Dimension of Quotient Spaces

The dimension formula for quotient spaces is not a curiosity — it is one of the load-bearing beams of linear algebra and its applications.

• Rank-nullity theorem. For a linear map $T: V \to W$ with $V$ finite-dimensional, $T$ factors as $V \to V/\ker T \xrightarrow{\sim} \operatorname{im} T$, and applying the dimension formula to $\ker T \subseteq V$ recovers $\dim V = \dim(\ker T) + \dim(\operatorname{im} T)$ — the quotient dimension formula is the conceptual heart of rank-nullity.
• Differential geometry and manifolds. Tangent spaces to quotient manifolds (like projective space $\mathbb{RP}^n = (\mathbb{R}^{n+1}\setminus\{0\})/\sim$) are computed using quotient-space dimension counts, and codimension of a submanifold is the direct geometric analogue of codimension of a subspace.
• Algebraic geometry and commutative algebra. Codimension of subvarieties, and the dimension of quotient rings $R/I$, generalize this exact linear-algebraic idea to more general rings and modules, with Noether's structural viewpoint providing the bridge.
• Coding theory. A linear code $C \subseteq \mathbb{F}_q^n$ has dual distance and syndrome computations governed by $\dim(\mathbb{F}_q^n / C) = n - \dim C$; the syndrome decoding algorithm literally works in the quotient space of the ambient space by the code.
• Statistics. In linear models, residual degrees of freedom are the dimension of the quotient of the observation space by the fitted model's column space, i.e. $n - \operatorname{rank}(X)$, directly instantiating the codimension formula.
• Physics and constrained systems. Gauge theories and constrained Hamiltonian mechanics describe physical phase space as a quotient of a larger configuration space by a subspace of "gauge directions" or constraints, and the dimension of physical phase space is computed exactly via the codimension formula.
• Cryptography. Lattice-based and code-based cryptosystems use quotient constructions $\mathbb{Z}^n / L$ (lattices) and their vector-space analogues over finite fields, where dimension/index counting of the quotient underlies security parameter estimates.

:::quiz[For $V = \mathbb{R}[x]$ (all polynomials) and $U = \{p : p(0)=0\}$, why can't we directly write $\dim(V/U) = \dim V - \dim U$ as ordinary subtraction?]

• Because $V/U$ is not a vector space in this case
• Because $U$ is not actually a subspace of $V$
• Because both $\dim V$ and $\dim U$ are infinite cardinals, and cardinal subtraction is not well-defined the way finite subtraction is ✓
• Because $U$ contains vectors not in $V$

Ask the tutor: How does the quotient-space dimension formula generalize when $U$ itself is infinite-dimensional but has finite codimension in $V$?

Think of $V$ as a large space and $U$ as a subspace sitting inside it, like a line or plane through the origin in $\mathbb{R}^3$. The cosets $v + U$ are translates of $U$; they partition $V$ into parallel copies of $U$, one for each "direction transverse to $U$." The quotient space $V/U$ treats each entire translate as a single point. Algebraically, this is exactly what happens when we declare every vector of $U$ to be zero: two vectors $v$ and $w$ become identified in $V/U$ exactly when their difference lies in $U$. Dimension counting confirms the picture. If $U$ is a plane through the origin in $\mathbb{R}^3$ (dimension 2), the cosets of $U$ are the planes parallel to $U$, and they are indexed by a single real parameter — how far up or down you shift. So $V/U$ should look one-dimensional, matching $\dim V - \dim U = 3 - 2 = 1$. If instead $U$ is a line through the origin (dimension 1), the cosets are lines parallel to $U$, and they are indexed by a two-dimensional family (any point in a transverse plane picks out a unique coset), matching $3 - 1 = 2$. The key structural fact underlying the theorem is that any choice of complement $W$ with $V = U \oplus W$ gives a "copy" of $V/U$ living inside $V$: the restriction of $\pi$ to $W$ is a linear isomorphism $W \to V/U$. This is why the answer does not depend on which complement you pick, even though complements themselves are wildly non-unique.

This corollary explains the informal claim in the intuition section: quotienting by $U$ is, up to a canonical isomorphism, the same as discarding $U$ from any fixed complement — but $V/U$ achieves this without ever choosing one.

Example 1: A concrete subspace of $\mathbb{R}^4$ defined by asymmetric linear conditions. Let $V = \mathbb{R}^4$ and let

We compute $\dim(V/U)$ two ways to confirm the formula. Finding $\dim U$ directly. Solve the system. From the second equation, $x_3 = 3x_4$. Substitute into the first: $x_1 - 2x_2 + 3x_4 = 0$, so $x_1 = 2x_2 - 3x_4$. Treat $x_2, x_4$ as free parameters $s, t$:

The vectors $(2,1,0,0)$ and $(-3,0,3,1)$ are linearly independent (neither is a scalar multiple of the other, e.g. compare second coordinates), so they form a basis of $U$ and $\dim U = 2$. Extending to a basis of $V$. We need two more vectors so that together with $u_1=(2,1,0,0)$, $u_2=(-3,0,3,1)$ they span $\mathbb{R}^4$. Try the standard basis vectors $e_1 = (1,0,0,0)$ and $e_4 = (0,0,0,1)$. Check independence of $\{u_1, u_2, e_1, e_4\}$ by forming the matrix with these as rows and row-reducing:

Swap row 1 and row 3, then eliminate: $R_1=(1,0,0,0)$; $R_2 \to R_2 + 3R_1 = (0,0,3,1)$; $R_3 \to R_3 - 2R_1 = (0,1,0,0)$ (using the original row 1 now in position 3); $R_4=(0,0,0,1)$. The resulting matrix

(after reordering rows) is upper triangular with nonzero diagonal entries $1,1,3,1$, so it has rank 4. Hence $u_1, u_2, e_1, e_4$ is a basis of $\mathbb{R}^4$. Reading off $\dim(V/U)$. By the theorem's proof, the cosets $e_1 + U$ and $e_4 + U$ form a basis of $V/U$, so

Direct check against the formula: $\dim V - \dim U = 4 - 2 = 2$. ✓ Note also that $U$ is literally the solution set of two independent linear equations, matching the general fact that a subspace cut out by $r$ independent linear conditions in $\mathbb{F}^n$ has codimension $r$ — here $r = 2$.

Example 2: The quotient of a polynomial space by an asymmetric subspace. Let $V = \mathcal{P}_4(\mathbb{R})$, the polynomials of degree at most $4$, so $\dim V = 5$ (basis $1, x, x^2, x^3, x^4$). Let

polynomials vanishing at $1$ and at $-2$. Finding $\dim U$. The evaluation map $T: \mathcal{P}_4(\mathbb{R}) \to \mathbb{R}^2$, $T(p) = (p(1), p(-2))$, is linear, and $U = \ker T$. The map $T$ is surjective: for instance $T(x-1) = (0,-3)$ and $T(x+2)=(3,0)$ give two vectors spanning $\mathbb{R}^2$ (they are non-parallel since one has zero first coordinate and the other zero second coordinate, and neither is the zero vector), and linear combinations of $x-1,x+2$ hit any target in $\mathbb{R}^2$. So $\operatorname{rank}(T) = 2$. By the Rank-Nullity Theorem applied to $T$,

Direct basis check. A basis of $U$ is given by polynomials each vanishing at both $1$ and $-2$, built from factors of $(x-1)(x+2) = x^2 + x - 2$:

These three are linearly independent (their leading terms $x^2, x^3, x^4$ are distinct and no cancellation can occur across degrees), and every polynomial in $U$ is divisible by $(x-1)(x+2)$ with quotient of degree $\le 2$, hence a combination of $q_1,q_2,q_3$. So $\dim U = 3$, confirming the rank-nullity computation. Computing $\dim(V/U)$. Extend $q_1, q_2, q_3$ to a basis of $V$ by adjoining, say, $1$ and $x$ (independence check: the list $q_1,q_2,q_3,1,x$ has "leading behavior" $x^2,x^3,x^4,1,x$ which after triangularization by degree is clearly a basis — five vectors in a five-dimensional space, and one checks no nontrivial combination vanishes by comparing coefficients degree by degree). Then by the theorem's proof, $1 + U$ and $x + U$ form a basis of $V/U$, so

This matches the codimension count directly: $U$ is cut out by exactly the two independent linear conditions $p(1)=0$, $p(-2)=0$, and codimension equals the number of independent conditions — exactly as $\operatorname{rank}(T) = 2$ told us via rank-nullity.