No, a shape does not have to be closed in geometry. The word "shape" covers open curves, open regions, and disconnected figures just as legitimately as closed polygons. What changes is which theorems apply: closure is a property that unlocks area, perimeter-as-boundary, and most classical polygon results, not a membership requirement for being a shape.
Geometry textbooks often muddle this because K–12 instruction spends years on polygons and circles, both closed by definition, so students absorb "closed" as part of the definition of shape rather than as a property some figures carry. Working mathematicians and applied modelers treat openness and closure as a toggle with consequences. The distinction matters in computer graphics, topology, computational geometry, and drafting standards, where an open path and a closed path behave very differently in code and in measurement.
This explainer separates the definitional question from the consequential one. First, what counts as a shape. Second, why closure changes what you can compute. Third, where the boundary between open and closed gets genuinely ambiguous, and how practitioners resolve it.
What "shape" means across geometry's subfields
In everyday usage, a shape is any distinguishable bounded form. In Euclidean plane geometry, the foundational objects are points, lines, rays, segments, curves, and regions. A line segment is a shape in the loose sense and is open at both ends. A circle is closed. A parabola is an open curve; the region it bounds with a chord is closed.
Topology sharpens this considerably. A set is closed if it contains all its limit points, and open if every point has a neighborhood inside the set. A circle as a point set is closed. An open disk, the interior without its boundary circle, is open. A half-open interval like [0, 1) is neither. None of these fail to be shapes; they are shapes with different topological characters.
The practical upshot: "shape" is a pre-theoretical category, and closure is a technical property you check on a per-figure basis. Geometry does not have a universal closedness axiom. It has definitions of specific classes, and closure appears inside some of them: a polygon is a closed polygonal chain, a polyhedron is a closed surface, and so on. Outside those classes, openness is normal.
There is also a distinction between a closed curve and a closed set that trips up even professionals. A closed curve is one whose endpoints coincide, like a loop of string tied together. A closed set is a topological property about limit points. These coincide for simple loops but diverge for messier figures. A figure-eight curve is a closed curve but self-intersecting; the set it traces is closed, but it is not a simple closed curve, and the Jordan curve theorem does not apply to it in the naive way.
Why closure unlocks area and boundary theorems
The mechanism behind most closure-dependent results is the Jordan curve theorem. A simple closed curve in the plane divides the plane into exactly two regions, an inside and an outside, with the curve as the shared boundary. That inside/outside split is what licenses the notion of enclosed area as a well-defined quantity.
Open curves lack this split. An open arc has no interior, so "area enclosed" is undefined rather than zero. You can still compute the area under a curve between two x-values, but that is a different operation: it uses the curve plus a reference axis to manufacture a closed region. The closure is supplied by the axis, not by the curve.
This is why Green's theorem, the planar form of Stokes' theorem, is stated for a positively oriented, piecewise-smooth, simple closed curve. The theorem converts a line integral around the boundary into a double integral over the enclosed region. Without closure, the boundary integral does not correspond to a region integral. The theorem is not false for open curves; it is inapplicable, because one side of the equation has no region to integrate over.
Polygon area formulas inherit the same dependency. The shoelace formula computes signed area from vertex coordinates and implicitly assumes the vertices form a closed loop, with the last vertex connecting back to the first. If you feed it an open chain, you get the area of the polygon formed by closing the chain with a straight segment from last to first vertex, whether or not you intended that. The formula silently closes the figure for you. This is a common source of error in computational pipelines: an open polyline's "area" is really the area of its implied closure.
Perimeter behaves differently and more subtly. For a closed figure, perimeter is the full boundary length. For an open curve, the natural measure is arc length, and it is perfectly well-defined. The issue is that arc length of an open curve is not a boundary in the Jordan sense. Practitioners often call it perimeter anyway, which is loose but harmless if everyone knows the figure is open.
Plain-language version: closure is what lets "inside" and "boundary" be meaningful at the same time. Take away closure and you can still measure length along the curve, but you lose a principled inside.
Where closure is genuinely ambiguous
Real figures often sit between open and closed, and the resolution depends on the modeling context rather than on pure mathematics.
Pixel and raster boundaries are the classic case. A shape drawn on a grid, like a region in an image, has a boundary that is ambiguous at the pixel level: the boundary can be assigned to the inside pixels, to the outside pixels, or to the edges between pixels. Digital topology handles this with paired conventions, such as treating foreground as 8-connected and background as 4-connected, so that the boundary is consistently owned by one side. The figure is closed in the discrete sense, but which pixels constitute the boundary depends on the convention chosen.
Vector graphics distinguish open paths from closed paths explicitly. A path has a start and end; whether a closing segment is drawn is a flag in the data. Filling an open path forces the renderer to close it implicitly. This is why the same coordinate list can render as an outline or as a filled blob depending on one boolean. Adobe's PostScript and SVG specifications both formalize this open-versus-closed path distinction, and it propagates into CAD drafting standards where an open contour is not a valid boundary for a filled region.
Limit cases in analysis add another layer. An open interval is open; its closure adds the endpoints and is closed; a half-open interval is neither open nor closed. These are all legitimate subsets of the real line and all are shapes in any reasonable sense. The lesson is that open/closed is not a binary over all figures; it is a classification that sometimes yields neither.
Polygons with repeated vertices or collinear points produce degenerate closures. A "closed" polygon whose vertices backtrack along the same edge is closed as a point set but has zero or negative signed area contributions in the shoelace formula. The closure flag is satisfied; the geometric integrity is not. Software that checks only the closed flag will accept a figure that fails area sanity checks downstream.
One professional caveat that surprises people: closedness is not preserved under common geometric operations in the way closure-preserving language suggests. The union of two closed sets is closed, but the union of infinitely many closed sets need not be closed. The intersection of closed sets is closed, but the union of open sets is open while the intersection of open sets need not be. In practice, boolean operations in CAD on closed solids can produce non-manifold or open results when inputs touch tangentially or share only an edge. The closure of the input does not guarantee the closure of the output. That is why mesh-repair tools exist and why watertightness checks are a separate step from solid modeling.
The misconceptions that cause real errors
Misconception one: "a shape must be closed to have an area." An open curve has no enclosed area, but many open figures do have well-defined areas once you specify the reference boundary. The area under a curve, the area between two curves, and the area of a region bounded by a curve and an axis are all standard and all involve open pieces. What fails without closure is area of the figure itself as an enclosed region.
Misconception two: "open means no boundary." Every open set in the plane can have a boundary; the open disk's boundary is the circle, which is not part of the disk. Openness is about which points are members, not about whether a boundary exists.
Misconception three: "closed and bounded are the same." Closedness and boundedness are independent. A line is closed and unbounded. An open disk is bounded and open. The Heine–Borel theorem links them only in finite-dimensional Euclidean space: a subset of R^n is compact exactly when it is closed and bounded. That is a theorem about compactness, not a definition of shape.
Misconception four: "a polygon is any multi-sided figure." Under the standard definition, a polygon is a closed chain of finitely many segments meeting only at endpoints. Self-intersecting closed chains are sometimes called complex polygons, but the simple-polygon theorems, including the Jordan curve theorem applied naively and many triangulation and point-in-polygon algorithms, do not hold for them without modification.
Red-flag warnings for anyone building or grading geometric work:
Treating an open polyline as if its shoelace "area" were the intended enclosed area, when the closing segment was never part of the design. Accepting a figure as a valid region because it carries a closed flag, without checking for self-intersection, repeated vertices, or zero-length edges. Assuming that boolean operations on closed inputs always yield closed outputs; tangential contact and coplanar face sharing routinely break this. Confusing "closed curve" with "closed set"; the properties coincide for simple loops but not for self-intersecting or degenerate figures. Using raster boundary conventions inconsistently between foreground and background, which produces off-by-one boundary ownership and leaks in flood-fill operations.
Open versus closed: what changes in practice
The comparison below is the practical core of the question. The same figure, toggled between open and closed, changes which operations are defined.
| Property or operation | Open curve or path | Closed curve or region |
|---|---|---|
| Enclosed area | Undefined without a supplied reference boundary | Well-defined, signed area via shoelace or Green's theorem |
| Inside/outside classification | Not meaningful | Meaningful for simple closed curves (Jordan curve theorem) |
| Arc length | Defined | Defined, and equals full boundary length |
| Green's / Stokes' theorem | Not directly applicable | Applicable to the boundary integral over the region |
| Point-in-figure testing | Requires alternative formulation | Standard point-in-polygon tests apply |
| Fill in vector graphics | Renderer closes implicitly or refuses | Native fill operation |
| Typical uses | Graphs, trajectories, splines, scans | Polygons, disks, solids, regions |
The counterintuitive detail worth internalizing: for many computations, the open figure is the more general object and the closed figure is a special case obtained by adding one segment or one limit. Area is the exception where closure is genuinely load-bearing. Practitioners who treat closure as universal get into trouble on area; those who treat it as never required underspecify regions.
Practical actions and thresholds
Concrete steps that resolve the open/closed question for a given figure before it causes downstream errors.
Decide the modeling intent first. If you need enclosed area, inside/outside, or a region integral, the figure must be closed or explicitly closed by a reference boundary. If you need length, trajectory, or a curve fit, openness is fine and often required. Check closure mechanically. Verify that first and last vertices coincide within a tolerance, and that consecutive segments meet only at shared endpoints. For vector paths, read the closed-path flag rather than inferring from coordinates. Detect self-intersection before trusting simple-closed results. Simple-polygon algorithms (sweep-line intersection detection and triangulation) assume no crossings; run an intersection check first and reject or repair crossings. Validate signed area direction. A consistent winding order (counterclockwise positive under standard conventions) is required for shoelace and Green's-theorem applications. Mixed winding produces signed area that understates or cancels true area rather than revealing an error, so compute the absolute value only after confirming orientation. For raster boundaries, fix the connectivity convention once and document it. The standard pairing is 8-connectivity for foreground with 4-connectivity for background, or the reverse; mixing them produces boundary leaks and double-counted pixels. For solid modeling, run a watertightness or manifold check as a separate validation step after every boolean operation, not as an assumption inherited from closed inputs. When a figure is half-open or otherwise neither, state the convention explicitly in documentation rather than defaulting to one side silently.
These steps reconcile the thresholds that appear in different tools. Closure tolerances in CAD are typically set as absolute distances on the order of the drawing units' precision (for a millimeter-based drawing, sub-millimeter coincidence criteria), while mesh watertightness is a topological condition on edge and vertex incidence rather than a distance threshold. Both are legitimate; they answer different questions, and conflating them is a frequent source of "the model looks closed but is not" failures.
What closure does not settle
Closure answers a narrow set of questions and leaves many others untouched. A closed figure can still be non-simple, non-manifold, self-intersecting, degenerate, or badly oriented. Closedness is neither necessary nor sufficient for most properties people care about in applied geometry, including convexity, regularity, and computability. A closed polygon can be concave, self-touching, or have collinear consecutive edges that break triangulation assumptions.
There is also a rank-of-operations caution. Topological closure, the smallest closed set containing a given set, is a different operation from geometric closing, a morphological operation that fills small gaps and smooths boundaries. The two share a word and nothing else. Confusing them in image processing or shape analysis leads to arguments where both parties are correct about different operations.
Finally, the original question has a clean answer that survives all the nuance: a shape does not have to be closed. Geometry defines many open figures, uses them routinely, and provides well-posed measurements for them. Closure is a property that matters for specific theorems and operations, chiefly area, region integrals, and inside/outside classification. The productive habit is to ask, for any given figure, which operations you intend to run, and then to check whether closure is required for those operations specifically. That question has a definite answer in every case where the underlying mathematics is well-defined, and stating the intent up front prevents nearly all of the confusion the loose phrasing of the original question invites.