Module problem 

The unified optimization problem.

Problem is the single interface every forge optimizer consumes: a box-constrained, real-valued search space plus a scalar objective. By convention forge minimizes; a candidate whose objective is non-finite (NaN/±∞) is treated as infeasible and rejected, so models that blow up on degenerate parameters (a common case in hydrological calibration) need no special handling.

Two adapters cover the everyday cases without writing a struct:

• func wraps a closure with explicit bounds.
• Maximize flips the sense for objectives that should be maximized (e.g. NSE/KGE in rainflow), so the minimizing core stays the single convention.

Combinatorial problems (Anvil’s bit-flip simulated annealing) are served by a separate abstraction migrated in a later milestone; v0.1 is the continuous, population/real-vector substrate that unblocks rainflow.

Structs

Func

Closure-backed problem produced by func.

Maximize

Wraps a problem so the engine maximizes its objective instead of minimizing it, by negating finite values (non-finite stay rejected).

MultiFunc

Closure-backed multi-objective problem produced by multi_func.

Enums

BoundsError

Why a problem’s bounds are invalid.

Traits

MultiProblem

A box-constrained, real-valued multi-objective minimization problem.

Problem

A box-constrained, real-valued minimization problem.

Functions

func

A Problem defined inline by a closure and a bounds vector.

multi_func

A MultiProblem defined inline by a closure, a bounds vector, and an objective count.

validate

Validates that a problem is well-formed: at least one variable and every bound strictly ordered (lower < upper), rejecting NaN bounds.

validate_multi

Validates that a multi-objective problem is well-formed: valid bounds (as in validate) plus at least two objectives, and an objectives() vector whose length matches n_objectives() at the box midpoint.

Type Aliases

Bound

Inclusive search bounds for one decision variable: lower <= x <= upper.