Struct NlProblem 

pub struct NlProblem {

    pub n: usize,
    pub m: usize,
    pub num_obj: usize,
    pub minimize: bool,
    pub obj_nonlinear: Expr,
    pub obj_linear: Vec<(usize, f64)>,
    pub obj_constant: f64,
    pub con_nonlinear: Vec<Expr>,
    pub con_linear: Vec<Vec<(usize, f64)>>,
    pub x_l: Vec<f64>,
    pub x_u: Vec<f64>,
    pub g_l: Vec<f64>,
    pub g_u: Vec<f64>,
    pub x0: Vec<f64>,
    pub lambda0: Vec<f64>,
    pub suffixes: NlSuffixes,
    pub ampl_options: Vec<i64>,
    pub imported_funcs: Vec<ImportedFunc>,
    pub var_names: Vec<String>,
    pub con_names: Vec<String>,
}

Parsed .nl problem in the form needed by NlTnlp.

Fields

n: usize

m: usize

num_obj: usize

minimize: bool

obj_nonlinear: Expr

obj_linear: Vec<(usize, f64)>

obj_constant: f64

con_nonlinear: Vec<Expr>

Per-constraint nonlinear part (length m).

con_linear: Vec<Vec<(usize, f64)>>

Per-constraint linear part (length m), each a list of (var, coef).

x_l: Vec<f64>

x_u: Vec<f64>

g_l: Vec<f64>

g_u: Vec<f64>

x0: Vec<f64>

lambda0: Vec<f64>

suffixes: NlSuffixes

AMPL suffix dictionaries. Variable / constraint / objective suffixes are stored as dense vectors (length n / m / num_obj) with the sparse .nl S-segment entries scattered in, default zero. The integer / real split matches the S-segment header’s kind bit (0x4 ⇒ real, else integer). See https://ampl.com/REFS/hooking2.pdf §6 and the upstream .nl reader in ref/Ipopt/src/Apps/AmplSolver/AmplTNLP.cpp.

ampl_options: Vec<i64>

The model’s own AMPL option words, taken verbatim from .nl header line 0 (g<count> <opt0> <opt1> ...). A solver echoes these back in the .sol Options block rather than interpreting them — see crate::sol_writer::format_sol_with_options. Empty for problems not built from a .nl file.

imported_funcs: Vec<ImportedFunc>

AMPL imported (external) functions declared via top-level F segments. Empty unless the .nl file calls compiled-C user functions (typically emitted by IDAES property packages — see issue #49).

var_names: Vec<String>

Variable names from the sibling .col file, index-aligned to x (one name per line, column order). Empty when no .col file was found — AMPL only emits it under option auxfiles rc;.

Carrying names lets diagnostics report flow_balance / T_reactor instead of c[3] / x[132]. Lee et al. (2024) identify the gap between detecting an issue and tracing it to a named equation as a central roadblock for equation-oriented model debugging; threading names through to the solver/debugger is the prerequisite for closing it. See https://doi.org/10.69997/sct.147875.

con_names: Vec<String>

Constraint names from the sibling .row file, index-aligned to g (one name per line, row order). Empty when no .row file was found. See NlProblem::var_names for why names are captured.

Implementations

impl NlProblem

pub fn from_expressions(parts: NlProblemParts) -> Result<NlProblem, String>

Assemble a problem from expression trees, with no .nl file anywhere in the loop (issue #469).

A modeling frontend that already has its own expression DAG should come in here rather than serialize to .nl and re-parse: the round trip is not only slower, it is lossy, because .nl writers routinely refuse operators this tape supports natively (atan2, min/max, and — with no .nl opcode at all — UnaryOp::Erf).

The result is an ordinary NlProblem, so it feeds NlTnlp::try_new and gets exactly the evaluators a parsed model does: objective, gradient, constraints, Jacobian + structure, Lagrangian Hessian + structure, and NlTnlp::hessian_vector_product.

Errors on a length mismatch or on a Var(i) index at or beyond n — the latter would otherwise be an out-of-bounds read in the tape’s forward sweep, so it must be caught while it is still a diagnosable user error.

Trait Implementations

impl Clone for NlProblem

fn clone(&self) -> NlProblem

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for NlProblem

fn fmt(&self, f: &mut Formatter<'_>) -> Result<(), Error>

Formats the value using the given formatter.