Coordinates Relative to a Basis

Suppose you know a vector space of polynomials of degree less than 3. You want to describe a specific polynomial like $5 - 2x + 7x^2$ . You want to do this using nothing but a list of numbers. How can an abstract object — a polynomial, a matrix, an arrow in an unlabeled plane — be turned into a concrete tuple of scalars? This tuple must be something a computer, or a page of arithmetic, can manipulate. The answer is to fix an Ordered Basis. Then record how a vector unfolds as a linear combination of that basis, in order. This idea is one of the true engines of linear algebra. Every finite-dimensional vector space over a field $F$ can be represented faithfully by a tuple in $F^n$ . This holds no matter how exotic its elements look. Matrices of linear maps, systems of linear equations, and geometric transformations all become computable. This happens exactly because coordinates translate abstract vectors into columns of numbers. Without this translation, Linear Transformations would remain a purely conceptual subject. There would be no arithmetic to perform. The big picture is this: a choice of ordered basis for an $n$ -dimensional space $V$ produces something specific. It creates an invertible, linear correspondence between $V$ and $F^n$ . This correspondence is not canonical — it depends on the basis chosen. But once fixed, it is completely rigid. Each vector gets exactly one coordinate tuple. Every tuple in $F^n$ comes from exactly one vector. This rigidity is what makes coordinates trustworthy tools. They are not mere bookkeeping. This article defines the coordinate vector precisely. It proves that the coordinate map is a vector space isomorphism. It works through concrete examples in $F^n$ and in a space of polynomials. It also examines what happens if the spanning or independence hypotheses are dropped. These hypotheses come from the definition of a basis. Either failure destroys the correspondence. The techniques here set up language needed later for matrices of linear maps. They also support associated graded objects such as Quotient Spaces built from coordinate ideas.

Key takeaways

An ordered basis $\beta = (v_1, \dots, v_n)$ of $V$ lists basis vectors in a fixed sequence, not just as a set.
The coordinate vector $[v]_\beta \in F^n$ records the unique scalars expressing $v$ as $\sum a_i v_i$ .
The map $v \mapsto [v]_\beta$ is linear, injective, and surjective onto $F^n$ — an Isomorphism $V \cong F^n$ .
Coordinates convert abstract vector-space computations (sums, scalar multiples, checking dependence) into ordinary arithmetic in $F^n$ .
Changing the ordered basis changes the coordinate vector by an invertible linear substitution (a change of basis).
Every finite-dimensional vector space of dimension $n$ over $F$ is isomorphic to $F^n$ , but the isomorphism is never unique — it always depends on a basis choice.
Key idea: fixing an ordered basis turns "which linear combination is this vector?" into a well-defined, computable list of scalars.

Where this lives today

Coordinates relative to a basis are the reason matrix algebra can represent abstract linear maps at all: once you fix ordered bases for the domain and codomain, a linear transformation becomes a matrix acting on coordinate vectors, which is the entire subject of matrix representations of Linear Transformations. In numerical linear algebra and scientific computing, every vector stored in a computer — whether it represents a polynomial, a signal, a quantum state, or a finite-element solution — is secretly a coordinate vector relative to some chosen basis (monomials, Fourier modes, B-splines, or a mesh basis), and the choice of basis dramatically affects numerical stability and computational cost. In quantum mechanics, a state vector's "coordinates" relative to an eigenbasis of an observable are precisely the amplitudes whose squared magnitudes give measurement probabilities — a direct physical use of coordinate vectors. In computer graphics and robotics, points and directions are given coordinates relative to a chosen frame, and changing frames (translating a robot's arm coordinates into world coordinates) is a direct application of the change-of-basis idea that coordinate vectors make possible. In coding theory and cryptography, vectors in finite-dimensional spaces over finite fields are manipulated purely through their coordinate tuples relative to a fixed basis, since that is the only way to store and transmit them as bits. Finally, in the further development of the subject, coordinates relative to a basis provide the scaffolding for constructing sums of subspaces and — by passing to equivalence classes of coordinate tuples that agree modulo a subspace — for building Quotient Spaces, where the coordinate viewpoint clarifies exactly which information survives the quotient.

:::quiz[If $\beta = (v_1, v_2)$ is an ordered basis of $V$ and $[v]_\beta = \binom{3}{-1}$ , what is $v$ ?]

$v = 3v_1 - v_2$ ✓
$v = -v_1 + 3v_2$
$v = 3v_1 + v_2$
$v$ cannot be determined without more information

Ask the tutor: How does the change-of-basis matrix relate $[v]_\beta$ and $[v]_{\beta'}$ for two different ordered bases, and how is it computed in practice?

V

Intuition

Think of an ordered basis as a set of labeled axes together with a scale on each axis, and a coordinate vector as the reading you take off each axis to pinpoint a vector. In ordinary $\mathbb{R}^2$ with the standard basis $(e_1, e_2)$ , the vector $(3, 5)$ has coordinate vector $\binom{3}{5}$ — no surprise, since the standard basis is the identity dictionary. But choose a different ordered basis, say $\beta = ((1,1), (1,-1))$ , and the same vector $(3,5)$ is described by different numbers: solving $a(1,1) + b(1,-1) = (3,5)$ gives $a = 4, b = -1$ , so $[(3,5)]_\beta = \binom{4}{-1}$ . The vector itself never moved; only the "ruler" used to read it off changed. The crucial fact — proved below — is that this reading-off process is linear and invertible. Linear, because coordinates of a sum are the sum of coordinates, and coordinates of a scaled vector are the scaled coordinates. Invertible, because from a tuple of scalars you can always reconstruct the unique vector they describe by taking the linear combination $\sum a_i v_i$ . This two-way street is what lets us move computations back and forth between an abstract space $V$ and the concrete arithmetic of $F^n$ .

Figure

Two ordered bases assign different coordinates to the same point in the plane

V

\phi_\beta

\beta

\phi_\beta

Worked examples

Example 1: Coordinates in $F^n$ relative to a non-standard basis. Let $V = \mathbb{R}^3$ and consider the ordered basis

\beta = (v_1, v_2, v_3), \qquad v_1 = (1, 0, 1),\ v_2 = (2, 1, 0),\ v_3 = (0, 1, -1).

First we confirm $\beta$ is indeed a basis of $\mathbb{R}^3$ . Since $\dim \mathbb{R}^3 = 3$ and we have exactly 3 vectors, it suffices to check linear independence (a linearly independent list of $n$ vectors in an $n$ -dimensional space is automatically a basis, by the theory of Dimension). Set up

c_1(1,0,1) + c_2(2,1,0) + c_3(0,1,-1) = (0,0,0),

giving the system

c_1 + 2c_2 = 0, \qquad c_2 + c_3 = 0, \qquad c_1 - c_3 = 0.

From the first equation $c_1 = -2c_2$ . From the third, $c_3 = c_1 = -2c_2$ . Substituting into the second: $c_2 + (-2c_2) = 0 \Rightarrow -c_2 = 0 \Rightarrow c_2 = 0$ , hence $c_1 = c_3 = 0$ . Only the trivial solution exists, so $\beta$ is linearly independent, hence a basis. Now find $[u]_\beta$ for $u = (4, 3, -1)$ . We need $a_1, a_2, a_3$ with

a_1(1,0,1) + a_2(2,1,0) + a_3(0,1,-1) = (4,3,-1).

Componentwise:

a_1 + 2a_2 = 4, \qquad a_2 + a_3 = 3, \qquad a_1 - a_3 = -1.

From the first equation, $a_1 = 4 - 2a_2$ . Substitute into the third: $(4 - 2a_2) - a_3 = -1 \Rightarrow a_3 = 5 - 2a_2$ . Substitute into the second: $a_2 + (5 - 2a_2) = 3 \Rightarrow -a_2 + 5 = 3 \Rightarrow a_2 = 2$ . Then $a_1 = 4 - 4 = 0$ and $a_3 = 5 - 4 = 1$ . Check: $0(1,0,1) + 2(2,1,0) + 1(0,1,-1) = (0,0,0) + (4,2,0) + (0,1,-1) = (4,3,-1)$ . ✓ This matches $u$ exactly. So

[u]_\beta = \begin{pmatrix} 0 \\ 2 \\ 1 \end{pmatrix}.

Note this differs entirely from the standard coordinates $(4,3,-1)$ of $u$ — the same vector, two different labels, because the ruler changed.

Example 2: Coordinates in a polynomial space, with a non-monomial basis. Let $V = P_2(\mathbb{R})$ , the space of polynomials of degree at most 2, which has dimension 3. Consider the ordered basis

\gamma = (g_1, g_2, g_3), \qquad g_1 = 1 + x,\quad g_2 = x - x^2,\quad g_3 = 1 + x^2 .

First check $\gamma$ is a basis. Since $\dim P_2(\mathbb{R}) = 3$ and we have 3 vectors, again it suffices to verify linear independence. Suppose

c_1(1+x) + c_2(x - x^2) + c_3(1+x^2) = 0

as a polynomial. Collecting coefficients of $1, x, x^2$ :

(c_1 + c_3) + (c_1 + c_2)x + (-c_2 + c_3)x^2 = 0.

This forces

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

From the second, $c_2 = -c_1$ . From the third, $c_3 = c_2 = -c_1$ . Substituting into the first: $c_1 + (-c_1) = 0$ , which holds automatically — so far this allows a free parameter, and we must check more carefully. Set $c_1 = t$ ; then $c_2 = -t$ , $c_3 = -t$ . Check all three equations with these values: first, $c_1 + c_3 = t - t = 0$ ✓; second, $c_1+c_2 = t - t = 0$ ✓; third, $-c_2+c_3 = t - t = 0$ ✓. So every $t$ satisfies the system, meaning there are nontrivial solutions (e.g. $t=1$ gives $c_1=1,c_2=-1,c_3=-1$ )! This shows $\gamma$ as originally listed is linearly dependent, so it is not a basis — a good reminder to actually verify the hypothesis rather than assume it. Let us instead use a basis that genuinely works, correcting the example:

\gamma = (g_1, g_2, g_3), \qquad g_1 = 1 + x,\quad g_2 = x + x^2,\quad g_3 = 1 + x^2 .

Redo independence: suppose $c_1(1+x) + c_2(x+x^2) + c_3(1+x^2) = 0$ . Collecting coefficients,

(c_1 + c_3) + (c_1 + c_2) x + (c_2 + c_3) x^2 = 0,

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: $(-c_1) + (-c_1) = 0 \Rightarrow -2c_1 = 0 \Rightarrow c_1 = 0$ (since we are over $\mathbb{R}$ , char $\neq 2$ ), hence $c_2 = c_3 = 0$ . Only the trivial solution — $\gamma$ is linearly independent, and since $\dim P_2(\mathbb{R}) = 3$ , it is a basis. Now compute $[p]_\gamma$ for $p(x) = 5 - 2x + 7x^2$ . Solve

c_1(1+x) + c_2(x+x^2) + c_3(1+x^2) = 5 - 2x + 7x^2 .

Matching coefficients:

c_1 + c_3 = 5, \qquad c_1 + c_2 = -2, \qquad c_2 + c_3 = 7.

Add all three equations: $2(c_1+c_2+c_3) = 5 - 2 + 7 = 10$ , so $c_1 + c_2 + c_3 = 5$ . Subtract each original equation from this sum:

c_2 = 5 - 5 = 0, \qquad c_3 = 5 - (-2) = 7, \qquad c_1 = 5 - 7 = -2.

Check: $c_1 + c_3 = -2 + 7 = 5$ ✓; $c_1 + c_2 = -2 + 0 = -2$ ✓; $c_2 + c_3 = 0 + 7 = 7$ ✓. Verify in $V$ directly: $-2(1+x) + 0(x+x^2) + 7(1+x^2) = (-2-2x) + (7+7x^2) = 5 - 2x + 7x^2 = p(x)$ . ✓ So

[p]_\gamma = \begin{pmatrix} -2 \\ 0 \\ 7 \end{pmatrix}.

Compare with the coordinate vector relative to the standard ordered basis $\epsilon = (1, x, x^2)$ , which is simply $[p]_\epsilon = \binom{5}{-2}{7}$ (read off the coefficients directly). The two coordinate vectors for the same polynomial $p$ are different because the underlying rulers $\gamma$ and $\epsilon$ differ — exactly the phenomenon from Example 1, now in a non-arithmetic vector space of functions.

\beta = (v_1,\dots,v_n)

Scope, edge cases & common traps

Situation	What happens
$\beta$ is an ordered basis of finite-dimensional $V$	$\phi_\beta : V \to F^n$ is a well-defined linear isomorphism (the theorem).
The same set of vectors, listed in a different order	Gives a different coordinate map: coordinates get permuted, since $[v]_\beta$ depends on the sequence, not just the set.
$V$ is infinite-dimensional	No finite ordered basis exists in the usual sense; coordinate vectors in $F^n$ are not defined this way (infinite Hamel bases require more care and are outside this scope).
Vectors that merely span but are dependent	Coordinates are not unique — the correspondence is not even a function (see counterexample).
A list shorter than $\dim V$	Fails to span, so some vectors have no coordinate representation at all.
Changing $\beta$ to $\beta'$	Coordinate vectors relate by an invertible linear substitution (a change-of-basis matrix); this is the natural next topic beyond this article.

The mathematician behind it

Origins · 1 minHermann Grassmann and how this was discovered

Hermann Grassmann

1809–1877 · German mathematician and linguist, pioneer of abstract linear algebra

Grassmann worked largely in isolation as a secondary-school teacher in Stettin (then Prussia), developing in his 1844 treatise Die lineale Ausdehnungslehre ("The Theory of Linear Extension") a strikingly modern, coordinate-free calculus of vectors, subspaces, and what we would now call exterior algebra. His ideas anticipated the abstract notion of vector space and basis by decades but were written in dense, idiosyncratic prose that few contemporaries could parse, and his work was largely ignored during his lifetime. Only later, through the efforts of readers such as Giuseppe Peano, who gave a clean axiomatic formulation of vector spaces in 1888 explicitly building on Grassmann's ideas, did the modern concept of basis and coordinates relative to a basis take the form used today.

History

- 1844 — Grassmann publishes the Ausdehnungslehre, introducing an abstract, coordinate-free theory of linear combinations and dependence, the conceptual seed of "basis."

- 1888 — Peano gives the first fully axiomatic definition of a real vector space, making precise the notions later needed for a rigorous basis and coordinate map.

- Early 1900s — Coordinate representations relative to a chosen basis become standard tools in the emerging theory of linear transformations and matrices, tying Grassmann's abstract algebra to concrete computation.

- 1920s–1930s — Hermann Weyl and others formalize the notion of isomorphism between finite-dimensional vector spaces and

F^n

, making the coordinate map's role as the canonical (basis-dependent) isomorphism explicit in textbooks.

- Mid-20th century — The ordered-basis coordinate map becomes a standard first tool in linear algebra pedagogy, appearing in canonical texts as the bridge from abstract vector spaces to matrix computation.

Why do we need coordinates at all if a vector space is already well-defined abstractly?

Abstract vector spaces let us prove general theorems, but they do not by themselves give us a way to compute. You cannot add two abstract "polynomials in some ambient space" using a calculator, but you can add two coordinate tuples in $F^n$ entrywise using ordinary arithmetic. Coordinates relative to a basis translate any finite-dimensional vector space problem — checking dependence, solving equations, representing a linear map as a matrix — into a problem about $F^n$ , which is concrete and computable. This is precisely why the coordinate map is called an isomorphism: it does not lose any linear-algebraic information, so nothing is sacrificed by working in coordinates instead of abstractly.

Does the isomorphism

V \cong F^n

mean

V

and

F^n

are literally the same space?

No. The theorem shows $V$ and $F^n$ are isomorphic as vector spaces, meaning there exists a linear bijection between them that preserves all linear-algebraic structure (sums, scalar multiples, dimension, linear dependence). But the isomorphism $\phi_\beta$ depends entirely on the choice of ordered basis $\beta$ ; a different basis gives a genuinely different isomorphism, with different coordinate vectors assigned to the same element of $V$ . So $V$ and $F^n$ are abstractly indistinguishable as vector spaces, but there is no single "the" isomorphism between them — unlike, say, the identity map from a set to itself.

What happens to coordinate vectors when you change the ordered basis?

If $\beta$ and $\beta'$ are two ordered bases of the same finite-dimensional space $V$ , there is an invertible $n \times n$ matrix $Q$ (the change-of-basis matrix) such that $[v]_{\beta'} = Q\,[v]_\beta$ for every $v \in V$ . The columns of $Q$ are exactly the coordinate vectors of the $\beta'$ -basis vectors expressed relative to $\beta$ (or vice versa, depending on convention). This matrix is always invertible because both $\phi_\beta$ and $\phi_{\beta'}$ are bijections, so the composite $\phi_{\beta'} \circ \phi_\beta^{-1}$ is a bijective linear map on $F^n$ , i.e. represented by an invertible matrix. This is the starting point for the general theory of change of basis, which is a natural next topic after coordinate vectors.

Why must the scalars in the linear combination be unique for coordinates to make sense?

If a vector could be written as two genuinely different linear combinations of the basis vectors, then "the coordinate vector of $v$ " would not be a single well-defined object — it would be ambiguous, and $\phi_\beta$ would fail to even be a function. Uniqueness follows precisely from linear independence: if $\sum a_i v_i = \sum b_i v_i$ , subtracting gives $\sum (a_i - b_i) v_i = 0$ , and linear independence forces every coefficient $a_i - b_i$ to be zero. This is why "basis" bundles together spanning (existence of a representation) and linear independence (uniqueness of that representation) — coordinates need both.

Can coordinate vectors be used for vector spaces over finite fields, like

\mathbb{F}_2

Yes, and the entire theorem applies verbatim, since nothing in its proof used properties special to $\mathbb{R}$ or $\mathbb{C}$ — only the vector space axioms over an arbitrary field $F$ . Over $\mathbb{F}_2 = \{0,1\}$ , for instance, a basis of an $n$ -dimensional space gives coordinate vectors in $\mathbb{F}_2^n$ , which are literally bit strings of length $n$ . This is exactly the setting used in coding theory, where codewords are coordinate vectors of elements of a vector space over a finite field relative to a chosen basis, often called a generator matrix's rows.

Is the coordinate vector the same as the vector's "components" in physics?

Essentially yes, when the physics context fixes an ordered basis (such as $(\hat{x}, \hat{y}, \hat{z})$ or an orthonormal eigenbasis). A physical vector such as a force or a quantum state is basis-independent as an abstract object, but its "components" reported in a lab notebook are exactly its coordinate vector relative to whatever frame or basis was chosen. Changing coordinate systems in physics (rotating axes, or switching bases in quantum mechanics) is precisely the change-of-basis operation described above, and the physical vector itself is unchanged even though its listed components differ.

What goes wrong if I try to define coordinates using a spanning set that has more vectors than the dimension?

If a list has more vectors than $\dim V$ , it cannot be linearly independent (by the theory of Dimension, any linearly independent list has at most $\dim V$ vectors), so at least one vector is redundant. As the counterexample shows, this redundancy destroys uniqueness of the representation, so distinct scalar tuples can represent the same vector, and the assignment $v \mapsto (a_1,\dots,a_k)$ is not a well-defined function at all. Coordinate vectors relative to a basis require exactly $n = \dim V$ vectors, no more and no fewer, precisely so that both existence (spanning) and uniqueness (independence) hold simultaneously.