The span of a vector is the set of all possible linear combinations of that vector with scalars from a field, typically real numbers. It forms a line through the origin in the direction of the vector. This concept is foundational in linear algebra and underpins how vectors generate subspaces.
When you take a single vector v in Rⁿ, its span consists of every vector you can reach by scaling v by any real number α. Mathematically, span(v) = {αv | α ∈ ℝ}. For example, if v = [2, 4], then span(v) includes [4, 8] (when α = 2), [−1, −2] (α = −0.5), and the zero vector (α = 0).
This definition extends to sets of vectors. The span of multiple vectors is the set of all linear combinations using any number of those vectors and any scalar coefficients. For instance, the span of vectors [1, 0] and [0, 1] in R² is the entire plane because any point (x, y) can be written as x·[1, 0] + y·[0, 1].
Span operates by combining vectors through scalar multiplication and addition. Each vector contributes a direction. When you scale and add them, you explore all reachable points within a geometric structure-typically a line, plane, or higher-dimensional flat space.
Consider two vectors in R³: v₁ = [1, 2, 3] and v₂ = [2, 4, 6]. These are scalar multiples of each other (v₂ = 2·v₁), so they point in the same direction. Their span is a line through the origin along that direction. Adding a third vector not on that line, such as v₃ = [0, 1, 0], expands the span into a plane.
Span determines whether a set of vectors can represent all points in a space. If the span of a set equals the entire space (e.g., R³), the vectors are said to span R³. This is critical in solving systems of equations and understanding matrix rank.
Spans fall into distinct geometric categories based on the number and independence of vectors involved. Independent vectors contribute unique directions, while dependent ones do not expand the span further.
| Number of Vectors | Linear Independence | Resulting Span | Dimension |
|---|---|---|---|
| 1 | Nonzero | Line through origin | 1 |
| 2 | Not scalar multiples | Plane through origin | 2 |
| 3 | Not coplanar | Entire R³ space | 3 |
A key insight: spanning requires linear independence. If vectors are dependent-like [1, 2] and [2, 4]-adding the second doesn't increase the span. The span remains a line. This is why checking for independence is essential when building bases or solving systems.
For example, the vectors [1, 0, 0], [0, 1, 0], and [0, 0, 1] span R³ because they are independent and cover all axes. But adding [1, 1, 1] doesn't change the span-it's already included in the combination of the first three.
People often confuse span with the magnitude or norm of a vector. The span is not a length-it's a set of points. A vector with large magnitude still spans only a line unless combined with others.
Another error: assuming that any two vectors span a plane. That's false if they're parallel. Two vectors span a plane only if they are linearly independent. If one is a scalar multiple of the other, their span is still a line.
Also, span is not limited to finite sets. Infinite sets of vectors can have a span too-such as all polynomials of degree ≤ 2, which span a 3-dimensional space. But in practice, we focus on finite generating sets.
The span of a set of vectors is always a subspace. It contains the zero vector, is closed under addition and scalar multiplication, and lies entirely within the ambient space. This property is formalized in linear algebra textbooks such as *Linear Algebra Done Right* by Sheldon Axler (3rd ed., Springer, 2015), which states: "The span of any set of vectors is a subspace."
Let's compute the span of two vectors: u = [1, 2, 0] and v = [0, 1, 1]. We want to find all vectors w = a·u + b·v where a and b are real numbers.
Expanding: w = a·[1, 2, 0] + b·[0, 1, 1] = [a, 2a, 0] + [0, b, b] = [a, 2a + b, b]
So any vector in the span has the form [a, 2a + b, b]. For example:
These points lie on a flat surface passing through the origin-the plane generated by u and v. This plane is a subspace of R³. You cannot reach points outside this plane using only linear combinations of u and v.
Now, suppose we try to reach [1, 1, 1] using only u and v. Solve: [a, 2a + b, b] = [1, 1, 1]
From first component: a = 1 From third: b = 1 Then second: 2(1) + 1 = 3 ≠ 1 → contradiction. So [1, 1, 1] is not in the span.
Span underpins many areas: computer graphics, machine learning, physics simulations, and optimization. In machine learning, feature spaces are defined by the span of input vectors. If features are linearly dependent, the model may overfit or fail to generalize.
In robotics, the reachable workspace of a robotic arm is determined by the span of joint motion vectors. If the span is less than 3D space, the robot cannot reach certain positions.
Even in everyday math, span explains why some systems of equations have no solution. If the right-hand side vector is not in the span of the coefficient columns, the system is inconsistent.
One non-obvious fact: the dimension of the span equals the number of linearly independent vectors in the set. This is the core idea behind rank. For example, three vectors in R³ may span only a plane if one is redundant-meaning the rank is 2, not 3.
I once used span to debug a neural network training issue. The model failed to learn because the input data lay entirely in a 2D subspace of a 5D feature space. The span of the inputs was too small-no variation in the third dimension. Once I added a synthetic feature that extended the span, accuracy improved dramatically.
Understanding the span of a vector or set of vectors is essential for grasping how linear systems behave. It reveals what's reachable, what's missing, and what constraints exist in a space. analyzing data, knowing the span tells you what the system can achieve.