SCIP solver control parameters

Create a list of control parameters for the SCIP solver. Parameters are organized into logical groups: output, limits, tolerances, presolving, LP, branching, heuristics, separating, and emphasis. Any SCIP parameter can also be set directly using its native path via ....

Usage

scip_control(
  verbose = TRUE,
  verbosity_level = 3L,
  display_freq = 100L,
  time_limit = Inf,
  node_limit = -1L,
  stall_node_limit = -1L,
  sol_limit = -1L,
  best_sol_limit = -1L,
  mem_limit = Inf,
  restart_limit = -1L,
  gap_limit = 0,
  abs_gap_limit = 0,
  feastol = 1e-06,
  dualfeastol = 1e-07,
  epsilon = 1e-09,
  presolving = TRUE,
  presolve_rounds = -1L,
  presolve_emphasis = "default",
  lp_threads = 1L,
  lp_iteration_limit = -1L,
  lp_scaling = TRUE,
  branching_score = "p",
  heuristics_emphasis = "default",
  separating_emphasis = "default",
  emphasis = "default",
  threads = 1L,
  ...
)

Arguments

verbose: Logical; print solver output. Default TRUE.
verbosity_level: Integer 0–5; verbosity detail. Default 3.
display_freq: Integer; node display frequency. Default 100.
time_limit: Numeric; time limit in seconds. Default Inf.
node_limit: Integer; max nodes. Default -1L.
stall_node_limit: Integer; stall nodes. Default -1L.
sol_limit: Integer; solution limit. Default -1L.
best_sol_limit: Integer; improving solution limit. Default -1L.
mem_limit: Numeric; memory limit in MB. Default Inf.
restart_limit: Integer; restart limit. Default -1L.
gap_limit: Numeric; relative MIP gap. Default 0.
abs_gap_limit: Numeric; absolute gap. Default 0.
feastol: Numeric; feasibility tolerance. Default 1e-6.
dualfeastol: Numeric; dual feasibility tolerance. Default 1e-7.
epsilon: Numeric; zero tolerance. Default 1e-9.
presolving: Logical; enable presolving. Default TRUE.
presolve_rounds: Integer; presolve rounds. Default -1L.
presolve_emphasis: Character; presolving emphasis. Default "default".
lp_threads: Integer; LP solver threads. Default 1L.
lp_iteration_limit: Integer; LP iteration limit. Default -1L.
lp_scaling: Logical; LP scaling. Default TRUE.
branching_score: Character; score function. Default "p".
heuristics_emphasis: Character; heuristic emphasis. Default "default".
separating_emphasis: Character; separating emphasis. Default "default".
emphasis: Character; global solver emphasis. Default "default". See the Emphasis section.
threads: Integer; parallel solving threads. Default 1L. See Parallel section for caveats.
...: Additional SCIP parameters as name-value pairs, using SCIP's native hierarchical parameter paths (e.g., "lp/fastmip" = 1, "conflict/enable" = FALSE). See the SCIP parameter documentation for the full list.

Value

A named list of class "scip_control" with components:

verbose: Logical; verbosity flag.
scip_params: Named list; all parameters as SCIP native paths.

Details

The four emphasis arguments (emphasis, presolve_emphasis, heuristics_emphasis, separating_emphasis) are SCIP meta-settings: each one adjusts a whole family of native parameters at once, the same way SCIP's interactive shell commands set emphasis, set presolving emphasis, set heuristics emphasis and set separating emphasis do. They are applied before any individual parameter, in that order, so a native parameter given via ... always overrides whatever the emphasis chose for it.

Output

verbose: Logical; print solver output. Default TRUE.
verbosity_level: Integer 0–5; verbosity detail (0 = none, 3 = normal, 5 = full). Default 3. Ignored if verbose = FALSE.
display_freq: Integer; display a status line every this many nodes (-1 = never). Default 100.

Termination Limits

time_limit: Numeric; maximum solving time in seconds. Default Inf (no limit).
node_limit: Integer; maximum number of branch-and-bound nodes. Default -1L (no limit).
stall_node_limit: Integer; nodes without improvement before stopping. Default -1L (no limit).
sol_limit: Integer; stop after finding this many feasible solutions. Default -1L (no limit).
best_sol_limit: Integer; stop after this many improving solutions. Default -1L (no limit).
mem_limit: Numeric; memory limit in MB. Default Inf (no limit).
restart_limit: Integer; maximum restarts. Default -1L (no limit).

Tolerances

gap_limit: Numeric; relative MIP gap tolerance; solver stops when the gap falls below this value. Default 0 (prove optimality).
abs_gap_limit: Numeric; absolute gap between primal and dual bound. Default 0.
feastol: Numeric; feasibility tolerance for LP constraints. Default 1e-6.
dualfeastol: Numeric; dual feasibility tolerance. Default 1e-7.
epsilon: Numeric; absolute values below this are treated as zero. Default 1e-9.

Presolving

presolving: Logical; enable presolving. Default TRUE.
presolve_rounds: Integer; maximum presolving rounds (-1 = unlimited). Default -1L.
presolve_emphasis: Character; presolving emphasis setting: "default", "aggressive", "fast", or "off". Default "default".

LP

lp_threads: Integer; number of threads for LP solver. Default 1L.
lp_iteration_limit: Integer; LP iteration limit per solve (-1 = no limit). Default -1L.
lp_scaling: Logical; enable LP scaling. Default TRUE.

Branching

branching_score: Character; branching score function: "s" (sum), "p" (product), "q" (quotient). Default "p".

Heuristics

heuristics_emphasis: Character; heuristic emphasis setting: "default", "aggressive", "fast", or "off". Default "default".

Separating

separating_emphasis: Character; cutting-plane separation emphasis setting: "default", "aggressive", "fast", or "off". Default "default".

Emphasis

emphasis: Character; global solver emphasis, choosing a predefined parameter profile for a problem class or solving goal. One of "default", "cpsolver" (constraint-programming style search, no LP relaxation), "easycip" (solve easy problems fast), "feasibility" (find a feasible solution fast), "hardlp" (cope with hard LP relaxations), "optimality" (prove optimality fast), "counter" (solution counting), "phasefeas", "phaseimprove", "phaseproof" (the three phases of SCIP's three-phase approach), "numerics" (increased numerical safety), or "benchmark" (do not try to avoid the memory limit). Default "default".

Parallel

threads: Integer; number of threads for concurrent solving. Default 1L. Note: concurrent solving may require a SCIP build compiled with parallel support (e.g., PARASCIP=true); not all installations provide this.

Examples

## Quick solve with 60-second time limit
ctrl <- scip_control(time_limit = 60)

## Quiet solve with 1% gap tolerance
ctrl <- scip_control(verbose = FALSE, gap_limit = 0.01)

## Aggressive heuristics, no presolving
ctrl <- scip_control(heuristics_emphasis = "aggressive", presolving = FALSE)

## Aggressive presolving and separation, as for hard combinatorial MIPs
ctrl <- scip_control(presolve_emphasis = "aggressive",
                     separating_emphasis = "aggressive")

## Global profile: find a feasible solution quickly
ctrl <- scip_control(emphasis = "feasibility")

## Pass a native SCIP parameter directly
ctrl <- scip_control("conflict/enable" = FALSE, "separating/maxrounds" = 5L)