{
  "id": "triangulation",
  "title": "Triangulation",
  "lang": "en",
  "source_lang": "en",
  "summary": "Triangulation determines positions from intersecting directions or decomposes geometric spaces into triangles and higher-dimensional simplices.",
  "category": "mathematics",
  "subcategory": "geometry",
  "type": "concept",
  "aliases": [],
  "generated_at": "2026-11-06",
  "model": "pro/gpt-6.1-sol",
  "url": "https://aiwiki.page/en/triangulation/",
  "translations": {
    "zh-Hans": {
      "url": "https://aiwiki.page/zh/triangulation/",
      "json": "https://aiwiki.page/zh/triangulation.json"
    }
  },
  "sections": [
    {
      "heading": "",
      "text": "Triangulation has two closely related meanings in geometry. In measurement, it is the determination of an unknown position from directions observed at known locations, together with a known distance or other information establishing scale. In mathematics, it is the decomposition of a region or space into triangles, or into their higher-dimensional analogues, arranged so that their intersections fit together consistently. Both meanings exploit the fact that complicated geometric relationships can be organized through simple, rigid building blocks.\n\nAngular triangulation underpinned national mapping before electronic distance measurement and satellite positioning became widespread. Its principles remain central to stereo vision, photogrammetry, optical metrology, and some forms of radio direction finding. Geometric triangulations, meanwhile, are fundamental to computational geometry, computer graphics, numerical simulation, and topology. These uses share geometric ideas, but their immediate objectives differ: surveying reconstructs locations from observations, whereas mathematical triangulation constructs a representation of an already specified region or space."
    },
    {
      "heading": "Geometric principles of position determination",
      "text": "The elementary surveying problem begins with two known stations, \\(A\\) and \\(B\\), and an unknown point \\(P\\) visible from both. The segment \\(AB\\) is the baseline. Its length fixes the scale of the construction, while the observed directions \\(AP\\) and \\(BP\\) determine the shape of triangle \\(ABP\\). Without a known baseline or equivalent scale information, angular observations determine similar triangles rather than absolute distances.\n\nLet \\(A=(0,0)\\), \\(B=(b,0)\\), and suppose that \\(P\\) lies above the baseline. If the interior angles at \\(A\\) and \\(B\\) are \\(\\alpha\\) and \\(\\beta\\), the angle at \\(P\\) is\n\n\\[\\gamma=\\pi-\\alpha-\\beta.\\]\n\nThe sine rule, a basic result of trigonometry, gives\n\n\\[AP=\\frac{b\\sin\\beta}{\\sin(\\alpha+\\beta)}, \\qquad BP=\\frac{b\\sin\\alpha}{\\sin(\\alpha+\\beta)}.\\]\n\nIn the associated Cartesian coordinate system, the unknown coordinates are\n\n\\[x=\\frac{b\\sin\\beta\\cos\\alpha}{\\sin(\\alpha+\\beta)}, \\qquad y=\\frac{b\\sin\\beta\\sin\\alpha}{\\sin(\\alpha+\\beta)}.\\]\n\nThese expressions also reveal geometric degeneracy: when the sight lines become nearly parallel or nearly collinear, small angular errors can produce large position errors.\n\nFor example, a baseline of \\(100\\) metres and two base angles of \\(45^\\circ\\) place the point \\(50\\) metres along the baseline and \\(50\\) metres perpendicular to it. Neither station has measured the distance to the point directly. This is the essential advantage of triangulation for inaccessible targets, such as a mountaintop, a point across a river, or a distant feature observed by cameras.\n\nThe terminology depends partly on which quantities are observed. Intersection commonly denotes fixing an unknown target from known observing stations. Resection determines the location of an observing station from directions to known control points. Trilateration instead uses distances: in a plane, three suitable distance observations generally locate a point, subject to consistency and possible ambiguity. Mixed angle-and-distance methods are often called triangulateration.\n\nThree-dimensional triangulation uses spatial sight rays rather than lines confined to a plane. A calibrated camera observation, for example, identifies a ray from the camera centre through an image point. Two ideal rays from distinct centres intersect at the object point. In practice they usually do not meet exactly, so reconstruction requires an estimation criterion. Angles alone still do not supply absolute scale unless the separation of the observing centres, or another physical length, is known."
    },
    {
      "heading": "Survey networks and the shape of Earth",
      "text": "A large survey rarely consists of a single triangle. It uses a network of stations connected by observed directions, with selected distances establishing and checking scale. Once one triangle has been solved, a computed side can become the baseline for an adjacent triangle. Repetition extends positional control across regions where direct measurement of every long distance would be difficult. Closed figures, overlapping observations, and additional baselines provide independent checks against accumulated error.\n\nClassical networks often distinguished primary, secondary, and lower-order control. Primary stations were widely spaced and measured with the highest available precision; denser networks then supported local mapping and engineering. Stations were placed on hills, towers, or other elevated structures to secure intervisibility. The theodolite, an instrument for measuring horizontal and vertical angles, became central to this work. Repeated observations in different instrument orientations helped reduce errors associated with circle graduations and imperfect alignment.\n\nA survey must distinguish physical observations from the mathematical surface on which coordinates are calculated. Over short distances, a plane approximation may suffice. Over large distances, the curvature of Earth matters. Geodesy commonly represents Earth by a reference ellipsoid and relates observed directions and distances to that surface. Instrument heights, terrain elevations, and the distinction between the direction of gravity and an ellipsoid normal can all affect the necessary reductions.\n\nFor a spherical triangle bounded by great-circle arcs, the angle sum exceeds \\(\\pi\\). On a sphere of radius \\(R\\), its spherical excess \\(E\\), measured in radians, is related to its area \\(A\\) by\n\n\\[E=\\frac{A}{R^2}.\\]\n\nThus a failure of large observed triangles to sum to \\(180^\\circ\\) is not automatically evidence of observational error. Ellipsoidal surveying requires more refined calculations involving a geodesic, the locally shortest path on the reference surface. A map projection introduces another distinction: an angle or distance on a flat map need not equal its counterpart on the ellipsoid.\n\nNetwork coordinates also require a reference framework, or geodetic datum. A freely floating network of angular observations has no specified translation, orientation, or scale. Appropriate constraints remove these freedoms and connect the network to an adopted coordinate system. Historically, a datum could be established through an origin, an initial azimuth, and a measured baseline; modern terrestrial reference frames are realized through coordinated observations at many stations.\n\nElectronic distance measurement and satellite techniques changed the balance of observations rather than abolishing network geometry. Modern control may combine angles, distances, height differences, and satellite-derived coordinates. Classical triangulation remains important for interpreting historical maps, transforming older coordinate systems, and understanding how errors and reference choices propagate through a national survey."
    },
    {
      "heading": "Error propagation and network adjustment",
      "text": "Triangulated coordinates are estimates derived from imperfect observations. Angular reading errors, uncertain station positions, baseline errors, atmospheric effects, and instrument misalignment all contribute to measurement uncertainty. Their effects depend strongly on geometry. A small angular error can cause little displacement perpendicular to one sight line but substantial displacement along it, producing elongated uncertainty regions rather than uniformly uncertain positions.\n\nA simple depth model illustrates the importance of baseline length. For a distant target near the perpendicular bisector of a short baseline, the convergence angle \\(\\theta\\) satisfies approximately \\(Z\\approx b/\\theta\\), where \\(Z\\) is target distance. First-order error propagation gives\n\n\\[\\sigma_Z\\approx\\frac{Z^2}{b}\\sigma_\\theta.\\]\n\nWith a \\(100\\)-metre baseline, a target about \\(1\\) kilometre away, and convergence-angle uncertainty of one arcsecond, this approximation gives depth uncertainty of roughly \\(4.8\\) centimetres. The same angular precision at ten times the distance produces about one hundred times the depth uncertainty. The example excludes baseline uncertainty, atmospheric effects, and other systematic contributions.\n\nNetworks are usually designed with more observations than are strictly necessary. Such redundancy makes incompatible observations detectable and permits joint adjustment. In a least-squares formulation, coordinates \\(\\mathbf{x}\\) are chosen to minimize\n\n\\[\\mathbf{r}(\\mathbf{x})^{\\mathsf T} W\\mathbf{r}(\\mathbf{x}),\\]\n\nwhere \\(\\mathbf{r}\\) contains discrepancies between observed and predicted quantities and \\(W\\) is a weight matrix. Angular observations require care because directions are periodic: a residual near the \\(0^\\circ/360^\\circ\\) boundary must represent the small angular difference, not an almost complete revolution.\n\nLinearization uses the Jacobian matrix \\(J\\) of the observation model. If observation errors have known covariance \\(\\Sigma\\), and the reference freedoms have been properly constrained, the local coordinate covariance is approximately\n\n\\[C_{\\mathbf{x}}\\approx \\left(J^{\\mathsf T}\\Sigma^{-1}J\\right)^{-1}.\\]\n\nThe resulting covariance matrix describes both individual coordinate uncertainty and correlations between coordinates. Its interpretation is local and model-dependent; large errors or strongly nonlinear configurations may invalidate a first-order approximation.\n\nRedundancy cannot by itself eliminate systematic error. Atmospheric refraction can bend sight paths, temperature can affect baseline apparatus, and imperfect centring can shift the effective observation point. Repeated measurements under nearly identical conditions may reproduce the same bias. Independent baselines, observations at different times, instrument reversals, and calibration therefore serve purposes distinct from simple repetition.\n\nThe mathematical issue is one of conditioning as well as observational precision. A highly precise instrument does not compensate for nearly parallel sight lines. Network design consequently balances station accessibility, visibility, cost, intersection angles, and redundancy rather than optimizing angular precision alone."
    },
    {
      "heading": "Historical development of triangulation",
      "text": "Geometric methods for finding inaccessible distances preceded organized triangulation networks by many centuries. Greek mathematical astronomy and surveying traditions used triangles, angular measurements, and proportional reasoning. Eratosthenes’ ancient estimate of Earth’s circumference also combined angular and distance information, but it was not a triangulation network in the later geodetic sense. The distinctive development was the systematic transfer of scale through connected triangles over an extended territory.\n\nGemma Frisius gave an influential account of this procedure in 1533, explaining how measured directions from selected stations could support mapping. Willebrord Snellius subsequently applied triangulation to a meridian measurement in the Netherlands. His Eratosthenes Batavus, published in 1617, described work connecting Alkmaar and Bergen op Zoom. The project helped establish a practical model in which a carefully measured baseline supported a much longer chain of triangles.\n\nDuring the seventeenth and eighteenth centuries, French surveys connected triangulation with investigations of Earth’s figure. Jean Picard’s meridian measurement of 1669–1670 used improved angular observations and baseline measurement. Later expeditions to Lapland and the region of present-day Ecuador investigated how meridian arc length changes with latitude. These measurements contributed to establishing Earth’s polar flattening and to resolving disagreements between competing interpretations of earlier surveys.\n\nThe Cassini family’s eighteenth-century mapping of France demonstrated the administrative and cartographic value of a coherent triangulation framework. Another major application was the meridian arc measured by Jean-Baptiste-Joseph Delambre and Pierre Méchain between Dunkirk and Barcelona, beginning in 1792. Their work supported the original definition of the metre through a fraction of the terrestrial meridian. The observations linked an abstract unit of length to a large-scale geodetic measurement, although later definitions of the metre no longer depended on Earth’s dimensions.\n\nIn Britain, the Principal Triangulation developed from work initiated under William Roy in the 1780s and continued into the nineteenth century. The Great Trigonometrical Survey of India began in 1802 under William Lambton and was later directed by George Everest. Such projects required not only angular instruments but also precision baseline equipment, permanent marks, towers, transport, and extensive computation. They supported mapping while reflecting the administrative and political contexts of the institutions that organized them.\n\nTriangulation also helped stimulate statistical methods for combining observations. Carl Friedrich Gauss worked on the triangulation of Hanover from 1821, and geodetic adjustment was an important setting for the development and application of least squares. During the twentieth century, electronic distance measurement, aerial photogrammetry, and satellite geodesy progressively displaced purely angular national networks. Historical triangulation nevertheless survives in monuments, maps, coordinate archives, and the mathematical foundations of modern adjustment."
    },
    {
      "heading": "Photogrammetry, stereo vision, and three-dimensional reconstruction",
      "text": "In photogrammetry and computer vision, triangulation reconstructs a spatial point from its projections into two or more images. The basic imaging model is a pinhole camera: an object point, the camera centre, and its image projection lie on one line. Camera calibration establishes the relationship between image coordinates and spatial rays, accounting for quantities such as focal length, principal point, and lens distortion. Known camera positions and orientations then place those rays in a shared coordinate system.\n\nUsing homogeneous coordinates, an ideal camera is represented by\n\n\\[\\lambda\\mathbf{x}=P\\mathbf{X},\\]\n\nwhere \\(P\\) is a \\(3\\times4\\) camera matrix, \\(\\mathbf{X}\\) represents a spatial point, and \\(\\mathbf{x}\\) its image. Eliminating the unknown scale \\(\\lambda\\) yields linear constraints on \\(\\mathbf{X}\\). Stacking constraints from several views gives a homogeneous system that can be solved approximately using singular value decomposition. This is a useful initialization, but minimizing an algebraic residual is not generally equivalent to minimizing an image measurement error.\n\nFor parallel, rectified stereo cameras, the depth relationship takes the especially simple form\n\n\\[Z=\\frac{fB}{d},\\]\n\nwhere \\(f\\) is focal length in pixel units, \\(B\\) is camera separation, and \\(d\\) is horizontal image disparity. If \\(f=1000\\) pixels, \\(B=0.10\\) metres, and \\(d=20\\) pixels, the reconstructed depth is \\(5\\) metres. A disparity uncertainty of \\(0.2\\) pixels corresponds, by first-order propagation, to about \\(0.05\\) metres of depth uncertainty at that distance.\n\nFinding the same physical feature in different images is often harder than intersecting the resulting rays. Repetitive patterns can produce false matches; smooth surfaces may offer few identifiable features; and occlusion can make a point visible in only one image. The geometry of image pairs, described through projective geometry, restricts a potential match to an epipolar line but does not uniquely determine it. Reflective and transparent objects can also violate assumptions that image features correspond to stable surface points.\n\nBecause noisy rays normally fail to intersect exactly, reconstruction often minimizes reprojection error: the discrepancy between observed image points and projections of the estimated spatial point. Joint refinement of camera parameters and many object points is called bundle adjustment. Its weighting should reflect image measurement uncertainty, while robustness to incorrect correspondences requires additional statistical or algorithmic treatment.\n\nImages alone also leave reference freedoms. For calibrated cameras, reconstruction without external scale information is generally determined only up to a similarity transformation, including an arbitrary scale. Known distances, surveyed control points, or independently measured camera motion can establish physical units. Pure camera rotation supplies no baseline and therefore cannot determine finite point depths through ordinary triangulation."
    },
    {
      "heading": "Optical metrology, direction finding, and astronomical parallax",
      "text": "Triangulation instruments need not use two passive cameras. A common optical arrangement projects a laser spot or stripe from a known position and observes it with an offset camera. The projector defines a known ray or light plane; the camera supplies another geometric constraint. Their intersection determines the illuminated surface location. A projected stripe can produce an entire profile, and repeated profiles can be assembled into a three-dimensional surface.\n\nThis principle is used in structured-light scanning, where projected patterns make correspondence easier than it would be on an untextured object. Encoded stripes or sequences of patterns identify which projector ray belongs to each camera pixel. The method depends on stable calibration of the projector-camera geometry. Shadows, occlusions, changing surface reflectance, and motion during a pattern sequence can create missing or incorrect measurements. Increasing the baseline may improve depth sensitivity while worsening visibility differences between projector and camera.\n\nOptical triangulation differs physically from time-of-flight ranging. A triangulation sensor infers distance from angular displacement, whereas a time-of-flight device infers it from signal travel time or an associated phase measurement. Both may use lasers and produce point clouds, but they have different error models and operating constraints. Similarly, satellite navigation is often called triangulation in everyday language, although its principal geometric observations are ranges or pseudoranges rather than measured sight angles.\n\nRadio direction finding applies angular intersection to transmitters. Two or more receivers estimate arrival directions, and the intersection of their bearings provides a transmitter location. Direction estimates may come from directional antennas or antenna arrays. Reflections from buildings and terrain can cause an estimated bearing to follow an indirect propagation path rather than the direct line to the source. Time-difference-of-arrival localization is another method: a measured difference in propagation time defines a hyperbola in a plane, or a hyperboloid in space, rather than a bearing line.\n\nAstronomical parallax uses the same baseline principle at a very different scale. Annual stellar parallax measures an apparent angular displacement associated with Earth’s changing position in its orbit. Conventionally, the parallax angle \\(p\\) corresponds to a baseline of one astronomical unit. The distance in parsecs is\n\n\\[d_{\\mathrm{pc}}=\\frac{1}{p_{\\mathrm{arcsec}}}.\\]\n\nThe first successful published stellar parallax measurements date to the late 1830s, including Friedrich Bessel’s 1838 measurement of 61 Cygni.\n\nModern astrometry separates parallax from proper motion and other effects through observations over time. For noisy measurements, simply taking the reciprocal of a measured parallax need not yield a statistically well-behaved distance estimate, especially when the uncertainty is large relative to the parallax. The geometry is elementary, but reliable inference depends on the observation model and its uncertainties."
    },
    {
      "heading": "Triangulating polygons and planar regions",
      "text": "In computational geometry, a triangulation of a simple polygon partitions its interior into triangles whose vertices are polygon vertices and whose interiors do not overlap. Internal triangle edges are diagonals lying inside the polygon. This construction does not infer unknown coordinates from measured angles; it reorganizes a known region into manageable pieces. It supports area calculation, rendering, geometric queries, and later construction of numerical meshes.\n\nEvery simple polygon with \\(n\\geq3\\) vertices has a triangulation. In the nondegenerate case, every such triangulation contains exactly \\(n-2\\) triangles and \\(n-3\\) internal diagonals. These counts follow from elementary planar counting or from the Euler characteristic. They are independent of whether the polygon is convex. Convexity affects which diagonals are permitted, but not the number of triangles required once a valid triangulation has been formed.\n\nA useful constructive result is the two-ears theorem. An ear consists of three consecutive polygon vertices whose triangle lies inside the polygon and contains no other polygon vertex in its interior. Removing an ear reduces the problem to a smaller polygon. Straightforward ear-clipping algorithms typically take quadratic time in the number of vertices, although more sophisticated approaches achieve better bounds. Bernard Chazelle established a linear-time algorithm for simple-polygon triangulation, published in 1991; its theoretical efficiency comes with substantial implementation complexity.\n\nA convex \\(n\\)-gon usually admits many distinct triangulations. Their number is the Catalan number\n\n\\[C_{n-2}=\\frac{1}{n-1}\\binom{2n-4}{n-2}.\\]\n\nThus a convex quadrilateral has two triangulations, a pentagon has five, and a hexagon has fourteen. A local edge flip replaces one diagonal of a convex quadrilateral formed by two adjacent triangles with the other diagonal. Such operations connect geometric triangulation to combinatorics and provide a mechanism for improving or navigating triangulations.\n\nRegions with holes require additional bookkeeping. If a connected planar region has \\(n\\) boundary vertices, \\(h\\) holes, and no added interior vertices, a triangulation has\n\n\\[T=n+2h-2\\]\n\ntriangles. Inserting \\(s\\) interior vertices changes the count to \\(T=n+2h+2s-2\\), under the usual conforming assumptions. A self-intersecting polygonal curve requires an explicit convention for which parts of the plane constitute the intended interior before triangulation is well defined.\n\nPractical algorithms must distinguish valid geometry from nearly degenerate input. Repeated vertices, extremely short edges, collinear vertices, and almost intersecting segments can complicate decisions. Robust orientation and intersection tests are therefore as important as the high-level decomposition method."
    },
    {
      "heading": "Delaunay triangulation and Voronoi duality",
      "text": "For a finite set of points in the plane, the problem is not merely to subdivide a prescribed polygon but to connect the points into a useful triangular network. The Delaunay triangulation is characterized by the empty-circumcircle condition: the open circumdisk of every triangle contains no other input point. It is named after Boris Delaunay, who published foundational work on the construction in 1934. For a noncollinear planar point set with no four points on one circle, the triangulation is unique.\n\nIts central geometric relationship is with the Voronoi diagram. Each Voronoi cell contains the locations at least as close to its associated input point as to any other. In the nondegenerate planar setting, two sites share a Delaunay edge when their Voronoi cells share an edge, while a Voronoi vertex corresponds to the circumcentre of a Delaunay triangle. This duality connects triangular meshes to nearest-neighbour geometry.\n\nDelaunay triangulations favour well-shaped triangles in a precise but limited sense: among triangulations of a fixed planar point set, they maximize the smallest angle. This helps avoid some unnecessarily thin triangles, but cannot guarantee good angles for arbitrarily distributed input points. Closely spaced or nearly collinear sites can force poor elements. If four or more points lie on the same empty circle, the underlying Delaunay subdivision can contain a polygonal face, and a choice of diagonals is needed to obtain a triangulation.\n\nA deeper explanation uses a lifting construction. Map each planar point \\((x,y)\\) to\n\n\\[(x,y,x^2+y^2)\\]\n\non a paraboloid in three dimensions. The lower faces of the lifted points’ convex hull, projected back into the plane, give the Delaunay subdivision. This turns a circle-emptiness problem into a convex-hull problem and generalizes naturally to higher dimensions. Weighted variants alter the lifting heights and lead to regular triangulations.\n\nPlanar Delaunay triangulations can be constructed in \\(O(n\\log n)\\) time by standard algorithms. Randomized incremental methods insert points and restore the empty-circle property through local changes. Divide-and-conquer methods build and merge smaller triangulations. Reliable implementations need robust orientation and in-circle predicates, because an incorrect sign caused by floating-point arithmetic can corrupt the mesh’s combinatorial structure.\n\nBoundary requirements lead to constrained Delaunay triangulations, which preserve specified segments while adapting the ordinary Delaunay condition. In three dimensions, the analogous elements are tetrahedra with empty circumspheres. The favourable planar angle property does not carry over straightforwardly: three-dimensional Delaunay meshes can contain slivers, tetrahedra with very small volume despite having edges of comparable length."
    },
    {
      "heading": "Simplicial complexes and topological triangulation",
      "text": "In topology, triangulation expresses a space through simple pieces and their pattern of attachment. A \\(k\\)-dimensional simplex is the convex hull of \\(k+1\\) affinely independent points. A point is a \\(0\\)-simplex, a segment a \\(1\\)-simplex, a triangle a \\(2\\)-simplex, and a tetrahedron a \\(3\\)-simplex. Higher-dimensional simplices extend the same definition beyond direct visualization.\n\nA simplicial complex is a collection of simplices containing every face of each simplex, with any two simplices intersecting either in a common face or not at all. These compatibility conditions distinguish a triangulation from an arbitrary collection of overlapping triangles. In particular, a vertex of one triangle cannot lie strictly inside an unbroken edge of another in a conforming simplicial complex; that edge must be subdivided.\n\nA triangulation of a topological space \\(X\\) consists of a simplicial complex \\(K\\) and a homeomorphism from its geometric realization \\(|K|\\) onto \\(X\\). This is more flexible than demanding that every piece lie flat inside the original space. A curved surface can be homeomorphic to a piecewise-flat complex even though its intrinsic lengths, angles, and curvature differ from those of the complex. The topological correspondence preserves continuity and connectivity, not necessarily metric geometry.\n\nTriangulations make global invariants accessible through finite combinatorial data. For a finite triangulated surface,\n\n\\[\\chi=V-E+F,\\]\n\nwhere \\(V\\), \\(E\\), and \\(F\\) count vertices, edges, and triangular faces. For a connected, closed, orientable surface of genus \\(g\\),\n\n\\[\\chi=2-2g.\\]\n\nThe sphere has Euler characteristic \\(2\\), while the torus has characteristic \\(0\\). Triangulations also supply chain complexes for computing homology, encoding cycles, boundaries, and holes through algebraic operations on oriented simplices.\n\nSubdivision changes the triangles without changing the underlying topological space. Barycentric subdivision, for example, introduces vertices associated with faces and builds smaller simplices from nested sequences of faces. Such operations make it possible to compare decompositions and approximate continuous maps by combinatorial ones. They also reveal a distinction between a space being triangulable and two triangulations being equivalent in a suitably piecewise-linear sense.\n\nTriangulability is not automatic in every dimension. Smooth manifolds admit compatible triangulations, and topological manifolds in dimensions at most three are triangulable. Edwin Moise established fundamental triangulation and uniqueness results for three-dimensional manifolds in the early 1950s. In contrast, nontriangulable topological manifolds exist in dimension four and in every dimension at least five. Triangulation therefore marks a substantive relationship between continuous and combinatorial structures, not merely a convenient drawing technique."
    },
    {
      "heading": "Numerical meshes, interpolation, and computational limits",
      "text": "Triangulations provide computational domains for approximating functions and solving physical equations. In the finite element method, a region is partitioned into elements, often triangles in two dimensions or tetrahedra in three. A complicated global problem is then represented through local functions and interactions between neighbouring elements. Applications include structural deformation, heat flow, electromagnetic fields, and fluid motion, although the suitability of a particular element type depends on the equation and discretization.\n\nWithin a triangle with vertices \\(A_1,A_2,A_3\\), every point has barycentric coordinates satisfying\n\n\\[\\mathbf{x}=\\lambda_1A_1+\\lambda_2A_2+\\lambda_3A_3, \\qquad \\lambda_1+\\lambda_2+\\lambda_3=1.\\]\n\nInside the triangle, all three coordinates are nonnegative. Given vertex values \\(u_1,u_2,u_3\\), piecewise-linear interpolation takes the form\n\n\\[u_h(\\mathbf{x})=\\lambda_1u_1+\\lambda_2u_2+\\lambda_3u_3.\\]\n\nOn a conforming mesh, neighbouring elements agree along shared edges when they use the same vertex values. The interpolant is continuous, although its gradient generally jumps across element boundaries.\n\nElement size alone does not determine accuracy. Very small angles or nearly degenerate elements can amplify coordinate errors and degrade derivative approximations. Standard interpolation estimates are often stated for shape-regular mesh families, in which elements do not become arbitrarily thin as they shrink. For sufficiently regular functions under appropriate assumptions, linear finite elements commonly yield \\(O(h)\\) error in an energy norm and \\(O(h^2)\\) error in an \\(L^2\\) norm. These rates depend on solution regularity, domain geometry, and the underlying equation.\n\nAdaptive refinement concentrates elements where the current discretization is inadequate, for example near a re-entrant corner or a sharply varying field. A posteriori error indicators guide local refinement, while conformity requirements may force additional subdivisions of neighbouring triangles. Anisotropic methods deliberately use elongated elements when a solution varies rapidly in one direction but slowly in another. Such elements are not intrinsically poor: their quality must be judged relative to the approximation task and the relevant directional scales.\n\nMeshing curved domains adds geometric approximation error. A straight-edged triangulation of a circular boundary does not reproduce the circle exactly. Refinement reduces this discrepancy, while curved or higher-order elements represent the boundary more accurately. A topological triangulation that captures connectivity is therefore not automatically an adequate numerical mesh.\n\nFinally, triangulation becomes computationally more demanding in higher dimensions. A planar Delaunay triangulation has linear output size, whereas a three-dimensional one can contain quadratically many tetrahedra, and higher-dimensional constructions can grow still more rapidly. Memory requirements, geometric predicates, boundary conformity, and element quality all become significant constraints. The central computational trade-off is not simply the number of simplices, but how effectively their arrangement represents the geometry and the quantities being approximated."
    }
  ],
  "links": [
    "geometry",
    "trigonometry",
    "cartesian-coordinate-system",
    "trilateration",
    "earth",
    "geodesic",
    "map-projection",
    "measurement-uncertainty",
    "least-squares",
    "jacobian-matrix",
    "covariance-matrix",
    "calibration",
    "condition-number",
    "carl-friedrich-gauss",
    "computer-vision",
    "homogeneous-coordinates",
    "singular-value-decomposition",
    "projective-geometry",
    "polygon",
    "euler-characteristic",
    "combinatorics",
    "delaunay-triangulation",
    "voronoi-diagram",
    "convex-hull",
    "floating-point-arithmetic",
    "topology",
    "simplex",
    "homeomorphism",
    "finite-element-method",
    "interpolation"
  ],
  "sources": [],
  "license": "https://creativecommons.org/licenses/by/4.0/"
}