Linear Algebra › Section D · Vector spaces › Topic D.3
Linearly independent sets and bases
A basis is a spanning set with no waste: every vector it builds is built in only one way. This topic carries linear independence into general vector spaces and shows how to find bases for column spaces and null spaces.
Spanning without waste
A subspace can usually be described by many different spanning sets, and some of them contain redundant vectors. If one vector is already a combination of the others, it adds nothing to the span; it only makes descriptions longer and representations non-unique. A basis is a spanning set from which nothing can be removed. To define it in any vector space, first carry over the idea of linear independence from "Span and linear independence", where it was defined for vectors in \(\mathbb{R}^n\). The definition is word for word the same; only the vectors change.
Linear independence in any vector space
An indexed set \(\{\mathbf{v}_1, \ldots, \mathbf{v}_p\}\) in a vector space \(V\) is linearly independent if the vector equation
has only the trivial solution \(c_1 = \cdots = c_p = 0\). Otherwise it is linearly dependent, and a non-trivial solution is a linear dependence relation. Familiar facts still hold. A set containing \(\mathbf{0}\) is dependent. Two vectors are dependent exactly when one is a multiple of the other. And a set of two or more vectors with \(\mathbf{v}_1 \ne \mathbf{0}\) is dependent exactly when some \(\mathbf{v}_j\) with \(j > 1\) is a combination of the vectors before it.
Independence among functions
In function spaces, \(\mathbf{0}\) is the zero function, so a dependence relation must hold for every value of \(t\). The functions \(\sin t\) and \(\cos t\) are independent: neither is a constant multiple of the other. At \(t = 0\), \(\cos t = c\sin t\) would give \(1 = 0\), and \(\sin t = c\cos t\) would force \(c = 0\), which fails at \(t = \pi/2\). By contrast, \(\{1, \cos 2t, \sin^2 t\}\) is dependent, because the identity \(\cos 2t = 1 - 2\sin^2 t\) gives the relation
for every \(t\). Similarly, the polynomials \(1, t, t^2, \ldots, t^n\) are independent, because a polynomial equal to zero for every \(t\) must have all its coefficients equal to zero.
What a basis is
Let \(H\) be a subspace of a vector space \(V\). An indexed set \(\mathcal{B} = \{\mathbf{b}_1, \ldots, \mathbf{b}_p\}\) in \(V\) is a basis for \(H\) if 1. \(\mathcal{B}\) is linearly independent, and 2. \(\mathcal{B}\) spans \(H\). Standard examples: the columns \(\mathbf{e}_1, \ldots, \mathbf{e}_n\) of \(I_n\) form the standard basis of \(\mathbb{R}^n\); \(\{1, t, t^2, \ldots, t^n\}\) is the standard basis of \(\mathbb{P}_n\); and by "The invertible matrix theorem", the columns of any invertible n x n matrix form a basis of \(\mathbb{R}^n\). A space has many bases. The set \(\{(1, 1), (1, -1)\}\) is another basis of \(\mathbb{R}^2\).
The spanning set theorem
Suppose \(S = \{\mathbf{v}_1, \ldots, \mathbf{v}_p\}\) spans a subspace \(H\). 1. If some \(\mathbf{v}_k\) is a linear combination of the other vectors in \(S\), removing it leaves a set that still spans \(H\): wherever \(\mathbf{v}_k\) appeared in a combination, substitute its expression in terms of the others. 2. If \(H \ne \{\mathbf{0}\}\), some subset of \(S\) is a basis for \(H\). Keep removing vectors that depend on the rest; the process stops at an independent set that still spans, because a single non-zero vector is always independent. So every finite spanning set can be trimmed down to a basis.
Worked example: a basis for Col A
Let
1. Row reduce: the reduced echelon form has rows (1, 2, 0, 3), (0, 0, 1, 2), (0, 0, 0, 0), with pivots in columns 1 and 3. 2. Read the column relations from it: column 2 is 2 times column 1, and column 4 is 3 times column 1 plus 2 times column 3. 3. Row operations do not change the solutions of \(A\mathbf{x} = \mathbf{0}\), so the columns of \(A\) satisfy the same relations. Check:
4. Remove the dependent columns. A basis for \(\operatorname{Col} A\) is \(\{(1, 2, -1), (0, -1, 3)\}\): the pivot columns of \(A\) itself.
Pivot columns of A, not of its echelon form
The rule is: the pivot columns of \(A\) form a basis for \(\operatorname{Col} A\). The echelon form only tells you which columns are pivot columns. Its own columns usually span a different space. In the example, every column of the reduced echelon form has third entry 0, so their span is the plane \(x_3 = 0\). But \(\operatorname{Col} A\) contains \((1, 2, -1)\), whose third entry is not 0. Row operations keep the dependence relations among the columns, but they move the columns themselves, so always go back to the original matrix for the basis vectors.
Worked example: a basis for Nul A
Use the same matrix \(A\). 1. From the reduced echelon form, the free variables are \(x_2\) and \(x_4\), with \(x_1 = -2x_2 - 3x_4\) and \(x_3 = -2x_4\). 2. The parametric vector form is
so these two vectors span \(\operatorname{Nul} A\). 3. They are independent: in entry 2 the first vector has 1 and the second 0, while in entry 4 the first has 0 and the second 1. So any combination equal to \(\mathbf{0}\) has both weights 0. 4. Hence \(\{(-2, 1, 0, 0), (-3, 0, -2, 1)\}\) is a basis for \(\operatorname{Nul} A\). This method always produces an independent set, one vector per free variable.
Two views of a basis
A basis is a spanning set that is as small as possible: delete any vector and it no longer spans, because the deleted vector is not a combination of the others. A basis is also an independent set that is as large as possible: add any other vector of the space and the set becomes dependent, because that vector is already a combination of the basis. For instance, \(\{(1, 0, 0), (0, 1, 0)\}\) is independent but too small to span \(\mathbb{R}^3\), while adding \((2, 5, 0)\) to it gives a set that is still not spanning but is now dependent. Only a set that is both independent and spanning is a basis.
Common mistakes to avoid
Taking the pivot columns of the echelon form instead of those of \(A\) gives a wrong basis for \(\operatorname{Col} A\). Including every column of \(A\) gives a spanning set, not a basis, unless the columns are independent. Mixing up the two spaces is another slip: the null space basis comes from the free variables, the column space basis from the pivot columns. For functions, a relation must hold for all \(t\), not just at one point. Finally, "independent" alone is not "basis": the set must also span.
Key takeaways
A set is linearly independent when the only combination giving \(\mathbf{0}\) is the trivial one; this works in any vector space, including function and polynomial spaces. A basis for \(H\) is an independent set that spans \(H\). Removing a vector that depends on the others keeps the span, so any finite spanning set of a non-zero subspace contains a basis. The pivot columns of \(A\), taken from \(A\) itself, form a basis for \(\operatorname{Col} A\). The vectors from the parametric vector form of \(A\mathbf{x} = \mathbf{0}\) form a basis for \(\operatorname{Nul} A\).
Flashcards
Flashcard
c₁v₁ + … + cₚvₚ = 0 only when every cᵢ = 0
Test your understanding
Question 1 of 5
A matrix A has reduced echelon form with pivots in columns 1 and 3. Which set is a basis for Col A?
Question 2 of 5
Which set of functions is linearly dependent?
Question 3 of 5
Ax = 0 has general solution x = x₂(-2, 1, 0, 0) + x₄(-3, 0, -2, 1). What is a basis for Nul A?
Question 4 of 5
S = {v₁, v₂, v₃} spans H, and v₃ = 2v₁ - v₂. What does the spanning set theorem say?
Question 5 of 5
Is {(1, 0, 1), (2, 1, 0), (0, 1, -2)} a basis of R³?
