TGLF-NN model in TRANSP

The TGLF-NN surrogates have been developed for the FUSE project.

Namelist options for the TGLF Neural Network Option

To use TGLFNN models, set TGLFMOD = 4 in the namelist file.

`NNENSEMBLE`

Description: Specifies the file containing the ensemble of neural network models to be used. Each ensemble contains 20 models.

Default: "sat3_em_nstx_azf-1" — recommended for NSTX H-mode and L-mode discharges.

Alternate settings (recommended use):

sat0quench_em_d3d_azf+1 — DIII-D H-mode discharges.
sat0quench_em_mastu_azf+1 — MAST-U H-mode discharges.
sat1_em_d3d_azf-1 — DIII-D L-mode discharges.
sat3_em_mastu_azf-1 — MAST-U L-mode discharges.
sat1_em_d3d_negD — DIII-D negative-D discharges.

Also available:

sat3_em_d3d_azf-1
sat3_em_d3d_azf+1
sat3_em_d3d+mastu+nstx_azf-1
sat3_es_d3d_azf-1
sat3_es_d3d_azf+1

`NNAVG`

Description: Number of models within the ensemble to evaluate and average for each calculation.

Default: 1

Special: Set to 0 to use all available models in the selected ensemble.

`NNMODIDX`

Description: Index of the starting model within the ensemble (1 to total number of models).

Default: 1

If NNAVG == 1, only this model is evaluated.
If NNAVG == 0, all models are evaluated and the starting index is irrelevant.

`NNBDCHK`

Description: Logical flag to check whether each input variable to the model lies within the model’s training bounds; prints a warning for each that does not.

Default: .TRUE.

Typically only a few variables in a few zones fall slightly outside the training bounds, but this can still produce many warnings; set NNBDCHK = .FALSE. to suppress them if desired.

Sample TGLFNN namelist

Sample TGLFNN namelist (tglf_control)

&tglf_control
 NLGRAD = .FALSE.          ! output flux gradients
 ADIABATIC_ELEC = .FALSE.  ! adiabatic electrons selection
 USE_BPER = .FALSE.        ! perpendicular B effects
 USE_BPAR = .FALSE.        ! parallel B effects
 USE_MHD_RULE = .TRUE.     ! MHD rules
 USE_BISECTION = .TRUE.    ! use bisection
 USE_INBOARD_DETRAPPED = .FALSE. ! inboard detrapped
 USE_AVE_ION_GRID = .FALSE.      ! if true compute charge density weighted average of the kinetic ion gyro-radius
 LFLUX = .TRUE.            ! IFLUX
 FIND_WIDTH = .TRUE.       ! find width

 TGLFMOD = 4           ! 1 for xptor version, 2 for tgyro version, 4 for NN version
 NNENSEMBLE = "sat3_em_d3d_azf-1"
 NNMODIDX = 1
 NNAVG = 0
 NNBDCHK = .TRUE.
 IBRANCH = -1          ! model branch
 GEOM_FLAG = 1         ! geometry flag
 NMODES = 2            ! tglf modes
 NSPEC = 3             ! tglf number of species
 NBASIS_MAX = 4        ! maximum base function used in tglf
 NBASIS_MIN = 2        ! minimum base function used in tglf
 NXGRID = 16           ! nx grid
 NWIDTH = 21           ! number of gaussian width
 SAT_RULE = 0          ! saturation rule in tglf
 KYGRID_MODEL = 1      ! model for kygrid
 XNU_MODEL = 2         ! model selection for collisionality
 VPAR_MODEL = 0        ! parallel velocity model
 VPAR_SHEAR_MODEL = 0  ! parallel velocity shear model
 KINETIC_FAST_ION = 0  ! fast ion species model in TGLF
 PEVOLVING = 0         ! evolving temperature and its gradients

 KY = 0.3            ! ky
 THETA_TRAPPED = 0.7 ! trapped theta
 WDIA_TRAPPED = 0.0  ! wdia trapped (bounce model)
 WIDTH = 1.65        ! gaussian width
 WIDTH_MIN = 0.3     ! minimum gaussian width
 WD_ZERO = 0.1       ! zero width
 PARK = 1.0          ! PARK in tglf
 GHAT = 1.0          ! ghat in tglf
 GCHAT = 1.0         ! gchat in tglf
 LINSKER_FAC = 0.0   ! liksker factor in tglf
 GRADB_FAC = 0.0     ! factor for gradB in tglf
 FILTER = 2.0        ! filter in tglf
 DAMP_PSI = 0.0      ! damping factor for psi
 DAMP_SIG = 0.0      ! damping factor for sig
 ALPHA_P = 1.0       ! parallel velocity shear factor
 ALPHA_ZF = 1.0      ! factor for SAT_RULE 1 zonal flow
 ALPHA_MACH = 0.0    ! factor for mach
 ALPHA_N = 0.0       ! factor for vn shear
 ALPHA_T = 0.0       ! factor for vt shear
 ALPHA_QUENCH = 0.0  ! quench model factor
 XNU_FAC = 1.0       ! factor for collisionality
 DEBYE_FAC = 1.0     ! factor for debye length
 ETG_FAC = 1.25      ! factor for the ETG model
 CBETAE = 1.0        ! factor for plasma beta
 CXNU = 1.0          ! scale factor for collisionality
 EM_STAB = 1.0       ! EM factor for the ion temperature gradient
 RLNP_CUTOFF = 18.0  ! Limits SAT2 factor of R/Lp < RLNP_CUTOFF
/