Parametric Vector Form of a Solution Set

Once you have a system in reduced row echelon form, you still need to translate that RREF into something geometrically meaningful. Parametric vector form does exactly that: it packages the entire solution set into a single expression that separates the particular solution from the degrees of freedom. The result is compact, readable, and directly tied to the underlying geometry of the system.

From RREF to Parametric Vector Form

After row-reducing a system, each row either pins a basic variable or disappears entirely. The free variables are the ones left uncontrolled -- they can take any real value, and every choice produces a different solution.

Suppose a system in three unknowns reduces to:

x1 - 2x3 = 4
x2 + 3x3 = -1
0  =  0

Here x3 is free (call it t), and x1, x2 are basic. Solving for the basics:

x1 = 4 + 2t
x2 = -1 - 3t
x3 = t

Writing the solution vector column by column:

x = [x1]   [4 + 2t ]   [4]       [2 ]
    [x2] = [-1 - 3t] = [-1] + t·[-3]
    [x3]   [  t    ]   [0]       [1 ]

That is the parametric vector form:

x = p + t·v

where p = (4, -1, 0) is one particular solution and v = (2, -3, 1) is the direction vector associated with the free variable.

The Geometry: Point Plus Span

The expression x = p + t·v describes a line through p in the direction of v. When t = 0 you land at p; as t varies, you slide along the line.

More generally, if there are two free variables s and t, the solution set is:

x = p + s·v1 + t·v2

This is a plane (or a flat of higher dimension if there are more free variables). The set Span{v1, v2} is called the solution set of the associated homogeneous system, and p shifts it away from the origin.

Key observation: p + Span{v1, v2} is not a subspace unless p = 0. It is an affine subspace -- a translated copy of a genuine subspace. Geometrically, you can think of it as nailing down one solution, then freely moving within the homogeneous solution space.

Homogeneous vs. Nonhomogeneous Systems

A homogeneous system Ax = 0 always has x = 0 as a solution, so the particular solution can be taken as p = 0. The parametric vector form simplifies to:

x = s·v1 + t·v2 + ...

This is literally Span{v1, v2, ...}, a subspace passing through the origin. That subspace is the null space of A. See how to find a basis for the null space for the full procedure.

For a nonhomogeneous system Ax = b (with b nonzero), the solution set is either empty or a shifted copy of that same null space. If you find any single particular solution p, then:

Solution set = {p + xh : xh satisfies Axh = 0}

This decomposition explains why solving the homogeneous system matters even when you care about the nonhomogeneous one: the homogeneous solution set describes the "shape" (the freedom), while p pins down the "location." You can read more about solving homogeneous systems for additional context.

Worked Example: Two Free Variables

Consider the augmented matrix:

[1  -3   0   2 | 5]
[0   0   1  -4 | 2]
[0   0   0   0 | 0]

This is already in RREF. Basic variables: x1 (column 1) and x3 (column 3). Free variables: x2 and x4.

Let x2 = s and x4 = t (both arbitrary real numbers).

From row 2: x3 - 4t = 2, so x3 = 2 + 4t. From row 1: x1 - 3s + 2t = 5, so x1 = 5 + 3s - 2t.

Writing the full solution vector:

x = [x1]   [5 + 3s - 2t]
    [x2] = [     s      ]
    [x3]   [2 + 4t      ]
    [x4]   [     t      ]

Split by terms:

x = [5]       [3]       [-2]
    [0] + s · [1] + t · [ 0]
    [2]       [0]       [ 4]
    [0]       [0]       [ 1]

So p = (5, 0, 2, 0), v1 = (3, 1, 0, 0), v2 = (-2, 0, 4, 1).

Sanity check for p: plug x = (5, 0, 2, 0) into the original equations.

• Row 1: 5 - 3(0) + 0(2) + 2(0) = 5. Correct.
• Row 2: 0 + 0 + 1(2) - 4(0) = 2. Correct.

Sanity check for v1: plug x = (3, 1, 0, 0) into Ax = 0.

• Row 1: 3 - 3(1) + 0 + 2(0) = 0. Correct.
• Row 2: 0 + 0 + 0 - 4(0) = 0. Correct.

Sanity check for v2: plug x = (-2, 0, 4, 1).

• Row 1: -2 - 0 + 0 + 2(1) = 0. Correct.
• Row 2: 0 + 0 + 4 - 4(1) = 0. Correct.

The solution set is a 2-dimensional flat (a plane in R^4) passing through p = (5, 0, 2, 0). You can also verify the result step by step with using RREF to solve linear systems.

Uniqueness and Consistency

If the RREF has a pivot in every column (no free variables), the only solution is p itself -- the solution set is a single point. If the RREF contains a row like [0 0 0 | 1], the system is inconsistent and there are no solutions at all.

Parametric vector form only makes sense in the consistent, non-unique case. That situation arises whenever the number of free variables is at least one, which happens when the rank of A is strictly less than the number of unknowns.

Frequently asked questions

What is parametric vector form, and why use it?

Parametric vector form expresses the entire solution set of a linear system as x = p + s·v1 + t·v2 + ..., where p is one particular solution and v1, v2, ... span the null space of the coefficient matrix. It is useful because it makes the geometry explicit: you can immediately see the dimension of the solution set (number of free variables) and the directions in which solutions vary.

How do I identify the particular solution p?

Set all free variables to zero in your RREF-derived equations and read off the values of the basic variables. The resulting vector is p. It satisfies the full nonhomogeneous system Ax = b. Any other valid choice of p gives an equivalent parametric form -- just shifted by a vector in the null space.

Are the direction vectors v1, v2, ... always linearly independent?

Yes, when you construct them by setting one free variable to 1 and all others to 0 (the standard procedure from RREF), the resulting vectors are automatically linearly independent. Each one has a 1 in a position where the others have 0, so no combination of them can produce the zero vector. They form a basis for the null space.

Does the particular solution p have to be unique?

No. There are infinitely many valid choices for p whenever the null space is nontrivial. Adding any null-space vector to p gives another particular solution. The parametric form itself is not unique (you can pick a different p, or even different spanning vectors for the null space), but the solution set it describes is always the same geometric object.