1 Feasibility and the notion of a feasible set

Feasibility is the property of a problem to have at least one assignment of variables that satisfies all imposed constraints. In many modeling tasks—such as optimization, systems of inequalities, and constraint satisfaction—determining feasibility is a prerequisite to further analysis, including finding an optimum or proving that a model is consistent with assumptions.

A feasible set is the collection of all variable values that meet the constraints. When constraints depend on parameters, the feasible set changes as those parameters vary, and the “feasibility range” describes where the set remains nonempty.

1.1 Definitions: feasible, infeasible, and boundary cases

A point (or assignment) is feasible if it satisfies every constraint simultaneously. A problem is infeasible if no such point exists, meaning the feasible set is empty.

Boundary cases occur when feasible points lie at the “edge” of satisfaction—such as when an inequality holds exactly with equality, or when an equality constraint forces variables to a lower-dimensional surface. These boundaries often govern transitions between feasibility and infeasibility as parameters shift.

1.2 Constraints as mathematical objects

Constraints define permissible variable values through mathematical conditions. They may be expressed as equations, inequalities, or more general relations.

1.2.1 Equality constraints

An equality constraint requires a function to match a specified value. In its simplest form, it can be written as \[ g(x)=0, \] or more generally \(g(x)=b\). Equality constraints typically reduce the degrees of freedom, potentially yielding feasible sets of lower dimension than the surrounding space.

1.2.2 Inequality constraints

An inequality constraint limits a function to be greater than, less than, or equal to a threshold: \[ h(x)\le 0,\quad h(x)\ge 0,\quad \text{or strict variants } h(x)<0,\; h(x)>0. \] Inequality constraints usually carve out regions (including open or closed regions depending on strictness). The inclusion or exclusion of boundary points strongly affects the resulting feasible set and, consequently, the feasibility range.

1.3 Feasibility in optimization versus feasibility-only problems

In optimization, feasibility ensures that at least one candidate solution exists for the objective to act on. Yet the objective can be irrelevant to feasibility: one may seek feasibility without minimizing or maximizing anything.

A feasibility-only problem asks only whether the feasible set is nonempty and, in parameterized versions, for which parameter values this remains true. Feasibility range analysis is central in such formulations, since it summarizes consistency across parameter variations without committing to a particular optimum.

2 Feasibility range for parameterized models

When constraints depend on parameters, feasibility becomes a function of those parameters. The feasibility range is the set (often an interval or region) of parameter values for which the parameterized model admits at least one feasible solution.

The range can be continuous (common with real-valued parameters), multi-dimensional (for several parameters), or a collection of components (when constraints produce disjoint feasible sets). In discrete settings, the “range” becomes a set of allowable parameter values rather than a continuous interval.

2.1 Parameter dependence of feasibility

Feasibility can change gradually or abruptly as parameters vary, depending on how constraints interact with the parameter.

2.1.1 Single-parameter feasibility intervals

With one real parameter \(p\), feasibility is often described by an interval or union of intervals: \[ p\in [p_{\min},p_{\max}],\quad \text{or}\quad p\in (p_{\min},p_{\max}), \] or even a disjoint union \(\bigcup_i I_i\). The interval endpoints are frequently determined by constraints becoming tight (inequalities holding at equality) or by equality constraints forcing incompatibilities.

2.1.1.1 Closed versus open intervals (strict versus non-strict inequalities)

Non-strict inequalities typically produce closed feasibility sets (including boundary points), while strict inequalities tend to yield open sets (excluding boundary points). For example, a condition \(h(x,p)\le 0\) may allow parameter values where \(h(x,p)=0\), but a condition \(h(x,p)<0\) excludes those parameter values even if a limiting feasible sequence exists.

This distinction matters in computed feasibility ranges, especially when numerical methods approximate boundaries with finite tolerance.

2.1.2 Multi-parameter feasibility regions

With two or more parameters, the feasibility range is a region in a higher-dimensional space. It can be convex or non-convex and may include holes or disjoint components. Geometric and algebraic structure of the constraints often determines the shape of this region.

For example, constraints that are linear in both variables and parameters can create polyhedral regions in parameter space. Nonlinear dependence can generate curved boundaries and complicated feasibility landscapes.

2.1.3 Discrete parameters and combinatorial feasibility

If a parameter takes values from a discrete set (integers, categories, or enumerations), feasibility range descriptions become combinatorial. Instead of intervals, one reports the subset of allowable parameter values for which feasible assignments exist. In such cases, the structure may resemble a truth table, a sparse set of options, or a grid that must be searched systematically.

3 Computing the feasibility range

Computing feasibility ranges translates the theoretical question “for which parameters is the feasible set nonempty?” into algorithms or analytical derivations. Methods vary based on whether constraints are linear or nonlinear, low- or high-dimensional, and whether exact or approximate answers are acceptable.

3.1 Analytical approaches

Analytical methods seek formulas or exact expressions for feasible parameter sets.

3.1.1 Inequality solving and interval arithmetic

For single-parameter problems, one may reduce constraints to inequalities in \(p\) and solve them directly. When expressions are complicated, interval arithmetic can propagate bounds to determine whether constraints can be satisfied within a parameter interval. This technique supports rigorous enclosure of possible feasibility, though it may become conservative.

3.1.2 Geometry-based reasoning for low-dimensional cases

In low-dimensional spaces, geometry can simplify reasoning. Intersections of half-spaces, discs, or other shapes can be analyzed to determine when an intersection becomes nonempty. For parameter sweeps, one can track how feasible regions appear or disappear as parameters move, often by identifying tangency or separation events.

3.2 Algorithmic approaches

Algorithmic approaches are used when analytical solutions are difficult or impossible.

3.2.1 Feasibility checks with constraint solvers

A common strategy is to repeatedly test feasibility using a constraint solver (e.g., for linear, nonlinear, or general constraint systems). For parameterized problems, the solver becomes an oracle: given a parameter value, it determines whether at least one solution exists. The feasibility range is then assembled by exploring parameter values systematically.

3.2.2 Search methods (bisection, scanning, continuation)

If feasibility is monotone in a parameter (or approximately so), bisection can locate boundary points efficiently by narrowing intervals where feasibility changes. Scanning evaluates feasibility on a grid and refines promising regions.

Continuation methods gradually vary parameters, using solutions at one parameter setting as initial guesses at nearby settings. This can be effective when feasible solutions persist smoothly, though it may fail near bifurcations or discontinuities.

3.2.3 Optimization-based reformulations

Sometimes feasibility range computation is reframed as optimization. For instance, one can introduce a measure of constraint violation and minimize it, treating feasibility as the event that the minimum equals zero. Alternatively, one may solve a related problem that identifies whether constraints can be satisfied simultaneously.

Reformulations must be handled carefully: a small violation in an approximate minimization might not correspond to true feasibility, especially under strict inequalities.

3.3 Numerical stability and tolerance handling

Real-world computation relies on floating-point arithmetic, so boundary detection must account for tolerances. Solvers typically treat constraints with a “soft” interpretation (e.g., \(h(x)\le \epsilon\) for small \(\epsilon\)) and report feasibility status accordingly.

A robust feasibility range computation therefore includes:

  • consistent tolerance selection,
  • verification that boundary points are correctly classified (within error bounds),
  • and, when needed, refinement using tighter tolerances or interval-based methods.

4 Characterizing the boundaries of the feasible set

The boundaries of the feasible set are where feasibility is most fragile. They are often the locations where small parameter changes can convert a feasible problem into an infeasible one.

4.1 Active constraints and constraint qualification intuition

A constraint is active at a feasible point if it holds exactly at the limiting value (e.g., \(h(x,p)=0\) for \(h(x,p)\le 0\)). Boundary points often have one or more active constraints. Intuitively, the structure of active constraints determines whether small perturbations retain a solution or destroy it.

While formal results require technical conditions, the key idea is that boundary behavior is governed by constraints that are “tight” rather than those with slack.

4.2 Detecting feasibility transitions

Feasibility transitions occur when the feasible set changes from empty to nonempty or vice versa as parameters vary. In many cases, transitions are associated with:

  • tangency between constraint surfaces,
  • collapse of a feasible region to a lower-dimensional set,
  • emergence or disappearance of intersections among constraint-defined sets.

Detecting these transitions often involves monitoring the minimum slack of inequalities or tracing the parameter values where a solver reports the smallest margin to feasibility.

4.3 Handling degeneracy and multiple solutions

Boundaries can be complicated by degeneracy (multiple constraints becoming simultaneously tight) and by the presence of multiple feasible solutions.

4.3.1 Uniqueness versus non-uniqueness near boundaries

If the feasible solution set near a boundary is a single point, transitions may be sharp and easier to localize. If many solutions exist (e.g., a continuum of feasible points), then the boundary may correspond to losing an entire region rather than a single point. In such situations, numerical procedures may show reduced sensitivity, making classification harder.

4.3.2 Flat regions and measure-zero feasible sets

Some systems permit “flat” feasible regions where constraints do not strongly constrain movement. Others may allow feasibility only on a set of parameter values of measure zero (for example, exact equalities in parameter-dependent constraints). In computational practice, such cases can be hard to capture, since random sampling may never land exactly on the narrow feasible set. Interval methods or symbolic reasoning can mitigate this issue when feasible.

5 Relationship to robustness and sensitivity

Feasibility range analysis provides insight into robustness: it shows how much parameter variation can be tolerated before constraints become unsatisfiable.

5.1 Robust feasibility under parameter perturbations

A model is robustly feasible if a neighborhood of parameter values around a nominal choice remains feasible. Here, the feasibility range directly quantifies robustness margin: larger feasible intervals or regions correspond to greater tolerance to perturbations.

Robustness is not absolute; it depends on how parameters are expected to vary and on whether strict or non-strict constraints are used.

5.2 Sensitivity of solutions versus sensitivity of feasibility

Solution sensitivity and feasibility sensitivity are related but distinct. A system may have feasible solutions that vary significantly with parameters (solution sensitivity), while feasibility itself might persist over a wide range. Conversely, feasibility may be highly sensitive even if the best feasible solutions change little.

Feasibility range focuses on the on/off property (nonemptiness) rather than the behavior of particular solutions. This can be advantageous when the objective is irrelevant or when solution quality is secondary to constraint satisfaction.

5.3 Worst-case versus average-case feasibility interpretation

In some applications, one interprets feasibility under:

  • worst-case perturbations: ensuring constraints hold for all parameter values in an uncertainty set, which aligns with conservative feasibility regions,
  • average-case perturbations: assessing typical feasibility probability under a distribution on parameters, which depends on how the feasibility set intersects the distribution’s support.

While both relate to feasibility range, they answer different questions about risk and reliability.

6 Feasibility range in common mathematical settings

Different constraint classes lead to different structures for feasibility sets and their parameter dependence.

6.1 Linear constraints and polyhedral feasible sets

When constraints are linear in decision variables, the feasible set is polyhedral: it can be represented as intersections of half-spaces and affine subspaces.

6.1.1 One-dimensional parameter sweeps

If only one parameter influences the linear system, feasibility often reduces to inequalities in \(p\), resulting in intervals or unions of intervals. The boundaries correspond to linear constraints becoming tight.

6.1.2 Higher-dimensional half-space intersections

With multiple parameters, the feasibility region in parameter space can still be polyhedral if constraints depend linearly on parameters as well. Then the feasibility range is computable via polyhedral operations or by reducing the feasibility question to linear programs parameterized by \(p\).

6.2 Nonlinear constraints and implicit feasible sets

For nonlinear constraints, the feasible set is typically not polyhedral and may be curved, disconnected, or even fractal-like in pathological cases.

6.2.1 Convex versus non-convex feasibility

If constraints define a convex feasible set (e.g., convex inequalities and convex equalities), feasibility range boundaries often behave more regularly, and certain optimization-based reformulations are more reliable.

Non-convex constraints can produce multiple disjoint feasible components in parameter space, along with transitions that are harder to detect due to local minima or solver dependence.

6.3 Systems of equations with feasibility constraints

Some problems combine strict satisfaction of equations with inequality feasibility conditions. Even when equations define a manifold, inequalities can carve away most of it, leaving feasible regions only where the manifold intersects permitted regions. Parameterized versions lead to feasibility ranges governed by intersection conditions between surfaces.

6.4 Feasibility under uncertainty (scenario interpretation)

Uncertainty modeling can interpret feasibility as the existence of a solution that works across multiple scenarios. In scenario-based feasibility, one enforces constraints for sampled realizations of uncertain parameters; feasibility range then relates to how long a consistent solution can be maintained as uncertainty grows. This can produce conservative but practically useful feasibility thresholds.

7 Visualization and interpretation

Visualization helps interpret what feasibility range means operationally and where boundaries lie.

7.1 Plotting feasible intervals and feasible regions

For one parameter, a plot of feasibility versus parameter typically uses shaded segments on the real line, with open or closed endpoints marked to reflect strictness. For multiple parameters, contour plots or slices can show which parameter pairs admit feasible solutions.

7.2 Interpreting disjoint feasible sets

Disjoint feasible components indicate that feasibility may reappear after an interval of infeasibility. This can happen in nonlinear systems where constraints create multiple intersections across parameter changes. Interpreting disjointness is important for avoiding overly simplistic “largest continuous interval” summaries.

7.3 Communicating results with examples and diagrams

Good communication ties the feasibility range to constraint interpretation:

  • what the boundaries represent (e.g., limiting slack),
  • whether strict inequalities exclude boundary parameter values,
  • and how computed tolerances affect classification.

Simple illustrative diagrams can clarify transitions and reduce ambiguity in reported feasible intervals.

8 Applications and use cases

Feasibility ranges appear across scientific and engineering workflows where constraints matter more than an objective value.

8.1 Engineering design margins (constraint satisfaction thresholds)

In engineering design, one often chooses parameters (materials, gains, geometry) subject to constraints on stresses, stability, or performance. The feasibility range gives a margin: values outside it cannot satisfy all constraints simultaneously, even if an optimization objective would suggest them.

8.2 Control and model tuning (maintaining feasibility during operation)

Control systems may require parameters to stay within allowable regions to avoid constraint violations such as saturation limits or stability conditions. Feasibility range analysis supports tuning that preserves admissibility during expected disturbances.

8.3 Data fitting with feasibility constraints (admissible parameter ranges)

In statistical modeling and system identification, fitted parameters may be restricted by physical plausibility or inequality conditions derived from theory. The feasibility range can summarize which parameter regions are admissible, guiding both model selection and interpretation of fitted results.

9 Pitfalls and best practices

Feasibility range computation can fail or mislead if common issues are overlooked.

9.1 Confusing feasibility range with optimality range

The set of parameters for which an optimum exists or achieves a particular value is not the same as the set for which the constraints are satisfiable. A model can be feasible but yield no meaningful optimum under additional requirements, or it can be infeasible while optimization formulations appear well-behaved.

9.2 Constraint scaling and numerical conditioning

Poorly scaled constraints can cause solvers to misjudge feasibility, particularly near boundaries. Scaling variables and normalizing constraint magnitudes helps ensure tolerances correspond to actual geometric or physical scales rather than artifacts of units.

9.3 Misinterpreting strict inequalities and solver tolerances

Strict inequalities are frequently handled by solvers with tolerances that effectively convert strictness into non-strictness or vice versa. As a result, a reported feasible set may include boundary parameter values that should be excluded under the original strict formulation—or may omit feasible points that lie within tolerance margins.

9.4 Reproducibility of computed ranges

Because numerical feasibility depends on solver settings, tolerances, and initializations, feasibility range computations should document:

  • solver type and version,
  • tolerance parameters,
  • stopping criteria,
  • and any post-processing rules for boundary classification.

This documentation enables consistent re-computation and comparison.

10 Worked examples (illustrative)

These examples illustrate how feasibility ranges arise in practice and how boundaries can be identified.

10.1 Example: feasible interval from a simple inequality

Consider the model with decision variable \(x\in\mathbb{R}\) and parameter \(p\in\mathbb{R}\) subject to \[ x^2 \le p. \] Feasibility requires the existence of some \(x\) satisfying the inequality. Since \(x^2\ge 0\) for all real \(x\), feasibility holds exactly when \(p\ge 0\). Thus the feasibility range is \[ p\in[0,\infty). \] If instead the constraint is strict, \(x^2 &lt; p\), then feasibility requires \(p&gt;0\), giving \[ p\in(0,\infty). \] The only difference is whether boundary parameter values are included, demonstrating the closed versus open interval effect.

10.2 Example: parameter sweep in a small linear program

Suppose variables \(x,y\) must satisfy \[ x+y \ge p,\quad x\ge 0,\quad y\ge 0. \] For a given \(p\), feasibility asks whether there exist nonnegative \(x,y\) with sum at least \(p\).

  • If \(p\le 0\), choose \(x=y=0\) to satisfy \(x+y\ge p\).
  • If \(p>0\), choose for example \(x=p,\; y=0\).

Therefore the feasibility range is again \[ p\in\mathbb{R}. \] In this example, feasibility never fails because the constraints allow arbitrarily large sums, so varying \(p\) does not empty the feasible set.

10.3 Example: boundary identification in a nonlinear constraint system

Let \(x\in\mathbb{R}\) and parameter \(p\in\mathbb{R}\) be constrained by \[ x^2-1 \le p,\quad x^2-1 \ge p. \] Combining them yields \[ x^2-1 = p. \] A feasible solution exists if there is a real \(x\) such that \(x^2 = 1+p\). Real solutions require \(1+p\ge 0\), i.e. \[ p\ge -1. \] Thus the feasibility range is \[ p\in[-1,\infty). \] At \(p=-1\), the condition becomes \(x^2=0\), so feasibility collapses to the single point \(x=0\), illustrating how the boundary can correspond to a degeneracy (a tight constraint with no slack).