Physics tables beta

[ckoerber]

Result of todays table meeting

Below you can find the current form of the physics tables (which is going to change).

models.pdf

Please take a look, ask questions and suggest modifications.

Specifically entries with a question mark or a / need suggestions.

Proposed changes

Ensemble -> GaugeConfig

• The Ensemble table will become a GaugeConfig table which will carry explicit gauge config ids. Thus we will have more rows in each propagator table because each propagator now knows about the specific id.
• action column will now point to new SeaAction table which carries quark mass informations. For this reason the Flavor211 table will be dropped.

Action

• Change existing action to the new base tables
  • FermionAction
    • mass -> float
    • quark -> str
    • gauge_smear -> GaugeSmear
  • GlueAction
    • beta -> float
    • name -> str
    • gauge_smear -> GaugeSmear
  • SeaAction
    • fermion_actions -> List[FermionAction]
    • glue_action -> GlueAction
    • function which generates m_pi based on actions
• Create the following new specializations of FermionAction (columns to be determined)
  • Hisq?
  • MDWF?
  • Clover?

Propagator

• rename ensemble to gauge_config and thus have an gauge_config.id
• mval stays in propagator and not in action
• action -> FermionAction
• Decide to specialize propagators instead of introducing an explicit Source table (which propagators point to). This has the reason that it basically just shifts work from Propagators to Sources while some sources will not exist as files and also querying becomes more difficult.

Current

• Add momentum_{x, y, z} -> int

Hadron -> Interpolator?

• This Table is still an open question: What is your most general definition of a lattice Hadron? What are conserved quantum numbers?
• Rename Hadron table to Interpolator
• Add n_up -> int
• Add n_down -> int
• Add n_strange -> int
• Add n_charm -> int
• Add n_bottom -> int
• Add n_top -> int
• Add momentum_{x, y, z} -> int
• Remove spin and spin_z and parity because they are not good quantum numbers
• Remove isospin and isospin_z because they are redundant (since n_{up, down})
• Add irrep -> str?
• Add irrep_row -> int?
• Specialize Interpolator later on into, e.g., Nucleon or others.
• Remove Basic table

Non-physics tables

According to https://github.com/callat-qcd/lattedb/issues/18#issuecomment-516631851

Job/Task/Data/Status

• physics -> PhysicsTable (like Correlator)
• machine -> Machine
• computation -> Computation
• status -> int

Machine

• topology -> str (distribution of lattice among processors)
• n_cpu -> int
• n_gpu -> int
• home -> str
• scratch -> str
• tape -> str

Computation?

• name -> str (descriptive info)
• details -> son/blob/file (infos to uniquely define the computation)

The reason for keeping Computation simple is that it seems hard to generalize. Also, it seems like that you do not specifically query for specific information -- you might want to backtrace previous runs though. The disadvantage is that the actual jobs must be created externally. E.g., given machine and physics object, generate me the actual job to submit, submit this job, push this job to the DB with status submitted.

[cchang5]

changelog

Action table changes: Split Action -> FermionAction and GaugeAction

Moved quark masses into FermionAction. Tag quark name. In particular, allows same mass to be called different names (light, up, down for example) Quark mass dropped from propagator and gaugeconfig

Moved beta into GaugeAction

Ensemble table changes: Renamed Ensemble -> GaugeConfig Changed Nconfig -> Config to specify specific config number GaugeConfig has children (e.g. Nf211) which FKeys to gaugeaction and fermionactions

Propagator table changes: Propagator now points to gaugeconfig and fermionaction Children of propagators specialize different RHS. (Decided to not make a source table and just specialize in Propagator).

Current table changes: removed flavor from Local(Current). The propagators explicitly contain quark mass info now. moved momentum into current to denote momentum transfer. created Local4D to spatial current (non FT’d) removed momentum from FHPropagator

Hadron table changes:
rename to Interpolator moved momentum into children tables. remove momentum from correlation functions (All momentum now defined at interpolating operator and current if in momentum space)

HadronSmear table changes: rename to interpolatorsmear

Todo: Need to think more about Interpolator table. Need to work on non-physics tables.

[ckoerber]

We settled for the following physics table layout with commit 0b94784e18309080f1ed40374a9aaca79df9d2a3 and following.

models.pdf

Note that the gauge config table now has a config column so that we will have many gaugeconfigs for a given ensemble (thus all tables depending on gaugeconfig will have many entries as well). The ensemble table groups gaugeconfigs.