FUSE meeting notes: Aug 8 2023

[orso82]

Orso

• Looked at time dependence
  • How to initialize current (eg. QED for N current relaxation time constants)
  • To start let's go from one stationary state to another stationary state
• Looked at running issues for v1.9.x on OMEGA
  • v1.8.5 works
  • export JULIA_REVISE_POLL=1 necessary when working on a Linux network filesystem
• We do not cost Li6 in our costing model
  • To start we Li6 should not be optimized --> use 90% like ARIES
  • INFUSE?
  • BpLi is 17% Li (by volume?)
  • Talked a lot in the blanked design group
  • US does not have any Li6 enrichment capability
  • Li6 used to be enriched in 50's and 60's at Y12
    • Mercury based process, very environmentally problematic --> illegal in the US
    • Military has a large Li6 inventory
  • Pure Li or FLiBe solves this issue
• Changed meshing of build layers

Kevin

• Amber can run Paramak build
• Progress on 2D deterministic solver
  • We can mesh (first wall issue that is too thin (2mm)) and 187 energy groups
  • Can run simulation
  • Visualize results in ParaView
  • Takes 2 minutes to run (single thread?)
    • Does this take into account speedup changes? probably yes, but check
  • Source is not realistic
  • Not very well converged
    • Convergence between successive runs
    • Residual is not really going down

Tim

• Fixed evolution of densities
• TAUENN with scaling laws is not necessarily robust
  • Can try using fixed profiles actor
  • Make it default

Adriana

• Discussion about costing with Leo (from Zabrina's group)
• Three priorities
  • Magnets --> LTS, HTS, learning rates
  • Focus on availability --> affects cost of electricity
  • BOP --> can we come up with a reduced costing model for BOP components
• We've eliminated some pri
  • Anything that scales with only electric power is not so important because we compare designs for the same electric power
• There is something about availability in FUSE

Weisberg

• Running optimizations
  • Fixed long runtime issues
  • Running with Solovev
  • TEQUILA is too slow and is not robust to over-driven cases
  • Should all solutions have positive current density on axis?
    • issue: psi is not a good independent variable
    • we're not interested: beta limits of current hole plasmas are very low, so not interested in building an FPP
    • we're not interested: on a realistic steady state scenario you'll never want to be overdriven
  • All Joey's STEP simulations have current drive on axis
    • We should have some on-axis current
    • This is true for simulation as well as reality
• Issue with GA algorithm running cases twice?
  • Need to investigate why

Tom

• Training of TGLF-NN on OMEGA with multi-threading and SLURM

JuliaHub

• export JULIA_REVISE_POLL=1
• Issues with Julia 1.9.2 on OMEGA
• debug where is issue