Linear Algebra · Linear span

Linear span

The span of some vectors is the set of all their linear combinations. In space, two vectors that are not collinear span a plane through the origin. A third vector adds a dimension only if it sticks out of that plane, even by a hair; if it lies in the plane it adds nothing, and the three are linearly dependent.

Linear span in this visualization

The span of some vectors is the set of all their linear combinations, with any real numbers as coefficients. It always contains the origin O, and in space it is O alone, a line or a plane through O, or all of space.

The span of three vectors
span⁡(v⃗,w⃗,u⃗)={ c1v⃗+c2w⃗+c3u⃗  :  c1,c2,c3∈R }\operatorname{span}(\vec v, \vec w, \vec u) = \{\, c_1\vec v + c_2\vec w + c_3\vec u \;:\; c_1, c_2, c_3 \in \mathbb R \,\}

In the opening view, the glass disc through O is WW, the span of v⃗\vec v and w⃗\vec w. The crossings of its engraved grid are the combinations with integer coefficients c1v⃗+c2w⃗c_1\vec v + c_2\vec w, and its rim fades because the plane goes on forever. The fainter discs are the layers W+ku⃗W + k\vec u for integers kk, and the blue disc is the plane W+su⃗W + s\vec u through the target point T. A chain of thin blue arrows, c1v⃗c_1\vec v, then c2w⃗c_2\vec w, then su⃗s\vec u, walks from O to T. The red segment from the tip of u⃗\vec u down to WW has length ∣h∣|h|, where hh is the signed height of u⃗\vec u off the plane.

Why can a third vector add a whole dimension? Every combination of v⃗\vec v, w⃗\vec w and u⃗\vec u is a point of WW moved by some multiple ss of u⃗\vec u, so the span is the union of the planes W+su⃗W + s\vec u over all real ss. If u⃗\vec u is off WW, the planes at ss and s+1s + 1 are ∣h∣|h| apart, and as ss runs through the real numbers the planes sweep through all of space.

Now lower u⃗\vec u until it is only a hair off the plane. With the default v⃗=(2,1,0)\vec v = (2, 1, 0) and w⃗=(0,1,2)\vec w = (0, 1, 2), u⃗=(1,1,1.01)\vec u = (1, 1, 1.01) is about 0.004 off WW, and the span is still all of space. Reaching (0,0,1)(0, 0, 1) takes s=100s = 100: (0,0,1)=−50v⃗−50w⃗+100u⃗(0, 0, 1) = -50\vec v - 50\vec w + 100\vec u. For a fixed target off WW, the smaller ∣h∣|h| is, the larger ∣s∣|s| has to be, but there is always an answer. Only when hh is exactly 0 does every plane W+su⃗W + s\vec u stay WW, and the span of v⃗\vec v, w⃗\vec w and u⃗\vec u drops back to WW.

Vectors are linearly dependent if some combination of them with coefficients not all zero is the zero vector. When the vectors are nonzero, the page draws this as a closed loop: multiples of the vectors, laid tip to tail, walk back to O. Equivalently, one of them is a combination of the others. No two of v⃗\vec v, w⃗\vec w and u⃗=(1,1,1)\vec u = (1, 1, 1) are parallel, yet the three are dependent, because u⃗\vec u lies in WW:

A third vector inside W closes a loop
12v⃗+12w⃗−u⃗=0⃗,u⃗=12v⃗+12w⃗\tfrac12\vec v + \tfrac12\vec w - \vec u = \vec 0, \qquad \vec u = \tfrac12\vec v + \tfrac12\vec w

A loop also means that a point has many recipes. Adding a nonzero multiple of the loop to a recipe for T changes the coefficients but not the point.

The dimension of the span is the largest number of independent vectors among them.

Things to notice

  • In perspective, a tip can look as if it touches a plane that it misses. The red segment and the height readout tell you, and Edge-on shows WW as a single line.

Controls

  • The red fader under Third vector u moves u⃗\vec u straight off the plane or back into it, and catches at 0. The wheels p and q move u⃗\vec u within its layer.

Related