Subspaces | KEVOS® Mathematics

Overview

A vector space rarely appears in isolation. Most of the sets an engineer actually works with — equilibrium displacement fields, admissible control inputs, balanced reaction vectors, deviatoric stress states — are carved out of a larger, obviously well-behaved space by a set of linear conditions. A subspace is the formal name for such a set when the carving leaves the vector space structure intact.

The definition is deliberately spare: $W$ is a subspace of $V$ when $W$ is a subset of $V$ and $W$ is itself a vector space using the same addition and the same scalar multiplication that $V$ uses. The phrase "the same operations" is not decoration. A subset endowed with a different addition might well be a vector space, but it would not be a subspace, and none of the relationships that make subspaces useful would follow.

What makes subspaces cheap to establish is inheritance. Commutativity, associativity of addition, associativity of scalar multiplication, both distributive laws and the identity rule $1\vec{u} = \vec{u}$ are statements of the form for all vectors, this identity holds. If an identity holds for every vector in $V$ , it certainly holds for every vector in the smaller collection $W$ . Six axioms therefore require no work at all.

The remaining four axioms are existence and closure statements, and those genuinely can fail on a subset. Adding two members of $W$ might produce something outside $W$ ; scaling a member might leave it; $W$ might not contain the zero vector; a member's additive inverse might be missing. Verifying a subspace is exactly the discipline of checking those four points, and the worked example on this page carries the check out in full before the standard shortcut is introduced.

Definition

Subspace

Let $V$ and $W$ be vector spaces with identical definitions of vector addition and scalar multiplication, and suppose $W \subseteq V$ . Then $W$ is a subspace of $V$ .

Both halves matter. $W$ must be contained in $V$ as a set, and $W$ must be a vector space in its own right under the operations it borrows from $V$ . Neither condition implies the other.

Trivial Subspaces

For any vector space $V$ , the whole space $V$ and the one-element set $\{\vec{0}\}$ are each subspaces of $V$ , called the trivial subspaces. Every other subspace is called non-trivial or proper.

The trivial subspaces are the extreme cases against which any structural statement about subspaces should be sanity-checked.

Concepts

Six axioms are inherited, four are not

Sort the ten axioms by logical shape. Commutativity, associativity of addition, associativity of scalar multiplication, distributivity over vector addition, distributivity over scalar addition and the identity rule are all universally quantified identities: they assert that an equation holds for all vectors and scalars. Restricting the range of a universal quantifier can never falsify it, so these six hold on any subset automatically. The other four — additive closure, scalar closure, existence of a zero vector, existence of additive inverses — assert that certain vectors exist inside the set, and shrinking the set can destroy exactly that.

Closure is the substantive question

In the parent space, closure is guaranteed by the axioms and is invisible. In a subset it becomes the whole content of the verification. If $W$ is described by a membership test — a condition that a candidate vector must satisfy — then closure asks whether the test survives the two operations. For a homogeneous linear condition such as $2x_1 - 5x_2 + 7x_3 = 0$ the answer is yes, because the left-hand side is a linear function of the entries and linear functions preserve sums and scalar multiples. For a non-homogeneous condition such as $3x_1 - 5x_2 = 12$ the answer is no.

The zero vector is the fastest disqualifier

A vector space must contain a zero vector, and in a subspace that zero vector is necessarily the zero vector of the parent space, because the zero vector of a vector space is unique. This yields an immediate screening test: compute $\vec{0}_V$ , apply the membership test of $W$ , and if $\vec{0}_V \notin W$ then $W$ is not a subspace and no further work is needed. Sets defined by a non-zero right-hand side fail here at once. The converse is not available — containing $\vec{0}$ proves nothing on its own.

Additive inverses come from scalar closure

It is tempting to treat the existence of additive inverses as an independent obligation. It is not. In any vector space $(-1)\vec{u}$ equals $-\vec{u}$ , so if $W$ is closed under scalar multiplication then for each $\vec{u} \in W$ the vector $(-1)\vec{u} = -\vec{u}$ is also in $W$ . Additive inverses therefore come free once scalar closure is established, which is the first step towards the three-condition shortcut used in practice.

Subspaces of subspaces, and chains

Being a subspace is transitive: if $U$ is a subspace of $W$ and $W$ is a subspace of $V$ , then $U$ is a subspace of $V$ , because containment is transitive and the operations never change along the chain. This makes it legitimate to build a hierarchy — a span inside a constrained set inside a polynomial space — and to reason at whichever level is convenient. It also means that once $W$ is known to be a subspace, every theorem about vector spaces may be applied to $W$ directly.

Geometric reading in $\mathbb{C}^3$ and beyond

In $\mathbb{R}^3$ the subspaces are exactly the origin, the lines through the origin, the planes through the origin, and the whole space. The recurring feature is the origin: a plane that misses the origin is not a subspace, because it fails to contain the zero vector and is closed under neither operation. This picture is a reliable intuition pump but not a proof technique, and it does not survive into $P_n$ , $M_{mn}$ or function spaces, where verification must return to the axioms.

Establishing that a subset is a subspace, from the definition

Confirm the containmentCheck that every element of

W

really is an element of

V

, and that

W

uses

V

's addition and scalar multiplication unchanged.

Screen on the zero vectorApply the membership test of

W

\vec{0}_V

. If it fails,

W

is not a subspace and the analysis stops.

Test additive closureTake arbitrary

\vec{x}, \vec{y} \in W

, form

\vec{x} + \vec{y}

and verify it passes the membership test.

Test scalar closureTake arbitrary

\alpha \in \mathbb{C}

and

\vec{x} \in W

, form

\alpha\vec{x}

and verify it passes the membership test.

Confirm additive inversesEither check

-\vec{x} \in W

directly or observe that

-\vec{x} = (-1)\vec{x}

, which scalar closure has already delivered.

Invoke inheritance for the restThe six identity axioms hold on

W

because they hold on all of

V

. All ten axioms are now in place, so

W

is a subspace.

Equations

Definition of subspace containment

EQ-SUB-01

W \subseteq V, W a vector space under the operations of V

Both conditions are required. Containment alone does not make a subset a subspace, and being a vector space under different operations does not either.

A plane through the origin in $\mathbb{C}^3$

EQ-SUB-02

A plane through the origin in $\mathbb{C}^3$

The archetypal non-trivial subspace: the solution set of one homogeneous linear equation in three unknowns.

Additive closure of a homogeneous condition

EQ-SUB-03

2 (x_{1} + y_{1}) - 5 (x_{2} + y_{2}) + 7 (x_{3} + y_{3}) = (2 x_{1} - 5 x_{2} + 7 x_{3}) + (2 y_{1} - 5 y_{2} + 7 y_{3}) = 0 + 0 = 0

The membership test is a linear function of the entries, so it distributes over the sum. This single line is why homogeneous linear conditions define subspaces.

Scalar closure of a homogeneous condition

EQ-SUB-04

2 (α x_{1}) - 5 (α x_{2}) + 7 (α x_{3}) = α (2 x_{1} - 5 x_{2} + 7 x_{3}) = α \cdot 0 = 0

Scalars factor out of the membership test, so the scaled vector passes it whenever the original does — for every scalar, including zero.

Additive inverse from scalar closure

EQ-SUB-05

- \vec{u} = (- 1) \vec{u} \in W whenever \vec{u} \in W and W is closed under scalar multiplication

Removes additive inverses from the list of obligations, reducing four substantive checks to three.

The two trivial subspaces

EQ-SUB-06

{\vec{0}} \subseteq W \subseteq V for every subspace W of V

Every subspace contains the zero subspace and is contained in the whole space. These bounds hold for every $W$ and are frequently the right sanity check on a claimed result.

Variable Definitions

Symbols used on this page
Symbol	Name	Meaning	Domain / type
$V$	Parent vector space	The ambient space supplying the two operations and the six inherited axioms	any vector space
$W$	Candidate subset	The subset under test, and the subspace if the test passes	subset of V
$\vec{x}$	Member vector	An arbitrary element of $W$ used in the closure arguments	element of W
$\vec{y}$	Second member vector	A second arbitrary element of $W$ , independent of $\vec{x}$	element of W
$α$	Scalar	Arbitrary multiplier used in the scalar closure test	complex number
$\vec{0}$	Zero vector	The unique additive identity, shared by $V$ and every subspace of $V$	element of V
$- \vec{x}$	Additive inverse	The vector satisfying $\vec{x} + (-\vec{x}) = \vec{0}$ ; equal to $(-1)\vec{x}$	element of V
$ℂ^{3}$	Ambient column space	Column vectors of size three with complex entries, used as the running parent space	vector space of dimension 3

Worked Numerical Example

Problem statement

A three-degree-of-freedom node in a finite element model is constrained by a single rigid tie, so its admissible displacement vectors $\vec{x} = (x_1, x_2, x_3)$ must satisfy $4x_1 - x_2 + 5x_3 = 0$ . Establish from the definition that the set of admissible displacements is a subspace of $\mathbb{C}^3$ , checking every axiom that can fail.

State the set and the ambient space
The parent space is $\mathbb{C}^3$ with ordinary column addition and scalar multiplication, which is a vector space. The admissible set is
$Z = {[\begin{matrix} x_{1} \\ x_{2} \\ x_{3} \end{matrix}] ∣ 4 x_{1} - x_{2} + 5 x_{3} = 0}$
Confirm containment
Every element of $Z$ is by construction a column vector of size three with complex entries, so $Z \subseteq \mathbb{C}^3$ . The operations are inherited unchanged — no new addition is being proposed — which is the second half of the subspace definition.
Verify the zero vector
Substitute $\vec{0} = (0,0,0)$ into the membership test: $4(0) - (0) + 5(0) = 0$ , so $\vec{0} \in Z$ . Since the zero vector of a vector space is unique, this same vector must serve as the zero vector of $Z$ , and it does, because addition is unchanged.
Verify additive closure
Let $\vec{x}, \vec{y} \in Z$ , so $4x_1 - x_2 + 5x_3 = 0$ and $4y_1 - y_2 + 5y_3 = 0$ . Apply the membership test to the sum:
$4 (x_{1} + y_{1}) - (x_{2} + y_{2}) + 5 (x_{3} + y_{3}) = (4 x_{1} - x_{2} + 5 x_{3}) + (4 y_{1} - y_{2} + 5 y_{3}) = 0 + 0 = 0$
Verify scalar closure
Let $\alpha \in \mathbb{C}$ and $\vec{x} \in Z$ . Apply the membership test to $\alpha\vec{x}$ , whose entries are $\alpha x_1$ , $\alpha x_2$ , $\alpha x_3$ :
$4 (α x_{1}) - (α x_{2}) + 5 (α x_{3}) = α (4 x_{1} - x_{2} + 5 x_{3}) = α \cdot 0 = 0$
Verify additive inverses
For $\vec{x} \in Z$ the candidate inverse is $-\vec{x} = (-x_1, -x_2, -x_3)$ . Testing it gives $-(4x_1 - x_2 + 5x_3) = -0 = 0$ , so $-\vec{x} \in Z$ . Equivalently, take $\alpha = -1$ in the previous step; the inverse is delivered by scalar closure at no extra cost.
Collect the inherited axioms
Commutativity, associativity of addition, associativity of scalar multiplication, both distributive laws and the rule $1\vec{x} = \vec{x}$ are identities that hold for all vectors of $\mathbb{C}^3$ . Restricting attention to the vectors of $Z$ leaves each identity true, so all six transfer with no verification.
Conclude
All ten axioms hold for $Z$ under the operations of $\mathbb{C}^3$ , and $Z \subseteq \mathbb{C}^3$ . Therefore $Z$ is a subspace of $\mathbb{C}^3$ , and every theorem about vector spaces now applies to $Z$ without restatement.

Result

The admissible displacement set is a genuine vector space, not merely a set of acceptable answers. Practically, this licenses superposition: any scalar multiple of an admissible displacement, and any sum of two admissible displacements, is admissible, so a solver may work with combinations of computed modes without re-checking the tie constraint each time. It also means the constrained problem has a well-defined dimension — here two, one fewer than the unconstrained three.

Applications & Industry Use

Structural engineering

Constrained displacement fields

Boundary conditions and rigid ties impose homogeneous linear relations on nodal displacements. The admissible displacements form a subspace of the full degree-of-freedom space, which is why constrained stiffness matrices can be assembled by restriction rather than by re-deriving the model.

Electrical engineering

Kirchhoff-admissible current vectors

Currents satisfying Kirchhoff's current law at every node form the null space of the incidence matrix, hence a subspace of the space of branch currents. Superposition of admissible current patterns is therefore automatically admissible, which underpins loop analysis.

Continuum mechanics

Deviatoric stress states

Symmetric stress tensors with zero trace form a subspace of the space of symmetric tensors. Because the trace condition is homogeneous and linear, sums and scalings of deviatoric states remain deviatoric, which is exactly what plasticity models rely on when decomposing stress.

Control systems

Reachable and unobservable sets

The reachable set of a linear system and the unobservable set are both subspaces of the state space. Their subspace status is what allows controllability and observability to be reported as a single integer — a dimension — rather than as a description of an arbitrary region.

Chemical engineering

Element-balanced composition vectors

Composition changes that conserve every chemical element satisfy a homogeneous linear system, so they form a subspace of the composition space. Any linear combination of balanced reactions is balanced, which is the formal basis for reaction-network reduction.

Machine learning

Zero-sum and centred feature spaces

Centring a data matrix constrains every feature vector to have zero mean, a homogeneous linear condition. The centred vectors form a subspace of the raw feature space, which is why principal component analysis operates on a space one dimension smaller than the ambient one.

Design Considerations

Homogeneous conditions define subspaces; non-homogeneous ones do not

A membership test of the form linear expression equals zero always yields a subspace. A test with a non-zero constant on the right never does, because the zero vector fails it. When a specification arrives in the form $A\vec{x} = \vec{b}$ with $\vec{b} \ne \vec{0}$ , the solution set is an affine translate of a subspace, and only the associated homogeneous set is a subspace.

Check the operations, not just the elements

Two sets can consist of the same objects and still differ as vector spaces if the operations differ. Before declaring a subspace relationship, confirm explicitly that the candidate uses the parent's addition and scalar multiplication. This trap is common in software, where a wrapper type may quietly redefine an operator.

Screen on the zero vector before doing any real work

The zero-vector test costs one substitution and disposes of a large class of candidates immediately. Make it the first step of any verification. Note carefully that it is a necessary condition only: passing it establishes nothing, and the closure checks still have to be done.

Beware conditions that are non-linear or use inequalities

Membership tests involving products of entries, absolute values, integrality or inequalities almost never define subspaces. A product condition typically fails additive closure; an integrality condition fails scalar closure once a non-integer scalar is admitted; an inequality region fails under multiplication by a negative scalar. Identify the shape of the condition before starting.

Prefer to recognise a subspace rather than prove one

Direct verification is instructive but slow. In practice most sets that arise are already known to be subspaces because they are null spaces, column spaces, row spaces or spans. Recasting a set into one of those forms usually settles the question in a line and gives a spanning set at the same time.

Do not assume the scalars are real

Scalar closure must hold for every scalar in the field, and in this development that field is $\mathbb{C}$ . A set that is closed under real scaling but not complex scaling is not a subspace of a complex vector space. If the application genuinely requires real scalars, say so and work over $\mathbb{R}$ throughout.

Standards & Codes

Notation, interchange and numerical standards that govern how this material is written down, stored and computed in production systems.

Applicable standards, conventions and reference implementations
Reference	Title	Relevance to this topic
`ISO 80000-2`	Quantities and units — Part 2: Mathematics	Prescribes the set-theoretic notation used here, including $\subseteq$ for containment, set-builder braces and the blackboard-bold symbol $\mathbb{C}$ for the complex scalars.
`ISO 10303 (STEP)`	Industrial automation systems and integration — Product data representation and exchange	Represents geometric constraints on assemblies as linear relations between degrees of freedom; the admissible motion set defined by such relations is precisely a subspace of the free-body configuration space.
`IEEE 754-2019`	IEEE Standard for Floating-Point Arithmetic	Governs the rounded arithmetic in which a constraint such as $4x_1 - x_2 + 5x_3 = 0$ is evaluated numerically, so exact membership in a subspace becomes a residual tolerance question in practice.
`ISO/IEC 40314`	Mathematical Markup Language (MathML) Version 3.0	Encodes the set-builder and column vector notation on this page so that constraints are machine-readable and searchable.
`W3C WCAG 2.1 AA`	Web Content Accessibility Guidelines	Requires the column vector and set notation to be presented semantically, so screen readers announce a constraint rather than skipping an image.

Material Selection

For a mathematical topic, "material" is the numeric representation: the scalar field, storage format and precision the computation is built from.

Representation and precision selection
Representation	Select when	Trade-off
Exact rational entries	Small constrained models where membership in the subspace must be certified rather than estimated.	Membership tests evaluate to exactly zero or not, removing tolerance decisions, at the cost of growing coefficients during elimination.
IEEE 754 binary64 entries	Finite element and network models where the constraint data come from measured or meshed geometry.	The membership residual is never exactly zero, so subspace membership becomes a scaled-tolerance test that must be documented.
IEEE 754 binary32 entries	Large-scale or GPU-resident constraint sets where memory bandwidth dominates.	Halves storage, but roughly seven significant digits makes near-membership and true membership hard to distinguish for ill-scaled constraints.
Complex scalars	The default here, and required whenever the constrained space feeds an eigenvalue or frequency-domain analysis.	Doubles memory and arithmetic cost for data that may be entirely real.
Sparse storage of the constraint matrix	The subspace is defined by many constraints, each touching only a handful of degrees of freedom.	Very large models become tractable, but the closure verification must be expressed as a matrix-vector product rather than as an entrywise formula.
Fixed-point entries	Embedded controllers enforcing admissibility constraints on-line with no floating-point unit.	Deterministic timing and small footprint, but scalar closure fails in the strict sense because scaling can overflow or truncate out of the representable set.

Manufacturing Notes

Implementation notes — how the result is actually produced by hand, by algorithm and by library, including cost and numerical behaviour.

Cost of a direct verification

Checking a membership test defined by an $m \times n$ constraint matrix costs one matrix-vector product, about $2mn$ floating-point operations, per candidate vector. The closure arguments themselves are symbolic and cost nothing at run time; they are performed once, on paper, to justify the structure.

Recognising a subspace from its description

In implementation terms, the fastest route is pattern recognition. A set written as $\{\vec{x} \mid A\vec{x} = \vec{0}\}$ is a null space; a set written as all linear combinations of stored vectors is a span. Both are subspaces by standing theorems, so no verification code is needed — only a check that the description really has that shape.

Testing closure numerically

A useful regression test draws random pairs from a claimed subspace, forms sums and random scalar multiples, and asserts that the constraint residual stays within a scaled tolerance. This will not prove a subspace, but it reliably detects an implementation that has quietly introduced a non-homogeneous term or an inequality.

Library representation

NumPy, SciPy and Eigen have no subspace type; a subspace is represented either by a basis matrix whose columns span it or by a constraint matrix whose null space it is. SymPy's nullspace and columnspace return basis lists. Choosing between the two representations early avoids repeated conversions, each of which costs a factorisation.

Guarding against affine sets in disguise

Constraint files exported from CAD and multibody tools often carry non-zero right-hand sides that encode an offset. Code that assumes a subspace will silently produce wrong superpositions. Assert that the right-hand side is zero, or split the description explicitly into a particular vector plus a subspace, before any linear combination is formed.

Failure Modes & Common Mistakes

Failure modes, root causes and prevention
Failure mode / mistake	Impact	Root cause	Prevention & detection
Treating a non-homogeneous solution set as a subspace	high	The set $\{\vec{x} \mid A\vec{x} = \vec{b}\}$ with $\vec{b} \ne \vec{0}$ omits the zero vector and is not closed under either operation.	Test the zero vector first; if it fails, describe the set as a particular solution plus a subspace instead.
Concluding subspace from zero-vector membership alone	high	Confusing a necessary condition with a sufficient one; many non-subspaces contain $\vec{0}$ .	Always follow the zero-vector screen with explicit additive and scalar closure checks.
Changing the operations	high	Defining a new addition or scaling on the subset, so the result may be a vector space but is not a subspace of the parent.	State the inherited operations explicitly; reject any candidate that redefines them.
Non-linear membership tests	medium	Conditions such as $x_1 x_2 = 0$ or $\det(A) = 0$ contain $\vec{0}$ but fail additive closure.	Check the degree of the condition; only conditions linear and homogeneous in the entries are safe.
Integrality or discreteness constraints	medium	Restricting entries to integers survives addition but fails scalar closure as soon as a non-integer scalar is used.	Test scalar closure with a fractional scalar such as $\tfrac{1}{2}$ before accepting a discretely defined set.
Inequality-defined regions	medium	Half-spaces, cones and boxes are closed under addition or under non-negative scaling but not under multiplication by a negative scalar.	Test with $\alpha = -1$ ; a region that is not symmetric about the origin cannot be a subspace.
Forgetting the empty set	low	A set defined by contradictory conditions is empty, has no zero vector, and cannot be a subspace despite satisfying closure vacuously.	Exhibit at least one member before starting the closure arguments.
Assuming a subspace of a subspace needs re-verification	low	Not recognising that the subspace relation is transitive, leading to duplicated work along a chain.	Verify at the innermost level once and cite transitivity for the rest of the chain.

FAQs

Why does a subspace get six of the ten axioms for free?

Those six are identities asserted for all vectors and all scalars, such as $\vec{x} + \vec{y} = \vec{y} + \vec{x}$ . A statement that is true for every vector in $V$ remains true when attention is restricted to the vectors in a subset $W$ . Only the axioms that assert the existence of a vector inside the set can be broken by shrinking it.

Must a subspace contain the zero vector of the parent space?

Yes. A vector space must have a zero vector, and the zero vector of a vector space is unique. Since $W$ uses the same addition as $V$ , the element of $W$ that acts as an additive identity must also act as one in $V$ , and therefore equals $\vec{0}_V$ . This is what makes the zero-vector screen valid.

Is a line in

\mathbb{R}^3

that does not pass through the origin a subspace?

No. It does not contain the zero vector, so it fails immediately. It is an affine subspace — a translate of a genuine subspace — and its structure is captured by writing it as a particular vector plus the line through the origin parallel to it.

Can a subspace equal the whole space?

Yes. Every vector space is a subspace of itself, since it is a subset of itself and is by assumption a vector space under its own operations. Together with the zero subspace $\{\vec{0}\}$ these are the two trivial subspaces, and both are genuine subspaces despite the name.

Do I have to check additive inverses separately?

No, provided scalar closure has already been established. In any vector space $(-1)\vec{u} = -\vec{u}$ , so scalar closure applied with $\alpha = -1$ places $-\vec{u}$ inside the set. This observation is one of the two steps that reduce the verification to a short three-condition test.

Is the set of all invertible matrices a subspace of

M_{nn}

No, and it fails in several ways at once. It does not contain the zero matrix, and it is not closed under addition — the identity and its negative are both invertible but sum to the zero matrix. The invertible matrices form a group under multiplication, which is a different and non-linear structure.

How does a subspace differ from a subset?

Every subspace is a subset, but very few subsets are subspaces. A subset is any collection of members of $V$ ; a subspace additionally has to be closed under both operations and to contain the zero vector, so that it is a vector space in its own right and supports superposition.

References

Beezer, R. A. A First Course in Linear Algebra, Version 0.70. University of Puget Sound, 2006. Section S, Definition S. Licensed under the GNU Free Documentation License v1.2.
Halmos, P. R. Finite-Dimensional Vector Spaces, 2nd edition. Springer, 1974.
Strang, G. Introduction to Linear Algebra, 5th edition. Wellesley-Cambridge Press, 2016.
ISO 80000-2:2019, Quantities and units — Part 2: Mathematics. International Organization for Standardization.
W3C. Mathematical Markup Language (MathML) Version 3.0, 2nd edition (ISO/IEC 40314).

AI Suggested Questions

Give me three subsets of $\mathbb{C}^3$ that each fail exactly one requirement for being a subspace, and identify which one fails.
Explain why the set of singular $2 \times 2$ matrices contains the zero matrix yet is still not a subspace of $M_{22}$ .
How do I convert the solution set of a non-homogeneous system into a particular vector plus a subspace?
Show that the intersection of two subspaces is always a subspace but the union usually is not.
Which subsets of the space of continuous functions on an interval are subspaces, and which common ones are not?
Prove that the subspace relation is transitive and give an example of a three-level chain of subspaces.

Related Calculators

Subspace Membership Checker

Test whether a given vector satisfies a set of homogeneous linear conditions, with a configurable residual tolerance.

Closure Property Tester

Sample pairs from a described set and report whether additive and scalar closure hold under the stated conditions.

Null Space Basis Calculator

Return a basis and dimension for the subspace defined by a homogeneous constraint matrix.