An axis-aligned bounding box encloses points with faces parallel to the coordinate axes.
dotnet add package RecursiveAI.OneMathRequires .NET 10 (net10.0); .NET 9 and earlier aren’t supported.
using RecursiveAI.OneMath; var box = Bounds3D.CreateFromPoints([ new(-2, 1, 0), new(4, 5, 6)]); Console.WriteLine(box.Center); Console.WriteLine(box.Extents);
(1, 3, 3) (3, 2, 3)
Extents are the half-sizes along X, Y and Z, not the full side lengths.
Unzip the example, then dotnet run.
The full program
using RecursiveAI.OneMath; Vector3D[] points = [new(-2, 1, 0), new(4, 5, 6), new(0, 3, 2)]; var box = Bounds3D.CreateFromPoints(points); Console.WriteLine($"Min: {box.Min}"); Console.WriteLine($"Max: {box.Max}"); Console.WriteLine($"Center: {box.Center}"); Console.WriteLine($"Extents: {box.Extents}"); Console.WriteLine($"Volume: {box.Volume()}"); var empty = Bounds3D.CreateFromPoints([]); Console.WriteLine($"Empty input: {empty.IsEmpty}"); Console.WriteLine($"Empty volume: {empty.Volume()}");
dotnet run prints:
Min: (-2, 1, 0) Max: (4, 5, 6) Center: (1, 3, 3) Extents: (3, 2, 3) Volume: 144 Empty input: True Empty volume: 0
Empty input returns Bounds3D.Empty, not an exception; check IsEmpty before reading the center or extents.
Variations
Accumulate points without storing the entire point set:
using RecursiveAI.OneMath; var box = Bounds3D.Empty; foreach (var point in new Vector3D[] { new(-2, 1, 0), new(4, 5, 6) }) box = box.Encapsulate(point); Console.WriteLine(box.Min); Console.WriteLine(box.Max);
(-2, 1, 0) (4, 5, 6)
Related recipes
The short answer
Use Bounds3D.CreateFromPoints(points) to calculate an axis-aligned bounding box (AABB) from 3D points in C#. The result stores the smallest and largest coordinates in Min and Max. Read Center for the midpoint and Extents for the half-size along each axis. For single-precision input, use Bounds3F and Vector3F instead.
The example’s third point lies inside the box, so adding it does not change the corners. The bounds source defines the construction and empty-input behavior.
How do you calculate an axis-aligned bounding box from points in C#?
For each axis, keep the minimum and maximum coordinate seen so far. OneMath does this in input order, starting with its empty box and calling Encapsulate for each point. The method accepts a ReadOnlySpan<Vector3D>, so an array can be passed directly without building a separate collection of corners.
For the program’s points, the X range is −2 to 4, Y is 1 to 5, and Z is 0 to 6. The full size is therefore (6, 4, 6). Extents are half that size: (3, 2, 3). Center is computed as Min + (Max - Min) * 0.5, giving (1, 3, 3). Extents are not full side lengths; confusing the two doubles the size when reconstructing a box.
If you already know a center and half-sizes, CreateFromCenterExtents is the corresponding factory. The two-corner constructor is different: it keeps the supplied corners as given and does not sort them. Use CreateFromPoints for unordered corners rather than assuming the constructor will repair their order.
How do you grow a bounding box one point at a time?
Use the variation above: start from Bounds3D.Empty and assign the result of Encapsulate back to the variable. Bounds are immutable values; calling the method without keeping its return value does not grow the box. This also lets you accumulate bounds without retaining every input point.
Do not start with default(Bounds3D) unless you intend to include the origin. Default fields are zero, so that value represents a zero-size box at the origin, not an empty box. A single-point input similarly produces a nonempty, zero-size box located at that point.
What happens if there are no points?
CreateFromPoints returns Empty: every minimum component is positive infinity and every maximum component is negative infinity. IsEmpty checks whether any minimum exceeds its matching maximum. Volume() and SurfaceArea() explicitly return zero for an empty box, but the center and half-sizes are arithmetic expressions over those infinite corners, not usable geometry.
Branch on IsEmpty before placing a camera or sizing an object from the result. Keep finite coordinates in point data meant for geometry; the factory is an accumulation operation, not an input-validation layer. The evaluation-order documentation specifies the component operations, including their handling of NaN. If you later serialize bounds to ordinary JSON, handle emptiness separately because its infinite corners cannot be written with default JSON options.