Broken-alpha: Genpdf snvar conditioning

[trivialTZ]

Extends GENPDF map conditioning so that extra dimensions can reference SN-generated variables (SALT2x1, SALT2c, RV, AV) in addition to HOSTLIB columns. This enables simulation of x1-dependent alpha distributions (broken-alpha) via GENPDF maps.

Changes:

• Add PTRVAL_SNVAR and USE_HOSTLIB fields to GENPDF struct
• Fall back to checkSNvar_HOSTLIB_WGTMAP() when HOSTLIB lookup fails
• Raise MXROW_GENPDF from 15000 to 50000 for dense conditioning maps
• Update funVal_genPDF and getRan_genPDF to read from SNVAR pointers

Build to achieve functionality for simulation similar to Ginolin et al. 2025 (ZTF DR2) broken-alpha finding.

Code changes

1. New struct fields (sntools_genPDF.h)

double *PTRVAL_SNVAR[MXVAR_GENPDF];  // pointer to live SN value (e.g., &GENLC.SALT2x1)
bool    USE_HOSTLIB[MXVAR_GENPDF];   // true=HOSTLIB lookup, false=SNVAR dereference

Each conditioning dimension of a GENPDF map now tracks which source to read from at runtime and, for SNVARs, stores a direct pointer to the simulator's internal value.

Also raised MXROW_GENPDF from 15000 to 50000 to support dense 2D maps (e.g., a fine alpha-vs-x1 grid can easily require 10k+ rows).

2. Safe initialization (two sites)

Both init_genPDF() (file-based maps) and init_genPDF_from_GenGauss() (Gaussian maps) now initialize the new fields:

for(ivar=0; ivar < MXVAR_GENPDF; ivar++) {
    GENPDF[NMAP].IVAR_HOSTLIB[ivar] = -9;
    GENPDF[NMAP].PTRVAL_SNVAR[ivar] = NULL;
    GENPDF[NMAP].USE_HOSTLIB[ivar]  = true;   // default: HOSTLIB
}

Defaults ensure backward compatibility: existing maps that only use HOSTLIB variables are unaffected.

3. Variable resolution with SNVAR fallback (init_genPDF)

The init-time loop that validates conditioning variables is changed from a single HOSTLIB-or-crash lookup to a three-step fallback:

Before:

for each conditioning variable:
    look up in HOSTLIB → not found? FATAL error

After:

for each conditioning variable:
    1. Try HOSTLIB lookup → found? store IVAR, set USE_HOSTLIB=true, done
    2. Try checkSNvar_HOSTLIB_WGTMAP() → recognized? store pointer, set USE_HOSTLIB=false, done
    3. Neither → FATAL error (with updated message mentioning both sources)

Step 2 reuses the existing upstream function checkSNvar_HOSTLIB_WGTMAP() (in sntools_host.c) which maps variable names to live pointers:

• "SALT2x1" / "x1" → GENLC.ptr_SHAPEPAR
• "SALT2c" / "c" → &GENLC.SALT2c
• "AV" → &GENLC.AV
• "RV" → &GENLC.RV

4. Runtime value retrieval (three sites)

Three functions that read conditioning values during simulation are updated with the same if/else pattern:

• funVal_genPDF() — evaluates P(x | conditions)
• getRan_genPDF() — accept/reject random draw, sets up conditioning values
• Debug dump in getRan_genPDF() — prints conditioning values when LDMP is on

Before (all three):

for(ivar=1; ivar < NDIM; ivar++) {
    IVAR_HOSTLIB = GENPDF[IMAP].IVAR_HOSTLIB[ivar];
    xval[ivar]   = get_VALUE_HOSTLIB(IVAR_HOSTLIB, IGAL);
}

After (all three):

for(ivar=1; ivar < NDIM; ivar++) {
    if (GENPDF[IMAP].USE_HOSTLIB[ivar]) {
        IVAR_HOSTLIB = GENPDF[IMAP].IVAR_HOSTLIB[ivar];
        xval[ivar]   = get_VALUE_HOSTLIB(IVAR_HOSTLIB, IGAL);
    } else {
        PTR_SNVAR = GENPDF[IMAP].PTRVAL_SNVAR[ivar];
        // NULL safety check with SEV_FATAL
        xval[ivar] = *PTR_SNVAR;
    }
}

Usage: simulating broken-alpha

Step 1: Write a GENPDF map file

The map defines P(SALT2ALPHA | SALT2x1). The first column is the generated parameter, middle columns are conditioning variables, and the last column is probability.

Minimal example (genpdf_alpha_vs_x1.DAT):

VARNAMES: SALT2ALPHA  SALT2x1  PROB
PDF: 0.11  -3.0  1.00
PDF: 0.19  -3.0  0.08
PDF: 0.11  -1.5  1.00
PDF: 0.19  -1.5  0.20
PDF: 0.11   0.0  1.00
PDF: 0.19   0.0  1.00
PDF: 0.11   1.5  0.20
PDF: 0.19   1.5  1.00
PDF: 0.11   3.0  0.08
PDF: 0.19   3.0  1.00

This encodes: at low x1 (x1=-3), alpha around 0.11 is strongly preferred; at high x1 (x1=+3), alpha~0.19 is preferred (broken-alpha population)

For precision studies (e.g., hard break at a specific x1), use a dense grid. Example with Ginolin+2024 truth values (alpha_low=0.280, alpha_high=0.088, break at x1=-0.49):

VARNAMES: SALT2ALPHA  SALT2x1  PROB
PDF: 0.088  -4.000  0.000
PDF: 0.091  -4.000  0.000
...
PDF: 0.280  -4.000  1.000
PDF: 0.088  -3.960  0.000
...

Dense maps require MXROW_GENPDF >= 50000 (included in this PR).

Step 2: Reference the map in snlc_sim input

GENMODEL:  SALT3.K21

GENPDF_FILE:    /path/to/genpdf_alpha_vs_x1.DAT
GENPDF_OPTMASK: 1

GENPEAK_SALT2x1:   0.0
GENSIGMA_SALT2x1:  1.0  1.0
GENRANGE_SALT2x1:  -3.0  3.0

Important: The x1 distribution (GENPEAK/GENSIGMA/GENRANGE_SALT2x1) must be defined so that x1 is generated first; the GENPDF map then draws alpha conditioned on the already-generated x1 value.

Step 3: Verify in the log

On successful init, snlc_sim prints:

Check GENPDF conditioning variables (HOSTLIB or SNVAR):
    Found SNVAR pointer for VARNAME='SALT2x1' (SALT2ALPHA)
    Found 0 HOSTLIB vars and 1 SNVARs in GENPDF map SALT2ALPHA

Pippin YAML example

SIM:
  BROKEN_ALPHA:
    IA:
      BASE: sim_broken_alpha.input   # references GENPDF_FILE above
    GLOBAL:
      NGEN_UNIT: 1
      BATCH_INFO: sbatch $SBATCH_TEMPLATES/SBATCH_LOCAL.TEMPLATE 1