -> 1000 step 100 GeV cos(θ): -1 -> -0.1 step 0.1 and

Slide18: 

e+ e-  jet, jet

In progress: 

In progress Interface to the “Les Houches Accords” Extension to other packages Grcft, a new fast EW tree level Grace system Grace 1-loop Objective-Perl ? Already more than 5000 Perl lines

Feynman@homeVolonteer Computing For Particle Physics: 

Feynman@home Volonteer Computing For Particle Physics BOINC Distributed Public Computing Berkeley Open Infrastructure for Network Computing http://boinc.berkeley.edu/ Follow-up of SETI@HOME http://setiathome.berkeley.edu/ Feynman@Home http://acpp.in2p3.fr/cgi-bin/twiki/bin/view/Feynman/WebHome

Slide22: 

1 credit=1/100 cpu PC hour 222 M CPU Hours 478,000 CPU Hours/day ~ 20000 CPUs full time Jan. 30 2007 39 projects

Slide23: 

Last 2 years Jan 30 2007 Large CPU power: 20,000 CPU and growing BUT Low reliability: redundant computations Not for time critical application Complementary to the GRID 222 M hours 914 K users

Feynman@HomeExploratory stage: 

Feynman@Home Exploratory stage Target and goal Public or/and Organization (KEK, IHEP, … Companies…) deployment Cross-section first then event generation Two applications: Small executables i.e.: 2->2,3,4 (100-1000 diag.) 100-500 Mb One set of processes/ many different parameters i.e. multi dimensional parameter phase space exploration (MSSM) Huge executables i.e.: 2->5…8, 1-loop (5,000-100,000 diag.) 10-50 Gb split the binaries into 100 small subsets each of 100-500 Mb. Each subset run in // on client PC The server run the integration algorithm At each iteration generate a new set of phase space points Hybrid system: BOINC + cluster/GRID Load balancing private cluster or the GRID

Feynman@Home: 

Feynman@Home International Collaboration France, KEK, CERN, … Feynman@home server operational in KEK, KEK intranet, no HEP application running yet Important Outreach for promoting LC and particle physics

Slide28: 

BDP (Beyond Double Precision) Quadruple/octuple precision is needed. Correct results. Faster algorithms. But software implementations are too slow. New hardware/software development needed.

Simple Exampleby J. Fujimoto (KEK): 

Simple Example by J. Fujimoto (KEK) f = 333.75 b6 + a2(11a2b2- b6- 121b4 – 2) + 5.5b8 + a/2b where a=77617.0, b=33096.0. (C. Hu, S. Xu and X. Yang) Double Precision f = 1.17260394005317863 Quadruple precision result f = 1.1726039400531786318588349045201801 Analytical result = - 54767/66192 f = - 0.82739605994682136814116509547981370 New Octuple precision library, H3Lib: f = - 0.827396059946821368141165095479816 lost bits = 121

Actual applicationBy J. Fujimoto KEK: 

Actual application By J. Fujimoto KEK Quadruple precision is required in some phase space points due to the Gram determinant happens in the reduction algorithm. t mass of mass of photon

Slide31: 

ReJ[1] = -1.49368718239238 ReJ[x] = - 6.86111482424926E-0002 ReJ[y] = - 6.86785270067264E-0002 ReJ[w] = - 1.39799775179174 ReJ[w**2] = - 1.36472026946296 ReJ[w*x] = - 2.708863236843683E-0002 ReJ[x*y] = - 3.048903558925384E-0002 … ReJ[w**3] = 93763.26727997246 … ReJ[1] = - 1.49368718238777512062307539882045 ReJ[x] = - 6.861114708877389206553392789958382E-0002 ReJ[y] = - 6.867852585600575199171661642779842E-0002 ReJ[w] = - 1.39799775496536042464289674154150 ReJ[w**2] = - 1.34746346742190735627641191119128 ReJ[w*x] = - 3.334744118868393382280835719751654E-0002 ReJ[x*y] = - 2.822377826411337874789947823777159E-0002 … ReJ[w**3] = - 1.60389378482142986480454883491878 … Double precision Quadruple precision Blow up !! J. Fujimoto

Slide32: 

Minimization algorithms, … Gambolatin et al. Single Double Quad.

Slide33: 

CPU bandwidth (Ghz) Memory size (Gbytes) Interconnection bandwidth (Ghz) Floating point precision (4-32 bytes) Instruction size (64 bits)

Slide34: 

High precision libraries quadruple/octuple (Hitatchi) Double-double, quad-double (Arprec) Multi-precision lib. (1000 digits and more) Interval arithmetic Exact arithmetic (XR, iRRAM) Linpack double/quad: 30 times slower High precision Arithmetic Parallel Processor Yoke HAPPY based on CELL processor (IBM,Sony,Toshiba) complex programing Investigating other possibilities

Grace SimulationSummary: 

Grace Simulation Summary Grace is producing tools for tree and one-loop SM and MSSM x-section calculations and event generation. (i.e. hep/ph 0701200) 3 Projects to overcome the computational and management difficulties of complex process calculations GraceFUL Grace User Interface Feynman@Home World-wide Public distributed computing for Feynman diag. calculations HAPPY High Arithmetic Precision: beyond double-precision. New collaborators welcome Perfect topics for international cooperations