Presentation on theme: "Evolving Architecture for Beyond the Standard Model"— Presentation transcript:
1 Evolving Architecture for Beyond the Standard Model KPS MeetingDCC, Daejeon, KoreaEvolving Architecture for Beyond the Standard Model~25Kihyeon CHO, Jangho KIM and Junghyun KIM(KISTI)
2 Contents Beyond Standard Model Simulation Computing Results Summary Simulation Tool Kit (MadGraph, Geant4)Evolving Computing Architecture
3 Higgs Discovery (July 4, 2012) Beyond Standard ModelHiggs Discovery (July 4, 2012)The SM is now complete?⇒ What is next?3
4 Beyond the Standard Model Beyond Standard ModelP5 Report ( )HiggsNeutrino MassAfter P5 Report ⇒DarkMatterDarkEnergyTheUnknownBefore P5 ReportEnergy FrontierIntensity FrontierCosmic FrontierSMBeyond the Standard Model
5 Beyond the Standard Model Beyond Standard ModelBeyond the Standard Model
6 SimulationHEP Simulation(physics)4We focus on MadGraph & Geant4.
7 Geant4 is the most successful model in HEP. SimulationGeant4 is the most successful model in HEP.HEP user community – BaBar(2001), LHC(2003), Belle IIOther community: Medical, Space, DNA physics, Solid
8 Vision for HEP Simulation To have a massively parallelized particle transportation engineTo comply with different architecture (GPU, MIC and etc.)To draw community interests for collateral effort
11 1. HTC module into MadGraph Results1. HTC module into MadGraphWe embed HTC module into MadGraph.⇒ Korean Economic News ( )Then, using it we study BSM.⇒ arXiv: [hep-ph]Beyond Standard Model(Z.’, Z’’)Simulation Tool Kit (MadGraph)Evolving Computing Architecture(HTC)
12 2. Finite Volume Effects on BK Results2. Finite Volume Effects on BKCP violation in Kaon SystemBeyond Standard Model (BK)Simulation (Finite Volume Effects)Evolving Computing Architecture(GPU)Calculated using Lattice QCDError Budget of BK⇒ To reduce the error of BK, we have to calculate the finite volume effect on the lattice.Reference: Kim, Jangho et al. Phys.Rev. D83 (2011) arXiv: [hep-lat]
13 GPU programming using CUDA ResultsGPU programming using CUDAAmount of calculation for FVX-axis : the number of norms.Y-axis : various quark masses.𝐾 1,0 ( 𝑛 𝑀𝐿), when the pion mass is small, the number of norms becomes large.Calculate Bessel function for every point in this plot and sum over them for same mass.CPU(single core) calculates point by point in serial order.It takes about two months to calculate FV correction in double precision for all the lattice samples we have.If we use GPU, it takes 1 day to calculate same thing.m_lowm_highParallel processing in GPU
14 3. NPR to calculate the matching factor of BK Results3. NPR to calculate the matching factor of BKBeyond Standard Model(BK)Simulation(NPR)Evolving Computing Architecture(GPU)One-LoopNPR (Non-perturbative Renormalization)BK in 𝑀𝑆 scheme on coarse lattice ( 20 3 ×64)Error Budget of BK
15 NPR to calculate the matching factor of BK (cont’d) ResultsNPR to calculate the matching factor of BK (cont’d)The calculation of the Green’s function for the one-color four-fermion takes 97% of total calculation time.H i;ABCD; c 1 c 2 c 3 c 4 𝛼 𝑝 1 + 𝜋 𝐴 , 𝑝 2 + 𝜋 𝐵 , 𝑝 3 + 𝜋 𝐶 , 𝑝 4 + 𝜋 𝐷 , 𝑘 =1 N conf 𝑖 𝑁 𝑐𝑜𝑛𝑓 2𝑎 4 𝑦 𝑒 −𝑖 𝑘 𝑦 𝐸𝐹𝐺𝐻 𝜓 i; AE;c 1 c 1 ′ 𝑝 1 , 𝑦 γ S 1 ⊗ ξ F EF 𝜙 i;FB; c 2 ′ c 2 y, p 2 U i;EH; c ′ 1 c ′ 4 y𝜓 i;CG; c 3 c 3 ′ p 3 ,𝑦 γ S 2 ⊗ ξ F GH 𝜙 i;HD; c 4 ′ c 4 y, p 4 U i;FG; c′ 2 c′ 3 (y)𝜓, 𝜙: quark propagator.𝑈 : gauge link.𝛼 : indicates operator in the Green’s function ( [𝑆 1 ⊗ 𝐹 1 ] [𝑆 2 ⊗ 𝐹 2 ]).𝑖 : gauge configuration index.𝑁 𝑐𝑜𝑛𝑓 : the number of gauge configurations.𝑎 : lattice spacing.A,B,C,D,E,F,G,H : hypercubic vectors:for example, A=(1,1,0,0). 16 cases are possible.c 1 , c 2 , c 3 , c 4 : color index (0 ∼ 2).z : lattice site whose spacing is 2𝑎.z A = 2z+A.Reference: Hwancheol Jeong, Jangho Kim et al. PoS(LATTICE2014)286 (2014) arXiv: [hep-lat]
16 GPU Performance NPR measurement code is optimized for GTX 480. Program ResultsGPU PerformanceProgramCPUGPUGPU vs. CPUCPU Spec.GFLOPSVGA(Peak Performancein double precision)Optimi-zationFinite Volume Effect of 𝐵 𝐾XeonE5-26200.5GTX 480(168 GFLOPS)64.338%128.6Non-perturbative Renormalization(NPR) measurementCorei7-4820K1.1366.640%58.9GTX 580(198 GFLOPS)76.1967.2GTX Titan Black(1707 GFLOPS)113.36100.3NPR measurement code is optimized for GTX 480.
17 Summary Physics goes beyond discovery. Computing needs solutions for the evolving architecture.⇒ To fulfill the gap between physics and computing, we need to focus on simulation R&D.
18 AcknowledgementDr. Soo-hyeon NamDr. Soon Yung JunProf. Weonjong Lee