Solver Options¶

EPM solves large-scale mixed-integer programs using CPLEX via GAMS. This page covers how to choose a solver type and tune the key parameters.

Solver types¶

Set via --modeltype on the CLI, or the MODELTYPE macro in GAMS:

Type	Description	When to use
`MIP`	Mixed-integer programming (discrete investment decisions)	Default for capacity expansion
`RMIP`	Relaxed MIP, integrality constraints dropped	Faster; good for bounds, debugging, or exploratory runs
`LP`	Linear programming, no integer variables	Only if the model has no binary/integer decisions

EPM ships ready-made option files in epm/input/data_test/cplex/. Point to one via the cplexfile row in config.csv:

File	Focus
`cplex_baseline.opt`	Balanced defaults, general-purpose
`cplex_rmip_fast.opt`	Fast relaxed solves (exploratory RMIP)
`cplex_rmip_precision.opt`	Accurate relaxed solves (reproducible RMIP)
`cplex_mip_fast.opt`	Fast integer solves (`optcr=0.05`, heuristic-heavy)
`cplex_mip_precision.opt`	High-precision integer solves (`optcr=0.01`, deterministic)
`cplex_test.opt`	Debug: verbose output, stable barrier settings

To run a solver comparison across scenarios:

python epm.py --folder_input data_test --config config.csv --scenarios scenarios_solver.csv

Focus on these four first, as they have the biggest impact:

Parameter	Recommended	Notes
`optcr`	`0.01`	Relative optimality gap. Tighten for production runs, loosen for speed. MIP only.
`threads`	`8–12`	More threads = faster, up to a point. Beyond ~16 threads, overhead increases.
`lpmethod`	`4` (auto)	Let CPLEX choose between simplex and barrier. Best default for power-system models.
`mipemphasis`	`0` (balanced)	`1` = find feasible solutions faster; `2` = prove optimality. Start balanced.

All parameters

Parameter	Recommended	Notes
`optcr`	`0.01`	MIP optimality gap
`mipemphasis`	`0`	MIP search strategy
`lpmethod`	`4`	LP algorithm
`threads`	`8–12`	Parallelism
`baralg`	`1`	Barrier + crossover
`barcrossalg`	`1`	Dual simplex crossover
`barcolnz`	`500`	Dense-column threshold for barrier
`barepcomp`	`1e-5`	Barrier complementarity tolerance
`parallelmode`	`1`	Deterministic (reproducible)
`startalg`	`4`	Root-start algorithm (auto)
`memoryemphasis`	`1`	Memory-saving mode
`predual`	`-1`	Primal/dual formulation (auto)
`scaind`	`1`	Automatic matrix scaling
`auxrootthreads`	`0`	Disable helper threads at root (avoids contention on large MIPs)
`heuristics`	`0.05`	Incumbent heuristics effort (MIP only)
`cuts`	`-1`	Cutting planes (auto)
`mipdisplay`	`4–5`	Branch-and-bound verbosity
`solutiontype`	`2` (RMIP), `1` (MIP)	Interior vs basic solution
`names`	`yes`	Keep symbol names for debugging
`iis`	`1`	Extract infeasibility report if model is infeasible
`bardisplay`	`2`	Barrier verbosity

Presolve hangs (log pauses after "CPLEX Presolve"): set auxrootthreads = 0
Infeasible model: enable iis = 1 to get a diagnostic report; see Debugging
Out of memory: enable memoryemphasis = 1; reduce problem size or move to a remote server