Bases

How many numbers do you need to pin down a point in space? In the plane, two coordinates suffice. In ordinary space, three are needed. This article makes that intuition precise by defining a basis. A basis is a list of vectors that is simultaneously efficient and complete. Efficient means no redundancy, i.e. linearly independent. Complete means it reaches every vector, i.e. it spans. The two ideas of Linear Independence and Span and Linear Combinations are each useful separately. Here they combine into the single most important structural concept in linear algebra. Bases matter because they give every vector space a coordinate system. Once a basis is fixed, every vector corresponds to exactly one list of scalars. This list is called its coordinates. Linear algebra then becomes computable. Matrices, systems of equations, and geometric transformations all secretly refer back to a choice of basis. Without a basis theorem guaranteeing existence and uniqueness of representation, none of this machinery would be justified. The big picture is that a basis is a minimal spanning set. It is simultaneously a maximal independent set. These turn out to be the same notion. This is itself a small miracle proved below. This equivalence is what lets us later prove something important. Any two bases of a finite-dimensional space have the same size. This gives rise to the notion of dimension. This article delivers a rigorous definition of basis. It presents the fundamental theorem that basis vectors give unique representations. It includes full proofs and worked examples with genuinely asymmetric numbers. It offers a counterexample showing why independence cannot be dropped. Finally, it maps where bases show up throughout mathematics and its applications.

Key takeaways

A basis is a list of vectors that is linearly independent and spans the vector space.
Equivalently, a basis is a list from which every vector can be written as a linear combination in exactly one way.
Every finite-dimensional vector space has a basis, and (proved elsewhere) all bases of the same space have the same length.
A basis turns abstract vectors into concrete coordinate lists, enabling matrix computations.
Dropping independence or dropping spanning each breaks the uniqueness-of-representation property.
Key idea: a basis is exactly a spanning list with no redundancy — independent enough to give uniqueness, big enough to give existence.

Where this lives today

Bases are the computational bridge between abstract vector spaces and concrete arithmetic with matrices: choosing a basis converts every linear map into a matrix and every vector into a coordinate column, which is why matrix algebra works at all. In numerical linear algebra, changing between bases (orthonormal bases especially) underlies algorithms like QR decomposition, least-squares fitting, and eigenvalue computation. In physics, choosing a basis of energy eigenstates in quantum mechanics turns an abstract Hilbert-space problem into a concrete matrix (Heisenberg) formulation; changes of basis correspond to changes of observer frame or representation. In signal processing and data compression (JPEG, MP3, wavelets), signals are expressed in bases like the Fourier or wavelet basis chosen so that most coordinates are near zero, enabling compression. In machine learning and statistics, principal component analysis constructs a new orthonormal basis aligned with directions of maximal variance, re-expressing data in coordinates that are more informative than the raw ones. In coding theory and cryptography, finite vector spaces over finite fields use bases to construct error-correcting codes and lattice-based cryptographic schemes, where the choice of "reduced" basis (short, nearly orthogonal vectors) is itself a deep algorithmic problem (e.g. the LLL algorithm).

Ask the tutor: If I have two different bases of the same space, how exactly do I convert one vector's coordinates from one basis to the other?

V

Intuition

A spanning list is enough material to build every vector in $V$ ; an independent list has no wasted material — no vector in the list is a "clone" expressible via the others. A basis asks for both at once: exactly enough material, with nothing to spare. Think of $\mathbb{R}^2$ . The list $\big((1,0),(0,1),(1,1)\big)$ spans $\mathbb{R}^2$ but is not independent — the third vector is the sum of the first two, so it's redundant weight. The list $\big((1,0)\big)$ is independent but does not span — it can only ever reach the $x$ -axis. The list $\big((1,0),(0,1)\big)$ threads the needle: independent and spanning, so it is a basis. So is the "tilted" list $\big((2,1),(1,1)\big)$ , showing bases are far from unique — but every vector still gets exactly one address relative to whichever basis you pick.

Figure

Two different bases of the plane give two different — but each unique — coordinate grids for the same point (3,2)

The central theorem

V

\beta = (v_1,\dots,v_n)

V

Worked examples

Example 1 — a genuinely tilted basis of $\mathbb{R}^3$ . Let $v_1 = (1,2,0)$ , $v_2 = (0,1,3)$ , $v_3 = (2,0,1)$ . We show $\beta = (v_1,v_2,v_3)$ is a basis of $\mathbb{R}^3$ and find the unique coordinates of $w = (5,1,7)$ . Independence. Suppose $a_1v_1 + a_2v_2 + a_3v_3 = 0$ . Componentwise:

a_1 + 2a_3 = 0,\qquad 2a_1 + a_2 = 0, \qquad 3a_2 + a_3 = 0.

From the first equation, $a_1 = -2a_3$ . Substituting into the second, $-4a_3 + a_2 = 0$ , so $a_2 = 4a_3$ . Substituting into the third, $12a_3 + a_3 = 13a_3 = 0$ , so $a_3 = 0$ . Then $a_2 = 0$ and $a_1 = 0$ . The only solution is trivial, so $\beta$ is linearly independent. Spanning. Since $\beta$ is an independent list of 3 vectors in $\mathbb{R}^3$ , and (as will be justified by the exchange lemma / dimension theory) any independent list of length equal to $\dim V$ automatically spans $V$ , $\beta$ spans $\mathbb{R}^3$ . To keep this example self-contained without invoking that theorem, we verify spanning directly by solving for arbitrary $(x,y,z)$ :

a_1 + 2a_3 = x,\qquad 2a_1 + a_2 = y,\qquad 3a_2 + a_3 = z.

Solve: from equation 1, $a_1 = x - 2a_3$ . Plug into equation 2: $2x - 4a_3 + a_2 = y \Rightarrow a_2 = y - 2x + 4a_3$ . Plug into equation 3:

3(y - 2x + 4a_3) + a_3 = z \;\Rightarrow\; 3y - 6x + 13a_3 = z \;\Rightarrow\; a_3 = \frac{z - 3y + 6x}{13}.

This determines $a_3$ (hence $a_1, a_2$ ) for every $(x,y,z)$ , so a solution always exists: $\beta$ spans $\mathbb{R}^3$ . Combined with independence, $\beta$ is a basis. Coordinates of $w=(5,1,7)$ . Using the formula above with $x=5, y=1, z=7$ :

a_3 = \frac{7 - 3(1) + 6(5)}{13} = \frac{7 - 3 + 30}{13} = \frac{34}{13}.

Then

a_2 = 1 - 2(5) + 4\cdot\frac{34}{13} = 1 - 10 + \frac{136}{13} = -9 + \frac{136}{13} = \frac{-117+136}{13} = \frac{19}{13},

a_1 = 5 - 2\cdot\frac{34}{13} = 5 - \frac{68}{13} = \frac{65-68}{13} = -\frac{3}{13}.

Check:

-\tfrac{3}{13}(1,2,0) + \tfrac{19}{13}(0,1,3) + \tfrac{34}{13}(2,0,1) = \left(\tfrac{-3+68}{13},\ \tfrac{-6+19}{13},\ \tfrac{57+34}{13}\right) = \left(\tfrac{65}{13}, \tfrac{13}{13}, \tfrac{91}{13}\right) = (5,1,7).

It checks out exactly. By the Unique Representation theorem, these are the only scalars representing $w$ in this basis — no other triple works.

Example 2 — polynomials: a non-monomial basis of $\mathcal{P}_2(\mathbb{R})$ . Let $V = \mathcal{P}_2(\mathbb{R})$ , polynomials of degree at most 2, with the standard basis $(1, x, x^2)$ well known to work. Consider instead

q_1 = 1 + x, \qquad q_2 = x + x^2, \qquad q_3 = 1 + x^2.

We check $\gamma = (q_1, q_2, q_3)$ is a basis and express $p(x) = 4 + 3x - x^2$ in it. Independence. Suppose $c_1 q_1 + c_2 q_2 + c_3 q_3 = 0$ (the zero polynomial). Expanding:

c_1(1+x) + c_2(x+x^2) + c_3(1+x^2) = (c_1+c_3) + (c_1+c_2)x + (c_2+c_3)x^2.

Matching coefficients of $1, x, x^2$ to zero gives the system

c_1 + c_3 = 0, \qquad c_1 + c_2 = 0, \qquad c_2 + c_3 = 0.

From the first, $c_3 = -c_1$ . From the second, $c_2 = -c_1$ . Substitute into the third: $-c_1 + (-c_1) = -2c_1 = 0 \Rightarrow c_1 = 0$ , hence $c_2 = c_3 = 0$ . Independent. Spanning / coordinates of $p$ . We solve $c_1 q_1 + c_2 q_2 + c_3 q_3 = p = 4 + 3x - x^2$ directly:

c_1 + c_3 = 4, \qquad c_1 + c_2 = 3, \qquad c_2 + c_3 = -1.

Add all three equations: $2(c_1+c_2+c_3) = 4 + 3 + (-1) = 6$ , so $c_1+c_2+c_3 = 3$ . Subtracting each original equation from this sum:

c_2 = 3 - 4 = -1, \qquad c_3 = 3 - 3 = 0, \qquad c_1 = 3 - (-1) = 4.

Check: $c_1 q_1 + c_2 q_2 + c_3 q_3 = 4(1+x) + (-1)(x+x^2) + 0(1+x^2) = 4 + 4x - x - x^2 = 4 + 3x - x^2 = p(x)$ . ✓. Since this general solving procedure (a linear system with a unique solution for arbitrary right-hand side, as the same elimination argument works for any target coefficients) succeeds for every polynomial in $\mathcal P_2(\mathbb R)$ , $\gamma$ spans $V$ ; combined with independence, $\gamma$ is a basis, and $(4,-1,0)$ are the unique $\gamma$ -coordinates of $p$ .

V = \mathbb{R}^2

Scope, edge cases & common traps

Situation	What to watch for
The zero vector space $\{0\}$	Its basis is the empty list $()$ ; the empty sum is defined as $0$ , so the theorem still holds vacuously/trivially.
A list containing $0$	Any list including the zero vector is automatically dependent ( $1\cdot 0 = 0$ is a nontrivial-looking but valid dependency once other coefficients are $0$ ... more precisely $1\cdot 0 = 0$ already witnesses dependence by itself), so it can never be part of a basis.
Order of vectors in $\beta$	A basis is a list (ordered), because coordinates are read off in that order; permuting the list gives a different (though related) coordinate system, so "the same set" with different orders is not "the same basis" for coordinate purposes.
Infinite-dimensional spaces	The definition adapts via indexed families and finite linear combinations; existence of a basis there needs the Axiom of Choice (Hamel bases) — genuinely more delicate and outside finite-dimensional scope here.
Confusing "spans" with "is a basis"	A spanning list that is too long is not a basis — check independence separately; a common student error is to stop after confirming spanning.

The mathematician behind it

Origins · 1 minHermann Günther Grassmann and how this was discovered

Hermann Günther Grassmann

1809–1877 · 19th-century Germany

Grassmann, a schoolteacher and largely self-taught mathematician in Stettin, developed in his 1844 treatise Die lineale Ausdehnungslehre ("Theory of Linear Extension") a strikingly general theory of vector spaces of arbitrary dimension, including the ideas of linear independence, span, and basis in essentially modern algebraic form, decades before the language of "vector space" was standardized. His work was so far ahead of contemporary notation and so densely written that it went largely unrecognized in his lifetime; he later shifted much of his energy to linguistics and Sanskrit scholarship, where he received more immediate recognition. Only after his death did mathematicians such as Peano formalize and popularize the axiomatic vector space concept that made Grassmann's basis theory the standard foundation of linear algebra.

History

- 1844 — Grassmann publishes the Ausdehnungslehre, introducing

n

-dimensional linear extension, independence, and spanning sets in essentially modern form.

- 1862 — Grassmann publishes a revised, clearer edition, still under-appreciated by his contemporaries.

- 1888 — Giuseppe Peano gives the first fully modern axiomatic definition of a vector space, crediting and clarifying Grassmann's ideas.

- 1920s–1930s — Hermann Weyl and others popularize coordinate-free, basis-theoretic linear algebra in physics and mathematics.

- Mid-20th century — bases and dimension become standard undergraduate curriculum via textbooks such as Halmos's Finite-Dimensional Vector Spaces.

Is a basis unique for a given vector space?

No — almost every nonzero vector space has infinitely many different bases. For instance $\mathbb R^2$ has the standard basis $((1,0),(0,1))$ , but also $((2,1),(1,1))$ , $((1,3),(2,5))$ , and infinitely many others, as long as the chosen vectors are independent and span. What is guaranteed (and proved via the Steinitz Exchange Lemma) is that every basis of a given finite-dimensional vector space has the same number of vectors, called the Dimension of the space. So while the specific basis is never unique, the count of vectors in any basis is a genuine invariant of $V$ itself.

Does every vector space have a basis?

Every finite-dimensional vector space that is spanned by finitely many vectors has a basis: start with any finite spanning list and discard vectors one at a time whenever one is a linear combination of the others, which strictly shrinks the list while preserving spanning; this process must terminate at an independent spanning list. Infinite-dimensional spaces (such as the space of all real sequences) also have bases (Hamel bases), but proving this in general requires the Axiom of Choice and such bases are typically impossible to write down explicitly.

Why does the order of vectors in a basis matter?

A basis is formally defined as an ordered list, not just a set, because the coordinates of a vector are read off in a specific slot-by-slot order: the $i$ -th coordinate corresponds to the $i$ -th basis vector. If you permute the list $(v_1, v_2)$ to $(v_2, v_1)$ , the same vector $v = 3v_1 + 5v_2$ now has coordinates $(5,3)$ instead of $(3,5)$ . The underlying spanning set is identical, but the coordinate map — crucial for matrix representations of linear maps — depends on the ordering, so linear algebra conventionally treats bases as ordered.

How is "basis" different from just "spanning set" or "independent set"?

A spanning set guarantees every vector can be reached, but possibly in more than one way if the set has redundant vectors. An independent set guarantees no redundancy, but possibly fails to reach every vector in $V$ . A basis is the intersection of these two properties: enough vectors to reach everything, but no more than needed, which is exactly the combination that forces every vector to have one and only one representation, as the central theorem proves.

Can a basis of

\mathbb R^n

have fewer or more than

n

vectors?

No. Although this article proves the basis definition and unique-representation equivalence without yet fixing the number of vectors, a companion result — the Steinitz Exchange Lemma — shows any independent list in $\mathbb R^n$ has at most $n$ vectors, and any spanning list has at least $n$ vectors. Combining these forces every basis of $\mathbb{R}^n$ to have exactly $n$ vectors, which is why we call $n$ the dimension of $\mathbb R^n$ .

What goes wrong if I only check independence and skip checking spanning?

An independent list that does not span $V$ is not a basis; it correctly gives unique representations, but only for the smaller subspace it spans, not for all of $V$ . For example, $((1,0,0),(0,1,0))$ is independent in $\mathbb R^3$ but only spans the $xy$ -plane, missing vectors like $(0,0,1)$ entirely — such a vector has zero representations in this list, not one, so the Unique Representation theorem's conclusion (which asserts existence and uniqueness for every vector in $V$ ) genuinely fails for vectors outside the span.

Is the empty list a basis of any vector space?

Yes: the empty list is the basis of the zero vector space $\{0\}$ . Vacuously, the empty list is linearly independent (there are no nontrivial coefficient choices to check), and its span is defined to be $\{0\}$ (the span of no vectors is, by convention, the singleton containing only the zero vector, since the empty sum equals $0$ ). This matches the theorem: the only vector in $\{0\}$ , namely $0$ itself, has the unique (empty) representation as the empty linear combination.