General vector space

Beyond Two Dimensions

Section 4.1: Vector Spaces and Subspaces

Proposition 4.1.7

Let V be a vector space.

  1. For all vV, we have 0+v=v
  2. For all v,wV, if v+w=0, then w+v=0
Proposition 4.1.8

Suppose that V is a vector space, and let u,v,wV.

  1. If v+u=w+u, then v=w
  2. If u+v=u+w, then v=w
Proposition 4.1.9

Let V be a vector space.

  1. If zV and v+z=v for all vV, then z=0, i.e. 0 is the only additive identity
  2. For all vV, there is a unique wV with v+w=0, i.e. additive inverses are unique
Proposition 4.1.11

Let V be a vector space.
3. 0v=0 for all vV
4. c0=0 for all cR
5. (1)v=v for all vV

Proposition 4.1.16

Let V be a vector space and let u1,u2,,unV. The set Span(u1,u2,,un) is a subspace of V

Section 4.2: Solving Linear Systems

In essence, Section 4.1 defines the abstract playing field (Vector Spaces) and important regions within it (Subspaces). Section 4.2 provides the essential computational toolkit (Gaussian Elimination on matrices) for answering concrete questions that arise when working within these abstract structures.

Proposition 4.2.4 (Elementary Row Operations are Reversible)

Suppose that we have a linear system, and we apply one elementary row operation to obtain a new system. We can then apply one elementary row operation to the new system in order to recover the old system.

Corollary 4.2.5

Suppose that we have a linear system A1, and we apply one elementary row operation to obtain a new system A2. If S1 is the solution set to the original system, and S2 is the solution set to the new system, then S1=S2.

Proposition 4.2.11

For any matrix A, there exists an echelon form of A.

Proposition 4.2.12

Suppose that we have a linear system with augmented matrix A, and that B is an echelon form of A.

  1. If the last column of B contains a leading entry, then the system is inconsistent
  2. If the last column of B contains no leading entry, but every other column of B has a leading entry, then the system is consistent and has a unique solution
  3. If the last column of B contains no leading entry, and there is at least one other column of B without a leading entry, then the system is consistent and has infinitely many solutions. Moreover, for each choice of values for the variables that do not correspond to leading entries, there is a unique solution for the system taking these values.
Proposition 4.2.15

Let a0,a1,,anR and b0,b1,,bnR. Let f:RR and g:RR be the polynomial functions: f(x)=anxn+an1xn1++a1x+a0 and g(x)=bnxn+bn1xn1++b1x+b0. If f(x)=g(x) for all xR, then ai=bi for all i.

Section 4.3: Spanning Sequences

Proposition 4.3.1

Let u1,u2,,unRm. Let A be the m×n matrix where the ith column is ui, and let B be an echelon form of A. We then have that Span(u1,u2,,un)=Rm if and only if every row of B has a leading entry.

Corollary 4.3.2

If u1,u2,,unRm and Span(u1,u2,,un)=Rm, then nm.

Corollary 4.3.3

If u1,u2,,unRm and n<m, then Span(u1,u2,,un)Rm.

Section 4.4: Linear Independence

Core Question of Section 4.4:

Section 4.3 asked: "Can we reach everything in the space?" Section 4.4 tackles the complementary question: "Do we have redundancy in our list of vectors?" In other words, is there a vector in our sequence (u1,u2,,un) that is "unnecessary" because it can already be created by combining the others?

The Initial Idea of Redundancy:

The most direct way to think about redundancy is to ask if any vector uk in the sequence is a linear combination of the other vectors in the sequence, i.e., is ukSpan(u1,,uk1,uk+1,,un)?

The Problem with the Initial Idea:

While intuitive, checking this directly is computationally inefficient. For a sequence of n vectors, you would potentially need to set up and solve n different linear systems (one for each vector uk to see if it's in the span of the others).

The Rationale - A More Elegant Perspective:

  1. Rewriting Redundancy: Suppose a vector uk is redundant. This means we can write uk=c1u1++ck1uk1+ck+1uk+1++cnun.
  2. The Homogeneous Equation: By simply moving uk to the other side, we get:
    c1u1++ck1uk11uk+ck+1uk+1++cnun=0.
    Notice this is a linear combination of all the vectors (u1,,un) that equals the zero vector 0. Crucially, at least one coefficient (the 1 in front of uk) is nonzero.
  3. Linear Dependence: This motivates the definition. A sequence (u1,,un) is called linearly dependent if there exist scalars c1,,cn, not all zero, such that c1u1+c2u2++cnun=0. This condition signals redundancy.
  4. Linear Independence: Conversely, the sequence is linearly independent if the only way to make the linear combination c1u1+c2u2++cnun=0 is by choosing the trivial solution where c1=c2==cn=0. This means there's no redundancy – no vector can be expressed in terms of the others (Proposition 4.4.2 proves this equivalence formally).

The Rationale - Connecting Linear Independence to Linear Systems:

How do we check if the only solution to c1u1+c2u2++cnun=0 is the trivial one c1==cn=0?

  1. Homogeneous System: This is equivalent to asking if the homogeneous linear system Ac=0 has only the trivial solution c=0, where A is the matrix whose columns are u1,,un.
  2. Gaussian Elimination Again: We use Gaussian elimination on the coefficient matrix A (we can ignore the augmented column of zeros as it never changes). Let B be an echelon form of A.
  3. Condition for Trivial Solution Only: The homogeneous system Ac=0 (which has the same solution set as Bc=0) has only the trivial solution if and only if there are no free variables.
  4. The Key Insight (Proposition 4.4.5): This happens exactly when the echelon form B has a leading entry (pivot) in every column. If there's a column without a pivot, that corresponds to a free variable, allowing for non-trivial solutions (meaning linear dependence). If every column has a pivot, there are no free variables, forcing the only solution to be the trivial one (meaning linear independence).

Rationale behind the Corollaries:

In summary, Section 4.4 defines linear independence as the absence of non-trivial relationships between vectors (meaning no redundancy). It connects this concept to checking if a homogeneous system Ac=0 has only the trivial solution. Gaussian elimination provides the tool (checking for a pivot in every column of the echelon form) to answer this efficiently. The corollaries establish fundamental limits on the number of linearly independent vectors based on the dimension of the ambient space.

Proposition 4.4.2

Let V be a vector space, and let u1,u2,,unV where n2. The following are equivalent:

  1. (u1,u2,,un) is linearly independent
  2. There does not exist an i such that uiSpan(u1,,ui1,ui+1,,un)
Proposition 4.4.3

Let V be a vector space, and let uV. We then have that (u) is linearly independent if and only if u0.

Proposition 4.4.5

Let u1,u2,,unRm. Let A be the m×n matrix where the ith column is ui, and let B be an echelon form of A. We then have that (u1,u2,,un) is linearly independent if and only if every column of B has a leading entry.

Corollary 4.4.6

If u1,u2,,unRm and (u1,u2,,un) is linearly independent, then nm.

Corollary 4.4.7

If u1,u2,,unRm and n>m, then (u1,u2,,un) is linearly dependent.

Section 4.5: Bases and Coordinates

Core Idea of Section 4.5: Finding the "Right" Set of Vectors

The goal is to identify the most "efficient" set of vectors to describe a vector space V. What does efficient mean here?

  1. Sufficiency (Spanning): The set must be large enough so that linear combinations of its vectors can create every vector in the space V. This is the idea of a spanning sequence from Section 4.3. We need Span(u1,,un)=V.
  2. No Redundancy (Linear Independence): The set should be "minimal" in the sense that it doesn't contain unnecessary vectors. If we can create one vector in the set using the others, it's redundant. This is precisely the idea captured by linear independence from Section 4.4. We need (u1,,un) to be linearly independent.

Rationale for "Basis" (Definition 4.5.1)

A basis is defined as a sequence of vectors (u1,,un) that satisfies both of the above conditions:

  1. It spans the entire space V.
  2. It is linearly independent.

Rationale for Theorem 4.5.2 (Basis Unique Representation)

This theorem provides the crucial insight into why bases are so fundamental:

Rationale for "Coordinates" (Definition 4.5.3)

Since Theorem 4.5.2 guarantees a unique list of scalars c1,,cn for each v relative to a basis α=(u1,,un), we give this list a name:

Rationale for Propositions 4.5.5 and 4.5.6 (Constructing Bases)

These results provide ways to find a basis:

In short, Section 4.5 defines the crucial concept of a basis as the optimal blend of spanning and linear independence. The core rationale is that a basis allows every vector in the space to be represented uniquely by a set of coordinates, effectively translating abstract vector space problems into the concrete setting of Rn. The section also provides methods for constructing a basis from a spanning set.

Theorem 4.5.2

Let V be a vector space and let α=(u1,u2,,un) be a sequence of elements of V. We then have that α is a basis for V if and only if every vV can be expressed as a linear combination of u1,u2,,un in a unique way.

Proposition 4.5.4

Let V be a vector space and let u1,u2,,un,wV. The following are equivalent:

  1. Span(u1,u2,,un,w)=Span(u1,u2,,un)
  2. wSpan(u1,u2,,un)
Proposition 4.5.5

Let V be a vector space, and let u1,u2,,unV be such that Span(u1,u2,,un)=V. There exists a basis of V that can be obtained by omitting some of the ui from the sequence.

Proposition 4.5.6

Let u1,u2,,unRm, and let W=Span(u1,u2,,un). Let A be the m×n matrix whose ith column is ui, and let B be an echelon form of A. If we build the sequence consisting only of those ui such that the ith column of B has a leading entry, then we obtain a basis for W.

Section 4.6: Dimension

Core Idea of Section 4.6: Defining the "Size" of a Vector Space

After establishing bases (Section 4.5) as "efficient" sets that both span the space and have no redundancy (linear independence), a natural question arises: Does every basis for a given vector space V have the same number of elements? Section 4.6 answers this question affirmatively and uses it to define the dimension of the space.

The Rationale - Why Dimension is Well-Defined:

  1. The Problem: Vector spaces can have many different bases. For the concept of "dimension" to be meaningful as an intrinsic property of the space itself, we need to know that the number of vectors in a basis doesn't depend on which basis we happen to choose.
  2. Key Result (Theorem 4.6.2): This theorem is the cornerstone. It states that if a vector space V can be spanned by n vectors (u1,,un), then any sequence (w1,,wm) with more than n vectors (i.e., m>n) must be linearly dependent.
    • Rationale behind Thm 4.6.2: The proof uses the Steinitz Exchange Lemma (Proposition 4.6.1) repeatedly. The idea is that you can systematically replace vectors in the spanning set (ui) with vectors from the linearly independent set (wj) while maintaining the span. Since you start with only n vectors spanning the space, you can swap in at most n of the wj's. If you have more than n vectors in the w sequence (m>n), then by the time you consider wn+1, it must be expressible as a linear combination of the preceding w1,,wn (which now span the space), proving the sequence (w1,,wm) is linearly dependent.
  3. Consequence (Corollary 4.6.3): Taking the contrapositive of Theorem 4.6.2, we get: If a space V is spanned by (u1,,un) and (w1,,wm) is linearly independent in V, then mn. In words: the size of any linearly independent set is less than or equal to the size of any spanning set.
  4. Invariance of Basis Size (Corollary 4.6.4): This directly uses the previous corollary. If you have two bases, α with n vectors and β with m vectors:
    • Since α spans and β is linearly independent, mn.
    • Since β spans and α is linearly independent, nm.
    • Therefore, m=n.
  5. Definition of Dimension (Definition 4.6.5): Because Corollary 4.6.4 guarantees that all bases for a given vector space V (if any exist) have the same number of elements, we can unambiguously define the dimension of V, denoted dim(V), as this common number.
    • Rationale: Dimension captures the intrinsic "size" or number of "degrees of freedom" or independent "directions" within the vector space, regardless of the specific basis chosen to measure it.

The Rationale - Consequences and Uses of Dimension:

Once dimension is well-defined, it becomes a powerful tool:

  1. Basis Shortcut (Proposition 4.6.7): If you know dim(V)=n, then checking if a sequence of exactly n vectors (u1,,un) is a basis becomes easier. You only need to check one of the two conditions:
    • If it's linearly independent, it must also span V.
    • If it spans V, it must also be linearly independent.
    • Rationale: This follows from Theorem 4.6.2 and its corollary. For instance, if you have n linearly independent vectors, but they didn't span V, you could add another vector wSpan(u1,,un) to get n+1 linearly independent vectors (Prop 4.6.9), which contradicts the fact that the n-dimensional space V cannot contain n+1 linearly independent vectors (Cor 4.6.3). A similar argument works the other way.
  2. Subspace Dimension (Proposition 4.6.8): Any subspace W of a finite-dimensional vector space V is also finite-dimensional, and dim(W)dim(V).
    • Rationale: You can construct a basis for W by starting with a linearly independent set in W and extending it (using Prop 4.6.10, below). This process must stop because W is contained in V, and you can't have more linearly independent vectors in W than the dimension of V. This guarantees W has a finite basis, and the number of vectors must be less than or equal to dim(V).
  3. Extending Linearly Independent Sets (Proposition 4.6.10): Any linearly independent sequence in a finite-dimensional vector space can be extended to form a basis for that space.
    • Rationale: Start with the linearly independent set. If it doesn't already span the whole space, find a vector not in its span and add it to the set; the new set remains linearly independent (Prop 4.6.9). Repeat this process. Since the dimension is finite, you can't keep adding vectors indefinitely while maintaining linear independence (by Thm 4.6.2). The process must stop when the set spans the space, at which point it is a basis. This confirms that linearly independent sets act as valid starting points or "skeletons" for building bases.

In summary, Section 4.6 leverages the interplay between spanning sets and linearly independent sets (specifically Theorem 4.6.2) to rigorously establish that the size of a basis is an invariant property of a vector space, allowing the definition of dimension. Dimension then becomes a key characteristic used to understand the structure of the space, relationships between spaces and subspaces, and provides useful criteria for determining if a set of vectors forms a basis.

Proposition 4.6.1 (Steinitz Exchange)

Let V be a vector space and let u1,u2,,un,wV. Suppose that we have d1,,dnR with w=d1u1++dkuk++dnun. If dk0, then Span(u1,,uk,,un)=Span(u1,,uk1,w,uk+1,,un).

Theorem 4.6.2

Let V be a vector space and let u1,,un,w1,,wmV. If m>n and Span(u1,,un)=V, then (w1,,wm) is linearly dependent.

Corollary 4.6.3

Let V be a vector space and let u1,,un,w1,,wmV be such that both Span(u1,,un)=V and (w1,,wm) is linearly independent. We then have mn.

Corollary 4.6.4

Suppose that V is a vector space and both (u1,,un) and (w1,,wm) are bases for V. We then have that m=n.

Proposition 4.6.7

Let V be a finite-dimensional vector space and let n=dim(V). Let u1,u2,,unV.

  1. If (u1,u2,,un) is linearly independent, then (u1,u2,,un) is a basis of V
  2. If Span(u1,u2,,un)=V, then (u1,u2,,un) is a basis of V
Proposition 4.6.8

Let V be a finite-dimensional vector space with dim(V)=n, and let W be a subspace of V. There exists k{0,1,,n} and w1,,wkW with W=Span(w1,,wk) and such that (w1,,wk) is linearly independent. In particular, W is a finite-dimensional vector space and dim(W)dim(V).

Proposition 4.6.9

Let V be a vector space, let (u1,,un) be a linearly independent sequence of vectors in V, and let wV. The following are equivalent:

  1. (u1,,un,w) is linearly independent
  2. wSpan(u1,,un)
Proposition 4.6.10

Let V be a finite-dimensional vector space, and let (u1,,uk) be a linearly independent sequence of vectors in V. There exists w1,,wmV (possibly with m=0) such that (u1,,uk,w1,,wm) is a basis of V.