Find A Basis For The Subspace

Finding a basis for a subspace is a fundamental skill in linear algebra, serving as the bridge between abstract vector spaces and concrete computational methods. Practically speaking, whether you are working with the column space of a matrix, the null space of a linear transformation, or a span of vectors in $\mathbb{R}^n$, the goal remains consistent: identify a minimal set of linearly independent vectors that still spans the entire subspace. Mastering this process allows you to determine the dimension of a space, simplify complex systems, and understand the structural geometry of linear mappings And that's really what it comes down to..

Understanding the Core Concepts

Before diving into algorithms, Make sure you solidify the definitions. It matters. A subspace $W$ of a vector space $V$ is a subset that is closed under vector addition and scalar multiplication Took long enough..

1. Linear Independence: No vector in the set can be written as a linear combination of the others. The only solution to $c_1\mathbf{v}_1 + \dots + c_k\mathbf{v}_k = \mathbf{0}$ is the trivial solution where all $c_i = 0$.
2. Spanning Set: Every vector in $W$ can be expressed as a linear combination of the basis vectors. $\text{Span}{\mathbf{v}_1, \dots, \mathbf{v}_k} = W$.

The number of vectors in any basis for $W$ is the dimension of $W$, denoted $\dim(W)$. While a subspace has infinitely many possible bases, they all share the same cardinality That's the part that actually makes a difference..

Scenario 1: Basis for the Column Space (Col A)

One of the most common tasks is finding a basis for the column space of a matrix $A$. The column space is the span of the column vectors of $A$. Because the columns of $A$ often contain redundancy (linear dependence), we must isolate the pivot columns.

Not the most exciting part, but easily the most useful Worth keeping that in mind..

The Algorithm:

1. Row reduce matrix $A$ to its Reduced Row Echelon Form (RREF), denoted $R$. (Row Echelon Form is sufficient, but RREF makes pivot identification unambiguous).
2. Identify the pivot columns in $R$. These are columns containing leading 1s.
3. Select the corresponding columns from the original matrix $A$. These columns form a basis for $\text{Col } A$.

Why this works: Row operations change the column space (the actual vectors change), but they preserve the linear dependence relations among the columns. If columns 1 and 3 are pivot columns in $R$, then columns 1 and 3 of $A$ are linearly independent and span the column space of $A$.

Example: Let $A = \begin{bmatrix} 1 & 2 & 3 \ 2 & 4 & 6 \ 1 & 2 & 1 \end{bmatrix}$. Row reducing yields $R = \begin{bmatrix} 1 & 2 & 0 \ 0 & 0 & 1 \ 0 & 0 & 0 \end{bmatrix}$. Pivot columns in $R$ are column 1 and column 3. Basis for Col A: $\left{ \begin{bmatrix} 1 \ 2 \ 1 \end{bmatrix}, \begin{bmatrix} 3 \ 6 \ 1 \end{bmatrix} \right}$. Note that the second column of $A$ (which is $2 \times \text{col}_1$) was dependent and correctly excluded That's the part that actually makes a difference..

Scenario 2: Basis for the Null Space (Nul A)

The null space of $A$ is the solution set of the homogeneous equation $A\mathbf{x} = \mathbf{0}$. Finding a basis here involves expressing the solution set in parametric vector form.

The Algorithm:

1. Row reduce the augmented matrix $[A \mid \mathbf{0}]$ to RREF.
2. Identify pivot variables (basic variables) and free variables.
3. Write the basic variables in terms of the free variables.
4. Express the solution vector $\mathbf{x}$ as a linear combination of vectors, where the coefficients are the free variables.
5. The vectors attached to the free variables form the basis for $\text{Nul } A$.

Example: Using the RREF from above: $R = \begin{bmatrix} 1 & 2 & 0 \ 0 & 0 & 1 \ 0 & 0 & 0 \end{bmatrix}$. Equations: $x_1 + 2x_2 = 0 \Rightarrow x_1 = -2x_2$; $x_3 = 0$. $x_2$ is free. $\mathbf{x} = \begin{bmatrix} -2x_2 \ x_2 \ 0 \end{bmatrix} = x_2 \begin{bmatrix} -2 \ 1 \ 0 \end{bmatrix}$. Basis for Nul A: $\left{ \begin{bmatrix} -2 \ 1 \ 0 \end{bmatrix} \right}$. The dimension of the null space (nullity) is the number of free variables.

Scenario 3: Basis for a Span of Given Vectors

Often, a subspace is defined explicitly as $W = \text{Span}{\mathbf{v}_1, \mathbf{v}_2, \dots, \mathbf{v}_p}$. To find a basis, you must discard dependent vectors Small thing, real impact..

Method A: Row Space Technique (Standard)

1. Make the vectors $\mathbf{v}_i$ the rows of a matrix $B$.
2. Row reduce $B$ to RREF.
3. The non-zero rows of the RREF form a basis for the row space of $B$, which is exactly $\text{Span}{\mathbf{v}_1, \dots, \mathbf{v}_p}$.

Method B: Column Space Technique (Preserves Original Vectors) If you prefer a basis consisting of vectors from the original set:

1. Make the vectors $\mathbf{v}_i$ the columns of a matrix $C$.
2. Row reduce $C$ to RREF.
3. Identify pivot columns in the RREF.
4. Select the corresponding columns from the original matrix $C$ (the original vectors).

Comparison: Method A produces vectors that are often "cleaner" (containing leading 1s and zeros), making them easier to work with for further calculations like orthogonal projections. Method B preserves the original data vectors, which might be necessary for interpretation in applied contexts Still holds up..

Scenario 4: Basis for the Row Space (Row A)

The row space of a matrix $A$ is the span of its row vectors. Row operations do not change the row space.

The Algorithm:

1. Row reduce $A$ to Row Echelon Form (REF) or RREF.
2. The non-zero rows of the echelon form constitute a basis for $\text{Row } A$.

This is computationally the fastest basis to find. If $A$ reduces to a matrix with $r$ non-zero rows, then $\dim(\text{Row } A) = r = \text{rank}(A)$ That's the part that actually makes a difference..

Scenario 5: Basis for a Subspace Defined by Equations

Subspaces are frequently defined by linear constraints, such as: $W = {(x, y, z) \in \mathbb{R}^3 \mid x - 2y + 3z = 0 \text{ and } 2x + y - z = 0}$.

This is essentially finding the null

space of a coefficient matrix Simple, but easy to overlook. Less friction, more output..

Method:

1. Construct the coefficient matrix $D$ where each row represents one linear equation. For the example: $D = \begin{bmatrix} 1 & -2 & 3 \ 2 & 1 & -1 \end{bmatrix}$

2. Row reduce $D$ to RREF: $RREF(D) = \begin{bmatrix} 1 & 0 & \frac{7}{5} \ 0 & 1 & \frac{1}{5} \end{bmatrix}$

3. The system becomes: $x = -\frac{7}{5}z$ $y = -\frac{1}{5}z$

4. Let $z = 5t$ (choosing 5 eliminates fractions). Then: $x = -7t$ $y = -t$ $z = 5t$

5. Express the general solution as: $\mathbf{w} = \begin{bmatrix} -7t \ -t \ 5t \end{bmatrix} = t\begin{bmatrix} -7 \ -1 \ 5 \end{bmatrix}$

Basis for W: $\left{ \begin{bmatrix} -7 \ -1 \ 5 \end{bmatrix} \right}$

The dimension equals the number of free variables, which is $n - r$ where $n$ is the number of columns and $r$ is the rank.

Summary Table

Conclusion

Finding bases for fundamental subspaces is a systematic process that relies on understanding the relationship between solutions, pivot positions, and free variables. By identifying pivot columns in a row-reduced matrix, we can determine which original vectors form a basis, and by treating free variables as parameters, we can express the entire solution space as a span of basis vectors. In real terms, the key insight is that row operations preserve the row space while revealing the structure of the null space. This unified approach—whether working with null spaces, row spaces, or spans of given vectors—demonstrates the elegant interconnectedness of linear algebra, where the same computational tools reveal the underlying structure of vector spaces across different contexts.