eq_rsc_INVlim infeasibility for geohydro_allkm when exogenous capacity exceeds m_rsc_dat * geo_discovery

[Yunzhi-Chen]

Summary

Infeasibility in eq_rsc_INVlim for geohydro_allkm for Western interconnect at county-level resolution. At p35023 in 2015, the right-hand side of eq_rsc_INVlim for geohydro_allkm_9 bin1 is negative by ~0.447 MW, so the model has no feasible value for INV_RSC (which is constrained to be ≥ 0).

Branch / commit: yc_pbitl (in ReEDS-2.0)

Case / scenario: cases_pbitl.csv → pbitl1_Western_Default_Stress

Solve year where it fails: 2015

Region / tech: county p35023, geohydro_allkm

Steps to reproduce:

1. Check out branch yc_pbitl (which I think is consistent with ReEDS main on this constraint)
2. Run county-level Western Interconnect
3. Model is infeasible at the 2015 solve

---

Observation

eq_rsc_INVlim enforces:

m_rsc_dat * geo_discovery  >=  sum(INV_RSC) + capacity_exog_rsc

geo_discovery(i,r,t) is the fraction of the geothermal supply curve treated as "discovered" in year t (e.g., 0.10–0.15 in early years, monotonically increasing toward 1).

In p35023 in 2015, the left-hand side m_rsc_dat * geo_discovery evaluates to slightly less than the right-hand-side terms, producing a negative RHS of ~0.447 MW and making the constraint infeasible together with the prescribed builds requirement (eq_forceprescription_power).

---

Solutions considered

A. Inflate m_rsc_dat so RHS becomes positive — e.g., extend the year-aware capacity top-up logic from commit 77fb3a6 in b_inputs.gms to add enough geothermal m_rsc_dat to bin1 so prescribed builds always fit, after accounting for geo_discovery.

• Pros: confirmed to resolve the p35023 infeasibility; works within the same framework used for wind and solar; no equation changes needed
• Cons: the "extra" MW added to bin1 are made-up numbers that show up later in capacity reports; piling everything into bin1 can mislead the optimizer's cost ranking across bins; treats the symptom rather than addressing why the constraint produced negative RHS in the first place

B. Disable eq_rsc_INVlim for offending bins via flag_eq_rsc_INVlim

• Pros: targeted; preserves data
• Cons: bin loses physical upper bound; may concentrate prescribed builds in one bin; conceptually a patch

C. Set geo_discovery = 1 wherever a bin has negative RHS

• Pros: trivially feasible
• Cons: opens all bins in the region (geo_discovery is region-level, not bin-level); requires updating rsc_capacity_scalar to stay consistent

D. Reformulate eq_rsc_INVlim so geo_discovery is applied symmetrically to both sides of the supply-curve accounting, by also scaling the capacity_exog_rsc term by rsc_capacity_scalar. Equivalent form:

   (m_rsc_dat - exog) * geo_discovery  >=  sum(INV_RSC)

• Pros: mathematically clean; no parameter fabrication; no special-case logic; backward-compatible (non-geothermal techs have rsc_capacity_scalar = 1, so unchanged); auto-converges as geo_discovery → 1
• Cons: touches the model equation itself, so future maintainers must understand the new convention; the conceptual interpretation of geo_discovery relative to existing capacity should be confirmed with the modeling team

---

Suggested Solution: D

Two synchronized changes:

1. c_supplymodel.gms — eq_rsc_INVlim: multiply the capacity_exog_rsc term by rsc_capacity_scalar.
2. d_solveoneyear.gms — rhs_eq_rsc_INVlim pre-solve calculation: apply the same geo_discovery factor to the exogenous term to match the new equation.

---

Validation

• Re-run pbitl1_Western_Default_Stress through 2015 — confirm infeasibility is resolved
• Spot-check INV_RSC.l for geohydro_allkm in p35023 across solve years to verify reasonable bin allocation
• Run a baseline (non-pbitl) Standard Scenarios case to confirm no regression for non-geothermal technologies
• Verify the geothermal branch of the 77fb3a6 pre-processing no longer adds capacity once D is in effect

---

Labels: bug