Visual Tools
Calculators
Tables
Mathematical Keyboard
Converters
Other Tools

Linear Algebra Formulas

Representation
Operations
Norm & Distance
Dot Product
Cross Product
Triple Products
Vector Space
Subspaces
Span
Linear Independence
Basis & Coordinates
Dimension & Rank
Transpose & Symmetry
Special Matrix Types
Inverse
Rank
Trace
Definitions
Cofactor Structure
Cofactor Expansion
Row Operation Effects
Algebraic Properties
Special Determinants
Adjugate & Inverse
Linear Systems
Geometric Interpretation
Eigenvalue Connection
Standard Forms
Echelon Forms
Elementary Row Operations
Solvability
Homogeneous Systems
Definition & Properties
Matrix Representation
Image & Kernel
Similarity & Basis Change
Geometric Transformations
Foundation
Characteristic Polynomial
Multiplicities
Eigenvalue Algebra
Diagonalizability
Special Spectra
Spectral
Applications
Complex
Inner Product
Orthogonal Complement
Orthogonal Sets
Projection
Gram-Schmidt
Least Squares
LU
Cholesky
QR
SVD
Cross-Decomposition
257 formulas

Representation

(4 formulas)

Vector Component Form

v=(v1,v2,,vn)Rn\mathbf{v} = (v_1, v_2, \ldots, v_n) \in \mathbb{R}^n
See details
Back to top
explanationnotationrelated formulasrelated definitions
An nn-dimensional vector is specified as an ordered tuple of nn real numbers. Each entry viv_i is a component, encoding signed displacement along the ii-th coordinate axis. The component form bridges the geometric picture (an arrow with magnitude and direction) and the algebraic object on which all subsequent operations act.
Back to top

Standard Basis Decomposition

v=v1e1+v2e2++vnen\mathbf{v} = v_1 \mathbf{e}_1 + v_2 \mathbf{e}_2 + \cdots + v_n \mathbf{e}_n
See details
Back to top
explanationnotationrelated formulasrelated definitions
Every vector in Rn\mathbb{R}^n is the linear combination of the standard basis vectors ei\mathbf{e}_i, with the components viv_i themselves serving as the coefficients. The decomposition makes the link between coordinates and basis explicit — coordinates are the weights in the expansion against a fixed reference frame.
Back to top

Direction Cosines

cosα=v1v,cosβ=v2v,cosγ=v3v\cos\alpha = \frac{v_1}{\|\mathbf{v}\|}, \quad \cos\beta = \frac{v_2}{\|\mathbf{v}\|}, \quad \cos\gamma = \frac{v_3}{\|\mathbf{v}\|}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The direction cosines are the cosines of the angles a nonzero vector makes with the three positive coordinate axes in R3\mathbb{R}^3. They package directional information into three scalars and depend only on the unit vector v^=v/v\hat{\mathbf{v}} = \mathbf{v}/\|\mathbf{v}\|, not on the length of v\mathbf{v}.
Back to top

Direction Cosines Identity

cos2α+cos2β+cos2γ=1\cos^2\alpha + \cos^2\beta + \cos^2\gamma = 1
See details
Back to top
explanationderivationrelated formulasrelated definitions
The three direction cosines of a vector in R3\mathbb{R}^3 are not independent — their squares sum to 11. This identity follows directly from v^2=1\|\hat{\mathbf{v}}\|^2 = 1 when v^\hat{\mathbf{v}} is expressed in components.
Back to top

Operations

(15 formulas)

Vector Addition

a+b=(a1+b1, a2+b2, , an+bn)\mathbf{a} + \mathbf{b} = (a_1 + b_1,\ a_2 + b_2,\ \ldots,\ a_n + b_n)
See details
Back to top
explanationconditionsrelated formulasrelated definitions
Addition pairs corresponding components and sums them. The result is again a vector in the same Rn\mathbb{R}^n. Geometrically, the tip-to-tail or parallelogram constructions both produce the same sum.
Back to top

Vector Subtraction

ab=a+(b)=(a1b1, a2b2, , anbn)\mathbf{a} - \mathbf{b} = \mathbf{a} + (-\mathbf{b}) = (a_1 - b_1,\ a_2 - b_2,\ \ldots,\ a_n - b_n)
See details
Back to top
explanationrelated formulasrelated definitions
Subtraction is addition combined with negation. Geometrically, with a\mathbf{a} and b\mathbf{b} drawn from a common tail, the difference ab\mathbf{a} - \mathbf{b} runs from the tip of b\mathbf{b} to the tip of a\mathbf{a}. This connects subtraction directly to distance: d(a,b)=abd(\mathbf{a}, \mathbf{b}) = \|\mathbf{a} - \mathbf{b}\|.
Back to top

Scalar Multiplication of Vectors

ca=(ca1, ca2, , can)c\mathbf{a} = (ca_1,\ ca_2,\ \ldots,\ ca_n)
See details
Back to top
explanationrelated formulasrelated definitions
Multiplying a vector by a scalar cc scales every component by cc. Geometrically, the result has length ca|c|\,\|\mathbf{a}\| and points in the same direction as a\mathbf{a} when c>0c > 0, the opposite direction when c<0c < 0, and collapses to 0\mathbf{0} when c=0c = 0.
Back to top

Linear Combination

c1v1+c2v2++ckvkc_1 \mathbf{v}_1 + c_2 \mathbf{v}_2 + \cdots + c_k \mathbf{v}_k
See details
Back to top
explanationrelated formulasrelated definitions
A weighted sum of vectors, with scalar weights ciRc_i \in \mathbb{R}. Vector addition and scalar multiplication are both special cases. Asking whether a vector b\mathbf{b} is a linear combination of v1,,vk\mathbf{v}_1, \ldots, \mathbf{v}_k is equivalent to asking whether the system Ax=bA\mathbf{x} = \mathbf{b} has a solution, where AA has the vi\mathbf{v}_i as columns.
Back to top

Span

Span{v1,,vk}={c1v1++ckvkciR}\text{Span}\{\mathbf{v}_1, \ldots, \mathbf{v}_k\} = \{c_1 \mathbf{v}_1 + \cdots + c_k \mathbf{v}_k \mid c_i \in \mathbb{R}\}
See details
Back to top
explanationrelated formulasrelated definitions
The span is the set of every linear combination of the given vectors. It is always a subspace of Rn\mathbb{R}^n containing the zero vector. The geometric shape reflects the input — a line for one nonzero vector, a plane for two non-parallel vectors, all of Rn\mathbb{R}^n for nn linearly independent vectors.
Back to top

Matrix Equality

A=B    aij=bij for all i,jA = B \iff a_{ij} = b_{ij} \text{ for all } i, j
See details
Back to top
explanationconditionsrelated formulasrelated definitions
Two matrices are equal precisely when they share the same dimensions and every corresponding entry matches. A single mismatched entry breaks equality. Matrices of different shapes are never equal regardless of contents — a 2×32 \times 3 matrix cannot equal a 3×23 \times 2 matrix.
Back to top

Matrix Addition

(A+B)ij=aij+bij(A + B)_{ij} = a_{ij} + b_{ij}
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
Matrix addition is performed entry by entry. The sum has the same shape as the operands. Equipped with addition and scalar multiplication, the set of m×nm \times n matrices forms a vector space of dimension mnmn.
Back to top

Matrix Subtraction

(AB)ij=aijbij(A - B)_{ij} = a_{ij} - b_{ij}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
Subtraction is defined as addition of the additive inverse: AB=A+(B)A - B = A + (-B). Entry by entry, this corresponds to subtracting matching components.
Back to top

Scalar Multiplication of Matrices

(cA)ij=caij(cA)_{ij} = c \cdot a_{ij}
See details
Back to top
explanationvariantsrelated formulasrelated definitions
Multiplying a matrix by a scalar scales every entry by that scalar. Together with addition, this operation gives the matrix space its vector space structure.
Back to top

Matrix Multiplication

(AB)ij=k=1naikbkj(AB)_{ij} = \sum_{k=1}^{n} a_{ik}\, b_{kj}
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
The (i,j)(i,j) entry of the product is the dot product of row ii of AA with column jj of BB. The number of columns of AA must equal the number of rows of BB, and the result has dimensions m×pm \times p.
Back to top

Matrix-Vector Product (Column Form)

Ax=x1a1+x2a2++xnanA\mathbf{x} = x_1 \mathbf{a}_1 + x_2 \mathbf{a}_2 + \cdots + x_n \mathbf{a}_n
See details
Back to top
explanationnotationrelated formulasrelated definitions
The product AxA\mathbf{x} is a linear combination of the columns of AA, weighted by the entries of x\mathbf{x}. This single observation underlies the theory of linear systems, transformations, and the column space.
Back to top

Matrix Multiplication Associativity

(AB)C=A(BC)(AB)C = A(BC)
See details
Back to top
explanationconditionsrelated formulasrelated definitions
Matrix multiplication is associative whenever both products are defined. The grouping of three or more factors does not affect the result, allowing chains ABCDABCD to be written without parentheses.
Back to top

Matrix Multiplication Distributivity

A(B+C)=AB+AC,(A+B)C=AC+BCA(B + C) = AB + AC, \qquad (A + B)C = AC + BC
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
Matrix multiplication distributes over matrix addition from both sides. Because multiplication is non-commutative, the left- and right-distributive laws are stated separately, but both hold whenever dimensions align.
Back to top

Matrix Multiplication Non-Commutativity

ABBAin generalAB \neq BA \quad \text{in general}
See details
Back to top
explanationderivationvariantsrelated formulasrelated definitions
Matrix multiplication does not generally commute, even when both products are defined and have matching dimensions. This asymmetry distinguishes matrix algebra from scalar arithmetic and has far-reaching consequences for inverses, powers, and transformations.
Back to top

Matrix Power

A0=I,Ak=AAAk factors,Ak=(A1)kA^0 = I, \qquad A^k = \underbrace{A \cdot A \cdots A}_{k \text{ factors}}, \qquad A^{-k} = (A^{-1})^k
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
Powers are defined for square matrices through repeated multiplication, with A0=IA^0 = I by convention. Negative powers exist only when AA is invertible. The standard exponent laws AjAk=Aj+kA^j A^k = A^{j+k} and (Aj)k=Ajk(A^j)^k = A^{jk} hold for all integers when AA is invertible, and for non-negative integers in general.
Back to top

Norm & Distance

(6 formulas)

Euclidean Norm

v=v12+v22++vn2=i=1nvi2\|\mathbf{v}\| = \sqrt{v_1^2 + v_2^2 + \cdots + v_n^2} = \sqrt{\sum_{i=1}^{n} v_i^2}
See details
Back to top
explanationderivationrelated formulasrelated definitions
The norm assigns a single non-negative real number — the length — to every vector in Rn\mathbb{R}^n. In R2\mathbb{R}^2 and R3\mathbb{R}^3 this matches the geometric length via the Pythagorean theorem; the formula extends the concept algebraically to any dimension.
Back to top

Distance Formula

d(a,b)=ab=(a1b1)2+(a2b2)2++(anbn)2d(\mathbf{a}, \mathbf{b}) = \|\mathbf{a} - \mathbf{b}\| = \sqrt{(a_1 - b_1)^2 + (a_2 - b_2)^2 + \cdots + (a_n - b_n)^2}
See details
Back to top
explanationrelated formulasrelated definitions
The Euclidean distance between two vectors is the norm of their difference. For position vectors, this is the straight-line distance between the points they identify. The distance is symmetric, non-negative, and zero only when a=b\mathbf{a} = \mathbf{b}.
Back to top

Vector Normalization

v^=vv\hat{\mathbf{v}} = \frac{\mathbf{v}}{\|\mathbf{v}\|}
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
Dividing a nonzero vector by its norm produces the unit vector pointing in the same direction. Normalization extracts pure direction by stripping out length information.
Back to top

Norm Scaling Property

cv=cv\|c\mathbf{v}\| = |c|\,\|\mathbf{v}\|
See details
Back to top
explanationderivationrelated formulasrelated definitions
Multiplying a vector by a scalar cc multiplies its norm by c|c|. The absolute value is required because a negative scalar reverses direction without producing a negative length.
Back to top

Triangle Inequality

a+ba+b\|\mathbf{a} + \mathbf{b}\| \leq \|\mathbf{a}\| + \|\mathbf{b}\|
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
The length of a sum never exceeds the sum of the lengths. Geometrically, the direct path from start to finish in the tip-to-tail construction is no longer than the path that traces both vectors end to end.
Back to top

Cauchy-Schwarz Inequality

abab|\mathbf{a} \cdot \mathbf{b}| \leq \|\mathbf{a}\|\,\|\mathbf{b}\|
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
The absolute value of the dot product is bounded by the product of the norms. This bound makes the angle formula well-posed — the ratio abab\frac{\mathbf{a} \cdot \mathbf{b}}{\|\mathbf{a}\|\,\|\mathbf{b}\|} always lies in [1,1][-1, 1], where arccos\arccos is defined.
Back to top

Dot Product

(8 formulas)

Dot Product (Algebraic)

ab=a1b1+a2b2++anbn=i=1naibi\mathbf{a} \cdot \mathbf{b} = a_1 b_1 + a_2 b_2 + \cdots + a_n b_n = \sum_{i=1}^{n} a_i b_i
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The dot product collapses two vectors into a single scalar by multiplying corresponding components and summing the results. Unlike addition or scalar multiplication, the output is not a vector — it is a number that measures how the two vectors relate.
Back to top

Dot Product (Geometric)

ab=abcosθ\mathbf{a} \cdot \mathbf{b} = \|\mathbf{a}\|\,\|\mathbf{b}\|\cos\theta
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
The dot product equals the product of the norms scaled by the cosine of the angle between the vectors. This form reveals what the dot product measures — directional alignment. It is positive for acute angles, zero for perpendicular vectors, and negative for obtuse angles.
Back to top

Angle Between Vectors

cosθ=abab\cos\theta = \frac{\mathbf{a} \cdot \mathbf{b}}{\|\mathbf{a}\|\,\|\mathbf{b}\|}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
Solving the geometric form of the dot product for cosθ\cos\theta extracts the angle between two vectors from their components. The right side equals a^b^\hat{\mathbf{a}} \cdot \hat{\mathbf{b}} — the dot product of the normalized vectors — confirming that the angle depends only on direction, not magnitude.
Back to top

Self Dot Product

vv=v12+v22++vn2=v2\mathbf{v} \cdot \mathbf{v} = v_1^2 + v_2^2 + \cdots + v_n^2 = \|\mathbf{v}\|^2
See details
Back to top
explanationrelated formulasrelated definitions
The dot product of a vector with itself equals the square of its norm. This identity ties the algebraic operation directly to length — squared length is not a separate concept but a special case of the dot product. It also gives an alternative norm formula: v=vv\|\mathbf{v}\| = \sqrt{\mathbf{v} \cdot \mathbf{v}}.
Back to top

Orthogonality Condition

ab=0    ab\mathbf{a} \cdot \mathbf{b} = 0 \iff \mathbf{a} \perp \mathbf{b}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
Two vectors are orthogonal precisely when their dot product is zero. In R2\mathbb{R}^2 and R3\mathbb{R}^3 this matches geometric perpendicularity; in higher dimensions, the algebraic condition serves as the definition. The zero vector is orthogonal to every vector by convention.
Back to top

Scalar Projection

compba=abb\text{comp}_{\mathbf{b}}\,\mathbf{a} = \frac{\mathbf{a} \cdot \mathbf{b}}{\|\mathbf{b}\|}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The signed length of the projection of a\mathbf{a} onto the direction of b\mathbf{b}. Positive when a\mathbf{a} leans toward b\mathbf{b}, negative when away, zero when orthogonal. The result is a scalar carrying signed-distance information along the direction of b\mathbf{b}.
Back to top

Vector Projection

projba=abb2b=abbbb\text{proj}_{\mathbf{b}}\,\mathbf{a} = \frac{\mathbf{a} \cdot \mathbf{b}}{\|\mathbf{b}\|^2}\,\mathbf{b} = \frac{\mathbf{a} \cdot \mathbf{b}}{\mathbf{b} \cdot \mathbf{b}}\,\mathbf{b}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The component of a\mathbf{a} in the direction of b\mathbf{b}, expressed as an actual vector parallel to b\mathbf{b}. The scalar projection rescaled by b/b\mathbf{b}/\|\mathbf{b}\| to produce a vector with the appropriate length and orientation.
Back to top

Orthogonal Decomposition

a=projba+a,a=aprojba\mathbf{a} = \text{proj}_{\mathbf{b}}\,\mathbf{a} + \mathbf{a}_{\perp}, \quad \mathbf{a}_{\perp} = \mathbf{a} - \text{proj}_{\mathbf{b}}\,\mathbf{a}
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
Every vector a\mathbf{a} splits uniquely into a component along b\mathbf{b} and a component perpendicular to b\mathbf{b}. The perpendicular part satisfies ab=0\mathbf{a}_{\perp} \cdot \mathbf{b} = 0. This decomposition underlies Gram–Schmidt orthogonalization and least-squares fitting.
Back to top

Cross Product

(8 formulas)

Cross Product (Component Form)

a×b=(a2b3a3b2, a3b1a1b3, a1b2a2b1)\mathbf{a} \times \mathbf{b} = (a_2 b_3 - a_3 b_2,\ a_3 b_1 - a_1 b_3,\ a_1 b_2 - a_2 b_1)
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The cross product takes two vectors in R3\mathbb{R}^3 and returns a third vector built from cyclic differences of pairwise products. Unlike the dot product, the output is a vector — perpendicular to both inputs — and the operation is defined exclusively in three dimensions.
Back to top

Cross Product (Determinant Form)

a×b=ijka1a2a3b1b2b3\mathbf{a} \times \mathbf{b} = \begin{vmatrix} \mathbf{i} & \mathbf{j} & \mathbf{k} \\ a_1 & a_2 & a_3 \\ b_1 & b_2 & b_3 \end{vmatrix}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
A symbolic 3×33 \times 3 determinant with the standard basis vectors in the first row. Cofactor expansion along that row reproduces the component formula term by term. Placing vectors in the top row makes this a notational device rather than a true determinant, but it is the standard mnemonic for organizing the six products and their signs.
Back to top

Cross Product Magnitude

a×b=absinθ\|\mathbf{a} \times \mathbf{b}\| = \|\mathbf{a}\|\,\|\mathbf{b}\|\sin\theta
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The length of a×b\mathbf{a} \times \mathbf{b} equals the area of the parallelogram spanned by a\mathbf{a} and b\mathbf{b}. The cross product peaks at θ=π/2\theta = \pi/2 (rectangle, maximum area) and vanishes at θ=0\theta = 0 or π\pi (parallel vectors, no area). This complements the dot product, which involves cosθ\cos\theta rather than sinθ\sin\theta.
Back to top

Standard Basis Cross Products

i×j=k,j×k=i,k×i=j\mathbf{i} \times \mathbf{j} = \mathbf{k}, \quad \mathbf{j} \times \mathbf{k} = \mathbf{i}, \quad \mathbf{k} \times \mathbf{i} = \mathbf{j}
See details
Back to top
explanationvariantsrelated formulasrelated definitions
The cross products of the standard basis vectors follow a cyclic loop ijki\mathbf{i} \to \mathbf{j} \to \mathbf{k} \to \mathbf{i}. Going forward around the loop yields a positive basis vector; reversing the order negates the result. Any cross product in R3\mathbb{R}^3 reduces to these nine cases via distributivity.
Back to top

Cross Product Anti-Commutativity

a×b=(b×a)\mathbf{a} \times \mathbf{b} = -(\mathbf{b} \times \mathbf{a})
See details
Back to top
explanationrelated formulasrelated definitions
Swapping the operands reverses the output. In the right-hand rule, exchanging which vector the fingers follow and which they curl toward sends the thumb the other way. This contrasts sharply with the dot product, where order does not matter.
Back to top

Parallelism Test (Cross Product)

a×b=0    ab\mathbf{a} \times \mathbf{b} = \mathbf{0} \iff \mathbf{a} \parallel \mathbf{b}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
Two vectors in R3\mathbb{R}^3 are parallel exactly when their cross product is the zero vector. The geometric reason: parallel vectors form an angle of 00 or π\pi, making sinθ=0\sin\theta = 0 and collapsing the cross product magnitude. The zero vector is parallel to every vector by convention.
Back to top

Vector Triple Product

a×(b×c)=(ac)b(ab)c\mathbf{a} \times (\mathbf{b} \times \mathbf{c}) = (\mathbf{a} \cdot \mathbf{c})\,\mathbf{b} - (\mathbf{a} \cdot \mathbf{b})\,\mathbf{c}
See details
Back to top
explanationvariantsrelated formulasrelated definitions
The triple cross product reduces to a linear combination of the inner two vectors, with coefficients given by dot products. The mnemonic "BAC minus CAB" captures the order: the middle vector b\mathbf{b} comes first, scaled by ac\mathbf{a} \cdot \mathbf{c}. Useful for simplifying nested cross products without expanding components.
Back to top

Lagrange Identity

(a×b)(c×d)=(ac)(bd)(ad)(bc)(\mathbf{a} \times \mathbf{b}) \cdot (\mathbf{c} \times \mathbf{d}) = (\mathbf{a} \cdot \mathbf{c})(\mathbf{b} \cdot \mathbf{d}) - (\mathbf{a} \cdot \mathbf{d})(\mathbf{b} \cdot \mathbf{c})
See details
Back to top
explanationrelated formulasrelated definitions
The dot product of two cross products expands into a 2×22 \times 2 determinant of dot products. Setting c=a\mathbf{c} = \mathbf{a} and d=b\mathbf{d} = \mathbf{b} recovers the magnitude identity a×b2=a2b2(ab)2\|\mathbf{a} \times \mathbf{b}\|^2 = \|\mathbf{a}\|^2 \|\mathbf{b}\|^2 - (\mathbf{a} \cdot \mathbf{b})^2, which is equivalent to sin2θ=1cos2θ\sin^2\theta = 1 - \cos^2\theta.
Back to top

Triple Products

(4 formulas)

Scalar Triple Product

a(b×c)=a1a2a3b1b2b3c1c2c3\mathbf{a} \cdot (\mathbf{b} \times \mathbf{c}) = \begin{vmatrix} a_1 & a_2 & a_3 \\ b_1 & b_2 & b_3 \\ c_1 & c_2 & c_3 \end{vmatrix}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
Three vectors in R3\mathbb{R}^3 combine into a single scalar via a cross product nested inside a dot product. The result is the signed volume of the parallelepiped with the three vectors as edges. The determinant arrangement of the nine components computes the same number.
Back to top

Parallelogram Area

Area=a×b\text{Area} = \|\mathbf{a} \times \mathbf{b}\|
See details
Back to top
explanationvariantsrelated formulasrelated definitions
The magnitude of the cross product equals the area of the parallelogram with adjacent sides a\mathbf{a} and b\mathbf{b}. From the magnitude formula absinθ\|\mathbf{a}\|\,\|\mathbf{b}\|\sin\theta — the standard base-times-height formula for parallelogram area, where bsinθ\|\mathbf{b}\|\sin\theta is the height perpendicular to a\mathbf{a}.
Back to top

Parallelepiped Volume

V=a(b×c)V = |\mathbf{a} \cdot (\mathbf{b} \times \mathbf{c})|
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The absolute value of the scalar triple product equals the volume of the parallelepiped with edges a\mathbf{a}, b\mathbf{b}, c\mathbf{c}. The signed version of the triple product encodes orientation — positive for right-handed triples, negative for left-handed. Taking the absolute value strips that information to leave the geometric volume.
Back to top

Pyramid Volume

V=16a(b×c)V = \tfrac{1}{6}|\mathbf{a} \cdot (\mathbf{b} \times \mathbf{c})|
See details
Back to top
explanationrelated formulasrelated definitions
The volume of the tetrahedron (triangular pyramid) with edges a\mathbf{a}, b\mathbf{b}, c\mathbf{c} from a common vertex is one-sixth of the corresponding parallelepiped volume. The factor 16\tfrac{1}{6} comes from 13\tfrac{1}{3} (general pyramid: V=13baseheightV = \tfrac{1}{3} \cdot \text{base} \cdot \text{height}) times 12\tfrac{1}{2} (the triangular base is half the parallelogram base).
Back to top

Vector Space

(3 formulas)

Vector Space Axioms

$$\begin{aligned}
&\text{For all } \mathbf{u}, \mathbf{v}, \mathbf{w} \in V \text{ and all } c, d \in \mathbb{F}: \\
&(1)\ \mathbf{u} + \mathbf{v} \in V \quad (2)\ \mathbf{u} + \mathbf{v} = \mathbf{v} + \mathbf{u} \\
&(3)\ (\mathbf{u} + \mathbf{v}) + \mathbf{w} = \mathbf{u} + (\mathbf{v} + \mathbf{w}) \\
&(4)\ \exists\, \mathbf{0} \in V: \mathbf{v} + \mathbf{0} = \mathbf{v} \\
&(5)\ \exists\, -\mathbf{v} \in V: \mathbf{v} + (-\mathbf{v}) = \mathbf{0} \\
&(6)\ c\mathbf{v} \in V \quad (7)\ c(d\mathbf{v}) = (cd)\mathbf{v} \\
&(8)\ c(\mathbf{u} + \mathbf{v}) = c\mathbf{u} + c\mathbf{v} \\
&(9)\ (c + d)\mathbf{v} = c\mathbf{v} + d\mathbf{v} \quad (10)\ 1\mathbf{v} = \mathbf{v}
\end{aligned}$$
See details
Back to top
explanationconditionsrelated formulasrelated definitions
A vector space over a field F\mathbb{F} is a set VV with two operations — vector addition and scalar multiplication — satisfying ten axioms. Five govern addition (closure, commutativity, associativity, zero, inverse), and five govern scalar multiplication (closure, scalar associativity, two distributive laws, multiplicative identity). Any structure satisfying all ten inherits the entire theory of linear algebra; any structure violating even one is disqualified.
Back to top

Scalar-Zero Property

0v=0,c0=00\mathbf{v} = \mathbf{0}, \quad c\mathbf{0} = \mathbf{0}
See details
Back to top
explanationderivationrelated formulasrelated definitions
The scalar zero annihilates every vector, and any scalar applied to the zero vector returns the zero vector. Both facts are theorems derived from the axioms — not separate assumptions. They establish that the two distinct "zeros" (scalar 00 in the field, vector 0\mathbf{0} in VV) interact through scalar multiplication in the expected way.
Back to top

Negative One Scalar

(1)v=v(-1)\mathbf{v} = -\mathbf{v}
See details
Back to top
explanationderivationrelated formulasrelated definitions
Scaling a vector by 1-1 produces its additive inverse. This identifies the additive inverse v-\mathbf{v} (introduced by axiom 5) with a specific scalar product, removing any ambiguity about what negation means in a vector space.
Back to top

Subspaces

(2 formulas)

Subspace Test

WV is a subspace    {Wu,vWu+vWvW, cFcvWW \subseteq V \text{ is a subspace} \iff \begin{cases} W \neq \emptyset \\ \mathbf{u}, \mathbf{v} \in W \Rightarrow \mathbf{u} + \mathbf{v} \in W \\ \mathbf{v} \in W,\ c \in \mathbb{F} \Rightarrow c\mathbf{v} \in W \end{cases}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
A nonempty subset of a vector space is a subspace exactly when it is closed under addition and under scalar multiplication. The other axioms (commutativity, associativity, distributivity) hold automatically because vectors in WW are vectors in VV. Closure is the only thing that can fail.
Back to top

Subspace Test Combined

WV is a subspace    W and cu+dvW for all u,vW, c,dFW \subseteq V \text{ is a subspace} \iff W \neq \emptyset \text{ and } c\mathbf{u} + d\mathbf{v} \in W \text{ for all } \mathbf{u}, \mathbf{v} \in W,\ c, d \in \mathbb{F}
See details
Back to top
explanationrelated formulasrelated definitions
The two closure conditions can be compressed into a single closure-under-linear-combinations condition. Setting d=0d = 0 recovers scalar closure; setting c=d=1c = d = 1 recovers additive closure. The combined form is more efficient in proofs and in algorithmic verification.
Back to top

Span

(3 formulas)

Span (Set Definition)

Span{v1,,vk}={c1v1+c2v2++ckvkciF}\text{Span}\{\mathbf{v}_1, \ldots, \mathbf{v}_k\} = \left\{c_1\mathbf{v}_1 + c_2\mathbf{v}_2 + \cdots + c_k\mathbf{v}_k \mid c_i \in \mathbb{F}\right\}
See details
Back to top
explanationrelated formulasrelated definitions
The span of a finite set of vectors is the set of all linear combinations of those vectors. As the scalar coefficients range over F\mathbb{F}, the span sweeps out an entire subspace. By convention, Span={0}\text{Span}\,\emptyset = \{\mathbf{0}\}.
Back to top

Span Membership Criterion

bSpan{v1,,vk}    Ac=b is consistent\mathbf{b} \in \text{Span}\{\mathbf{v}_1, \ldots, \mathbf{v}_k\} \iff A\mathbf{c} = \mathbf{b} \text{ is consistent}
See details
Back to top
explanationnotationrelated formulasrelated definitions
Testing whether b\mathbf{b} lies in the span of given vectors reduces to a linear-system solvability question. Arrange the spanning vectors as columns of a matrix AA; then b\mathbf{b} is in the span iff the system Ac=bA\mathbf{c} = \mathbf{b} has at least one solution. Row-reducing the augmented matrix [Ab][A \mid \mathbf{b}] decides the question: a contradiction row [00d0][0\,\cdots\,0 \mid d \neq 0] means bSpan\mathbf{b} \notin \text{Span}, no contradiction means bSpan\mathbf{b} \in \text{Span}.
Back to top

Span Is Smallest Subspace

Span(K)=W subspaceKWW\text{Span}(K) = \bigcap_{\substack{W \text{ subspace} \\ K \subseteq W}} W
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The span of a set KK is the smallest subspace containing KK. Equivalently, it is the intersection of all subspaces that contain KK — a subspace itself, since intersections of subspaces are subspaces. Any subspace containing KK must also contain every linear combination of vectors in KK, so Span(K)\text{Span}(K) is contained in every such subspace, making it the minimal one.
Back to top

Linear Independence

(5 formulas)

Linear Independence Equation

{v1,,vk} is independent    (c1v1++ckvk=0c1==ck=0)\{\mathbf{v}_1, \ldots, \mathbf{v}_k\} \text{ is independent} \iff \bigl(c_1\mathbf{v}_1 + \cdots + c_k\mathbf{v}_k = \mathbf{0} \Rightarrow c_1 = \cdots = c_k = 0\bigr)
See details
Back to top
explanationconditionsrelated formulasrelated definitions
A set is linearly independent when the only linear combination producing the zero vector is the trivial one — every coefficient must be zero. Any nontrivial relation (some coefficient nonzero) means at least one vector is redundant: it can be expressed as a combination of the others.
Back to top

Linear Independence Matrix Test

{v1,,vk}Rm is independent    Ac=0 has only the trivial solution\{\mathbf{v}_1, \ldots, \mathbf{v}_k\} \subset \mathbb{R}^m \text{ is independent} \iff A\mathbf{c} = \mathbf{0} \text{ has only the trivial solution}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
For column vectors in Rm\mathbb{R}^m, independence is equivalent to triviality of the homogeneous system whose coefficient matrix has those vectors as columns. Row-reduce A=[v1  vk]A = [\mathbf{v}_1\ \cdots\ \mathbf{v}_k]: independence holds iff every column has a pivot (no free variables). If any column is free, a nontrivial null-space vector gives a dependence relation.
Back to top

Linear Independence Determinant Test

{v1,,vn}Rn is independent    det[v1  vn]0\{\mathbf{v}_1, \ldots, \mathbf{v}_n\} \subset \mathbb{R}^n \text{ is independent} \iff \det[\mathbf{v}_1\ \cdots\ \mathbf{v}_n] \neq 0
See details
Back to top
explanationconditionsrelated formulasrelated definitions
When the number of vectors equals the dimension of the ambient space, independence reduces to a single number — the determinant of the matrix whose columns are the vectors. Nonzero determinant means independence; zero determinant means dependence. This follows from the invertible matrix theorem: AA is invertible iff its columns are independent iff det(A)0\det(A) \neq 0.
Back to top

Max Independent Set Size

S>dimVS is dependent|S| > \dim V \Rightarrow S \text{ is dependent}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
In an nn-dimensional vector space, no independent set can have more than nn elements. Any collection of n+1n+1 or more vectors is automatically dependent — independence imposes a hard ceiling at the dimension. Conversely, any independent set with exactly dimV\dim V elements is automatically a basis: the spanning condition comes for free at the magic count.
Back to top

Wronskian Test

W(f1,,fn)(x)=det(f1(x)fn(x)f1(x)fn(x)f1(n1)(x)fn(n1)(x))W(f_1, \ldots, f_n)(x) = \det\begin{pmatrix} f_1(x) & \cdots & f_n(x) \\ f_1'(x) & \cdots & f_n'(x) \\ \vdots & \ddots & \vdots \\ f_1^{(n-1)}(x) & \cdots & f_n^{(n-1)}(x) \end{pmatrix}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The Wronskian is a determinant built from successive derivatives of nn functions, each differentiable at least n1n-1 times. If W(f1,,fn)(x0)0W(f_1, \ldots, f_n)(x_0) \neq 0 at any single point x0x_0, the functions are linearly independent on the entire interval. The Wronskian provides a usable independence test in function spaces, where the column-matrix approach does not apply directly.
Back to top

Basis & Coordinates

(7 formulas)

Basis Definition

B={v1,,vn} is a basis for V    {B is linearly independentSpan(B)=V\mathcal{B} = \{\mathbf{v}_1, \ldots, \mathbf{v}_n\} \text{ is a basis for } V \iff \begin{cases} \mathcal{B} \text{ is linearly independent} \\ \text{Span}(\mathcal{B}) = V \end{cases}
See details
Back to top
explanationrelated formulasrelated definitions
A basis is a set that is independent and spans the space — the two conditions hold simultaneously. Independence guarantees no vector is wasted; spanning guarantees no vector in VV is unreachable. Equivalently, a basis is a maximal independent set, or a minimal spanning set.
Back to top

Unique Basis Representation

vV, !(c1,,cn):v=c1v1+c2v2++cnvn\forall\, \mathbf{v} \in V,\ \exists!\, (c_1, \ldots, c_n): \mathbf{v} = c_1\mathbf{v}_1 + c_2\mathbf{v}_2 + \cdots + c_n\mathbf{v}_n
See details
Back to top
explanationderivationrelated formulasrelated definitions
Every vector in VV admits exactly one representation as a linear combination of basis vectors. Existence follows from the spanning condition, uniqueness from independence: if two coefficient sets both produced v\mathbf{v}, their difference would give a nontrivial relation equal to 0\mathbf{0}. This unique representation is what makes coordinates well-defined.
Back to top

Coordinate Vector

[v]B=(c1c2cn)where v=c1v1++cnvn[\mathbf{v}]_\mathcal{B} = \begin{pmatrix} c_1 \\ c_2 \\ \vdots \\ c_n \end{pmatrix} \quad \text{where } \mathbf{v} = c_1\mathbf{v}_1 + \cdots + c_n\mathbf{v}_n
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The coordinate vector of v\mathbf{v} relative to basis B\mathcal{B} packages the unique expansion coefficients into a column in Fn\mathbb{F}^n. Coordinates depend on the choice of basis — the same vector has different coordinates in different bases. For the standard basis of Rn\mathbb{R}^n, coordinates coincide with components.
Back to top

Standard Basis (Rn)

ei=(0,,0,1i-th,0,,0),i=1,,n\mathbf{e}_i = (0, \ldots, 0, \underset{i\text{-th}}{1}, 0, \ldots, 0), \quad i = 1, \ldots, n
See details
Back to top
explanationnotationrelated formulasrelated definitions
The standard basis of Rn\mathbb{R}^n consists of nn vectors, each with a 11 in one position and zeros elsewhere. Independence is immediate: no vector is a combination of the others. Spanning follows from (v1,,vn)=v1e1++vnen(v_1, \ldots, v_n) = v_1\mathbf{e}_1 + \cdots + v_n\mathbf{e}_n. Coordinates relative to this basis coincide with the components themselves — the reason it is the default choice.
Back to top

Change of Basis Formula

[v]C=PCB[v]B[\mathbf{v}]_\mathcal{C} = P_{\mathcal{C} \leftarrow \mathcal{B}}\, [\mathbf{v}]_\mathcal{B}
See details
Back to top
explanationnotationrelated formulasrelated definitions
Coordinates of the same vector in two different bases are related by left-multiplication by the change-of-basis matrix. The columns of PCBP_{\mathcal{C} \leftarrow \mathcal{B}} are the C\mathcal{C}-coordinate vectors of each B\mathcal{B}-basis vector. Knowing [v]B[\mathbf{v}]_\mathcal{B}, this matrix produces [v]C[\mathbf{v}]_\mathcal{C} in one multiplication.
Back to top

Change of Basis Inverse

PBC=(PCB)1P_{\mathcal{B} \leftarrow \mathcal{C}} = \bigl(P_{\mathcal{C} \leftarrow \mathcal{B}}\bigr)^{-1}
See details
Back to top
explanationrelated formulasrelated definitions
Reversing the direction of a basis change inverts the matrix. The two change-of-basis matrices in opposite directions are inverses of each other, so once one is computed, the other is obtained by matrix inversion. Both matrices are invertible because basis vectors are linearly independent.
Back to top

Coordinate Map Linearity

[u+v]B=[u]B+[v]B,[cv]B=c[v]B[\mathbf{u} + \mathbf{v}]_\mathcal{B} = [\mathbf{u}]_\mathcal{B} + [\mathbf{v}]_\mathcal{B}, \qquad [c\mathbf{v}]_\mathcal{B} = c\,[\mathbf{v}]_\mathcal{B}
See details
Back to top
explanationrelated formulasrelated definitions
The coordinate map v[v]B\mathbf{v} \mapsto [\mathbf{v}]_\mathcal{B} preserves both vector operations. Together with bijectivity, this makes it an isomorphism VFnV \to \mathbb{F}^n. Every nn-dimensional vector space over F\mathbb{F} is therefore structurally identical to Fn\mathbb{F}^n — polynomials, matrices, and ODE solution spaces all reduce to coordinate computations once a basis is fixed.
Back to top

Dimension & Rank

(8 formulas)

Dimension Definition

dim(V)=Bfor any basis B of V\dim(V) = |\mathcal{B}| \quad \text{for any basis } \mathcal{B} \text{ of } V
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The dimension of VV is the number of vectors in any basis. The basis size theorem guarantees that all bases have the same cardinality, so the count is intrinsic to VV, not an artifact of basis choice. By convention, dim{0}=0\dim\{\mathbf{0}\} = 0 (the empty set is a basis for the zero space).
Back to top

Subspace Dimension Inequality

WVdim(W)dim(V),with equality    W=VW \subseteq V \Rightarrow \dim(W) \leq \dim(V), \quad \text{with equality} \iff W = V
See details
Back to top
explanationconditionsrelated formulasrelated definitions
A subspace cannot exceed its parent space in dimension, and matches it only when the subspace is the whole space. Any basis of WW is an independent set in VV, hence has size at most dimV\dim V. Equality means the basis of WW already spans VV, forcing W=VW = V.
Back to top

Dimension Sum Formula

dim(W1+W2)=dim(W1)+dim(W2)dim(W1W2)\dim(W_1 + W_2) = \dim(W_1) + \dim(W_2) - \dim(W_1 \cap W_2)
See details
Back to top
explanationderivationrelated formulasrelated definitions
The dimension of a sum of subspaces is the sum of their dimensions minus the dimension of their intersection — the linear-algebra analogue of inclusion-exclusion for set sizes. Vectors in W1W2W_1 \cap W_2 are counted once in each summand, so the intersection is subtracted to avoid double-counting.
Back to top

Direct Sum Criterion

V=W1W2    V=W1+W2 and W1W2={0}V = W_1 \oplus W_2 \iff V = W_1 + W_2 \text{ and } W_1 \cap W_2 = \{\mathbf{0}\}
See details
Back to top
explanationvariantsrelated formulasrelated definitions
A sum of two subspaces is direct exactly when the intersection is trivial. The trivial-intersection condition is what guarantees uniqueness of the decomposition v=w1+w2\mathbf{v} = \mathbf{w}_1 + \mathbf{w}_2: any two decompositions would differ by a nonzero vector lying in both W1W_1 and W2W_2, contradicting W1W2={0}W_1 \cap W_2 = \{\mathbf{0}\}.
Back to top

Direct Sum Dimension

V=W1W2dim(V)=dim(W1)+dim(W2)V = W_1 \oplus W_2 \Rightarrow \dim(V) = \dim(W_1) + \dim(W_2)
See details
Back to top
explanationderivationrelated formulasrelated definitions
For a direct sum, dimensions add cleanly. The dimension formula for sums reduces to plain addition because the intersection has dimension zero. Conversely, if V=W1+W2V = W_1 + W_2 and dimV=dimW1+dimW2\dim V = \dim W_1 + \dim W_2, the sum is automatically direct.
Back to top

Rank-Nullity Theorem (Matrix Form)

dim(ColA)+dim(NullA)=n\dim(\text{Col}\,A) + \dim(\text{Null}\,A) = n
See details
Back to top
explanationnotationvariantsrelated formulasrelated definitions
For an m×nm \times n matrix AA, the dimensions of the column space and the null space sum to nn — the dimension of the domain Rn\mathbb{R}^n. The nn degrees of freedom in the input split between what survives the map (column space) and what is annihilated (null space). No dimensions are created or destroyed.
Back to top

Four Fundamental Subspaces Dimensions

dim(ColA)=rdim(RowA)=rdim(NullA)=nrdim(NullAT)=mr\begin{aligned} \dim(\text{Col}\,A) &= r & \dim(\text{Row}\,A) &= r \\ \dim(\text{Null}\,A) &= n - r & \dim(\text{Null}\,A^T) &= m - r \end{aligned}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
For an m×nm \times n matrix of rank rr, all four fundamental subspaces have dimensions determined by rr. The column space and row space share dimension rr (despite living in different ambient spaces). The null space takes the remaining nrn - r dimensions of the domain, the left null space takes the remaining mrm - r dimensions of the codomain. Domain dimensions sum to nn, codomain dimensions sum to mm.
Back to top

Row Rank Equals Column Rank

dim(RowA)=dim(ColA)=rank(A)\dim(\text{Row}\,A) = \dim(\text{Col}\,A) = \text{rank}(A)
See details
Back to top
explanationderivationrelated formulasrelated definitions
The row space and column space of any matrix have the same dimension, despite living in different ambient spaces (Rn\mathbb{R}^n and Rm\mathbb{R}^m respectively). This common value is the rank. The result is unexpected — the rows and columns of a matrix encode different information, yet the number of independent rows always equals the number of independent columns.
Back to top

Transpose & Symmetry

(9 formulas)

Transpose Definition

(AT)ij=aji(A^T)_{ij} = a_{ji}
See details
Back to top
explanationrelated formulasrelated definitions
The transpose flips a matrix across its main diagonal, converting rows into columns. An m×nm \times n matrix becomes n×mn \times m, with the (i,j)(i,j) entry of ATA^T taken from the (j,i)(j,i) entry of AA.
Back to top

Transpose Involution

(AT)T=A(A^T)^T = A
See details
Back to top
explanationrelated formulasrelated definitions
Transposing twice returns the original matrix. The transpose is its own inverse operation.
Back to top

Transpose of Sum

(A+B)T=AT+BT(A + B)^T = A^T + B^T
See details
Back to top
explanationconditionsrelated formulasrelated definitions
Transposition distributes over matrix addition. Together with (cA)T=cAT(cA)^T = cA^T, this makes transposition a linear operation on the space of matrices.
Back to top

Transpose of Scalar Multiple

(cA)T=cAT(cA)^T = c\, A^T
See details
Back to top
explanationrelated formulasrelated definitions
Scalar multiplication commutes with transposition. The scalar passes through unchanged.
Back to top

Transpose of Product

(AB)T=BTAT(AB)^T = B^T A^T
See details
Back to top
explanationderivationvariantsrelated formulasrelated definitions
The transpose of a product is the product of the transposes in reversed order. The order reversal is essential — it accommodates the dimension matching that the product requires.
Back to top

Symmetric Matrix Definition

A=AT    aij=aji for all i,jA = A^T \iff a_{ij} = a_{ji} \text{ for all } i, j
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
A symmetric matrix equals its own transpose. The matrix is mirrored across its main diagonal, fully determined by entries on and above the diagonal. Symmetric matrices have all-real eigenvalues and admit orthogonal diagonalization.
Back to top

Skew-Symmetric Matrix Definition

AT=A    aij=ajiA^T = -A \iff a_{ij} = -a_{ji}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
A skew-symmetric matrix negates under transposition. Setting i=ji = j in aii=aiia_{ii} = -a_{ii} forces every diagonal entry to zero. Real skew-symmetric matrices have eigenvalues that are zero or purely imaginary.
Back to top

Symmetric Skew Decomposition

A=12(A+AT)+12(AAT)A = \tfrac{1}{2}(A + A^T) + \tfrac{1}{2}(A - A^T)
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
Every square matrix splits uniquely into a symmetric part 12(A+AT)\tfrac{1}{2}(A + A^T) and a skew-symmetric part 12(AAT)\tfrac{1}{2}(A - A^T). The decomposition mirrors how every function of two variables splits into symmetric and antisymmetric components.
Back to top

Gram Matrix Symmetry

(ATA)T=ATA,(AAT)T=AAT(A^T A)^T = A^T A, \qquad (A A^T)^T = A A^T
See details
Back to top
explanationderivationrelated formulasrelated definitions
For any matrix AA of any shape, the products ATAA^T A and AATA A^T are symmetric. These Gram matrices appear in least squares, the singular value decomposition, and inner-product computations.
Back to top

Special Matrix Types

(14 formulas)

Identity Matrix Definition

In=[δij],δij={1i=j0ijI_n = [\delta_{ij}], \qquad \delta_{ij} = \begin{cases} 1 & i = j \\ 0 & i \neq j \end{cases}
See details
Back to top
explanationnotationrelated formulasrelated definitions
The identity matrix has ones on the main diagonal and zeros elsewhere. The Kronecker delta δij\delta_{ij} is shorthand for this pattern. The subscript nn is dropped when size is clear from context.
Back to top

Identity Matrix Property

AI=IA=AAI = IA = A
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The identity matrix is the multiplicative identity for matrix multiplication. Multiplying any matrix by II on either side returns the original matrix unchanged, provided dimensions are compatible.
Back to top

Diagonal Matrix Definition

D=diag(d1,d2,,dn)    dij=0 for ijD = \operatorname{diag}(d_1, d_2, \ldots, d_n) \iff d_{ij} = 0 \text{ for } i \neq j
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
A diagonal matrix has nonzero entries only on the main diagonal. Diagonal matrices are the easiest to compute with: products, powers, and inverses all reduce to operations on the diagonal entries alone.
Back to top

Diagonal Matrix Power

Dk=diag(d1k,d2k,,dnk)D^k = \operatorname{diag}(d_1^k, d_2^k, \ldots, d_n^k)
See details
Back to top
explanationrelated formulasrelated definitions
Each diagonal entry is raised to the kk-th power independently. This trivial behavior is the principal reason diagonalization is so useful: writing A=PDP1A = PDP^{-1} converts an expensive matrix power into a cheap diagonal power, Ak=PDkP1A^k = PD^kP^{-1}.
Back to top

Diagonal Matrix Determinant

det(diag(d1,,dn))=d1d2dn\det\bigl(\operatorname{diag}(d_1, \ldots, d_n)\bigr) = d_1 d_2 \cdots d_n
See details
Back to top
explanationrelated formulasrelated definitions
The determinant of a diagonal matrix is the product of its diagonal entries. The matrix is invertible precisely when every diagonal entry is nonzero.
Back to top

Triangular Matrix Determinant

det(T)=t11t22tnn\det(T) = t_{11} \, t_{22} \cdots t_{nn}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
For an upper or lower triangular matrix, the determinant is the product of its diagonal entries. The eigenvalues are also the diagonal entries, both readable directly without further computation.
Back to top

Orthogonal Matrix Definition

QTQ=QQT=IQ^T Q = Q Q^T = I
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
An orthogonal matrix has its transpose as its inverse. Equivalently, the columns form an orthonormal set, and so do the rows. Orthogonal matrices preserve lengths and angles — they are the linear isometries.
Back to top

Orthogonal Matrix Determinant

det(Q)=±1\det(Q) = \pm 1
See details
Back to top
explanationderivationrelated formulasrelated definitions
Every orthogonal matrix has determinant +1+1 or 1-1. The value +1+1 corresponds to a rotation and 1-1 to an orientation-reversing transformation involving a reflection.
Back to top

Idempotent Matrix Definition

A2=AA^2 = A
See details
Back to top
explanationconditionsrelated formulasrelated definitions
An idempotent matrix is unchanged by squaring — applying it twice equals applying it once. Idempotent matrices are precisely the projections: they project Rn\mathbb{R}^n onto their column space along their null space. The eigenvalues are restricted to 00 and 11.
Back to top

Idempotent Rank Trace

A2=A    rank(A)=tr(A)A^2 = A \implies \operatorname{rank}(A) = \operatorname{tr}(A)
See details
Back to top
explanationderivationrelated formulasrelated definitions
For an idempotent matrix, the rank equals the trace. Both quantities count the number of eigenvalues equal to 11 — the trace by the eigenvalue-sum identity, the rank as the dimension of the image.
Back to top

Nilpotent Matrix Definition

Ak=Ofor some k1A^k = O \quad \text{for some } k \geq 1
See details
Back to top
explanationconditionsrelated formulasrelated definitions
A nilpotent matrix becomes the zero matrix at some power. The smallest such kk is the index of nilpotency. Every eigenvalue of a nilpotent matrix is zero, forcing both the determinant and the trace to vanish.
Back to top

Neumann Series Nilpotent

Ak=O    (IA)1=I+A+A2++Ak1A^k = O \implies (I - A)^{-1} = I + A + A^2 + \cdots + A^{k-1}
See details
Back to top
explanationderivationrelated formulasrelated definitions
When AA is nilpotent of index kk, IAI - A is invertible with inverse equal to a finite geometric series. The series terminates at the (k1)(k-1)-th term because higher powers are zero.
Back to top

Involutory Matrix Definition

A2=IA^2 = I
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
An involutory matrix is its own inverse. Applying it twice returns every vector to its starting point. The eigenvalues are restricted to ±1\pm 1. Reflections are the prototypical examples: reflecting twice across the same line or plane returns the identity.
Back to top

Cross Product Skew Matrix

a×b=[a]×b,[a]×=(0a3a2a30a1a2a10)\mathbf{a} \times \mathbf{b} = [\mathbf{a}]_\times \mathbf{b}, \qquad [\mathbf{a}]_\times = \begin{pmatrix} 0 & -a_3 & a_2 \\ a_3 & 0 & -a_1 \\ -a_2 & a_1 & 0 \end{pmatrix}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The cross product in R3\mathbb{R}^3 can be expressed as a matrix-vector multiplication. The 3×33 \times 3 skew-symmetric matrix [a]×[\mathbf{a}]_\times, built from the components of a\mathbf{a}, acts on b\mathbf{b} to produce the cross product. This reformulation lets cross-product computations participate in the algebra of matrix products.
Back to top

Inverse

(15 formulas)

Inverse Definition

AA1=A1A=IA A^{-1} = A^{-1} A = I
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
The inverse of a square matrix AA — when it exists — is the unique matrix A1A^{-1} that produces the identity from both sides. A matrix possessing an inverse is called invertible (or nonsingular); a matrix without one is singular. Uniqueness follows from a short associativity argument.
Back to top

Inverse 2x2 Formula

(abcd)1=1adbc(dbca)\begin{pmatrix} a & b \\ c & d \end{pmatrix}^{-1} = \frac{1}{ad - bc}\begin{pmatrix} d & -b \\ -c & a \end{pmatrix}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
For a 2×22 \times 2 matrix, the inverse is obtained by swapping the diagonal entries, negating the off-diagonal entries, and dividing by the determinant. This is the smallest case where the inverse has a simple closed form.
Back to top

Inverse via Adjugate

A1=1det(A)adj(A)A^{-1} = \frac{1}{\det(A)}\, \operatorname{adj}(A)
See details
Back to top
explanationconditionsnotationrelated formulasrelated definitions
When AA is invertible, the inverse equals the adjugate divided by the determinant. The adjugate is the transpose of the cofactor matrix, so each entry of A1A^{-1} is a signed minor of AA divided by det(A)\det(A). The formula is exact and fully symbolic, but expensive — for numerical work, row reduction is far cheaper.
Back to top

Inverse via Row Reduction

[AI]  row ops  [IA1][A \mid I] \;\xrightarrow{\text{row ops}}\; [I \mid A^{-1}]
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
Form the augmented matrix [AI][A \mid I] and apply row operations until the left half becomes the identity. The right half then holds A1A^{-1}. If the left half develops a row of zeros at any stage, AA is singular and no inverse exists.
Back to top

Inverse Involution

(A1)1=A(A^{-1})^{-1} = A
See details
Back to top
explanationconditionsrelated formulasrelated definitions
Inverting twice returns the original matrix. Inversion is its own inverse operation, mirroring the analogous property of transposition.
Back to top

Inverse of Product

(AB)1=B1A1(AB)^{-1} = B^{-1} A^{-1}
See details
Back to top
explanationconditionsderivationvariantsrelated formulasrelated definitions
The inverse of a product is the product of the inverses in reversed order. The reversal mirrors the rule for transpose of a product: to undo "first BB, then AA," undo AA first, then BB.
Back to top

Inverse of Transpose

(AT)1=(A1)T(A^T)^{-1} = (A^{-1})^T
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
Transposing and inverting commute. The two operations can be applied in either order with the same result. A useful corollary: the inverse of a symmetric invertible matrix is symmetric.
Back to top

Inverse of Scalar Multiple

(cA)1=1cA1(cA)^{-1} = \frac{1}{c}\, A^{-1}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
Scaling a matrix by a nonzero scalar scales its inverse by the reciprocal. The scalar passes through inversion in the natural way.
Back to top

Inverse of Power

(Ak)1=(A1)k=Ak(A^k)^{-1} = (A^{-1})^k = A^{-k}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The inverse of a power is the power of the inverse, and both equal the corresponding negative power. With this identity, the exponent laws AjAk=Aj+kA^j A^k = A^{j+k} and (Aj)k=Ajk(A^j)^k = A^{jk} extend to all integers.
Back to top

Determinant of Inverse

det(A1)=1det(A)\det(A^{-1}) = \frac{1}{\det(A)}
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
The determinant of the inverse is the reciprocal of the determinant. The identity provides a quick consistency check on inverse computations.
Back to top

Diagonal Matrix Inverse

diag(d1,,dn)1=diag ⁣(1d1,,1dn)\operatorname{diag}(d_1, \ldots, d_n)^{-1} = \operatorname{diag}\!\left(\frac{1}{d_1}, \ldots, \frac{1}{d_n}\right)
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The inverse of an invertible diagonal matrix is obtained by reciprocating each diagonal entry. The matrix is invertible if and only if every diagonal entry is nonzero.
Back to top

Orthogonal Matrix Inverse

Q1=QTQ^{-1} = Q^T
See details
Back to top
explanationconditionsrelated formulasrelated definitions
For an orthogonal matrix, the inverse equals the transpose. This is the cheapest matrix inverse to compute: no arithmetic is required, only a re-indexing of entries.
Back to top

Solve System via Inverse

Ax=b    x=A1bA\mathbf{x} = \mathbf{b} \implies \mathbf{x} = A^{-1}\mathbf{b}
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
When the coefficient matrix is invertible, the linear system has the unique solution A1bA^{-1}\mathbf{b}. The formula is the matrix analogue of dividing both sides by AA. Computationally, however, row reduction or LU factorization is preferred — computing A1A^{-1} explicitly is roughly three times more expensive and less numerically stable.
Back to top

Invertible Matrix Theorem

$$\begin{aligned}
\text{For } A \in \mathbb{R}^{n \times n}, \text{ the following are equivalent:} & \\
(1)\ A \text{ is invertible} \quad (2)\ \det(A) \neq 0 \quad (3)\ \operatorname{rank}(A) = n & \\
(4)\ \text{columns of } A \text{ are linearly independent} & \\
(5)\ \text{rows of } A \text{ are linearly independent} & \\
(6)\ \text{columns of } A \text{ span } \mathbb{R}^n \quad (7)\ \text{columns form a basis of } \mathbb{R}^n & \\
(8)\ A\mathbf{x} = \mathbf{0} \text{ has only the trivial solution} & \\
(9)\ A\mathbf{x} = \mathbf{b} \text{ has a unique solution for every } \mathbf{b} & \\
(10)\ \operatorname{Null}(A) = \{\mathbf{0}\} & \\
(11)\ \operatorname{rref}(A) = I \quad (12)\ A \text{ is a product of elementary matrices} & \\
(13)\ 0 \text{ is not an eigenvalue of } A &
\end{aligned}$$
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The Invertible Matrix Theorem collects equivalent characterizations of invertibility, each approaching it from a different angle — algebraic, geometric, computational, spectral. Proving any one condition automatically establishes all the others. Checking the determinant is often the fastest hand test; row reduction is preferred for large-scale computation.
Back to top

Singular Matrix Definition

A singular    det(A)=0    rank(A)<nA \text{ singular} \iff \det(A) = 0 \iff \operatorname{rank}(A) < n
See details
Back to top
explanationconditionsrelated formulasrelated definitions
A singular matrix has determinant zero, equivalently rank less than nn, equivalently no inverse. Its columns are linearly dependent, and as a transformation it collapses at least one dimension — its image is a proper subspace of Rn\mathbb{R}^n. Singularity is the negation of invertibility.
Back to top

Rank

(8 formulas)

Rank Bounds

0rank(A)min(m,n)0 \leq \operatorname{rank}(A) \leq \min(m, n)
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
The rank is bounded above by the smaller dimension of the matrix. When equality holds, the matrix has full rank — every row and every column contributes information that no combination of the others can reproduce. The lower bound zero is achieved only by the zero matrix.
Back to top

Rank of Transpose

rank(AT)=rank(A)\operatorname{rank}(A^T) = \operatorname{rank}(A)
See details
Back to top
explanationrelated formulasrelated definitions
Transposition preserves rank. This is a restatement of the deeper theorem that the column rank and row rank of any matrix are equal — transposition swaps the roles of rows and columns but leaves the common value unchanged.
Back to top

Rank Product Inequality

rank(AB)min(rank(A),rank(B))\operatorname{rank}(AB) \leq \min\bigl(\operatorname{rank}(A), \operatorname{rank}(B)\bigr)
See details
Back to top
explanationconditionsrelated formulasrelated definitions
Multiplying two matrices cannot create independent directions that were not already present in both factors. The rank of a product is bounded by the smaller of the two factor ranks.
Back to top

Sylvester Rank Inequality

rank(A)+rank(B)nrank(AB)\operatorname{rank}(A) + \operatorname{rank}(B) - n \leq \operatorname{rank}(AB)
See details
Back to top
explanationconditionsrelated formulasrelated definitions
Sylvester's lower bound on the rank of a product. The rank cannot drop too far below the sum of factor ranks — at most by nn, the inner dimension. Combined with the upper bound, this constrains rank(AB)\operatorname{rank}(AB) from both sides.
Back to top

Rank Sum Inequality

rank(A+B)rank(A)+rank(B)\operatorname{rank}(A + B) \leq \operatorname{rank}(A) + \operatorname{rank}(B)
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The rank of a sum is bounded by the sum of the ranks. Equality holds when the column spaces of AA and BB intersect only at the origin.
Back to top

Rank Invariance Invertible

rank(PAQ)=rank(A)\operatorname{rank}(PAQ) = \operatorname{rank}(A)
See details
Back to top
explanationconditionsrelated formulasrelated definitions
Multiplying AA by invertible matrices on either side preserves rank exactly. Invertible matrices neither collapse dimensions nor create new ones — they reshape the row and column spaces without altering their dimension.
Back to top

Gram Rank Identity

rank(ATA)=rank(AAT)=rank(A)\operatorname{rank}(A^T A) = \operatorname{rank}(A A^T) = \operatorname{rank}(A)
See details
Back to top
explanationderivationrelated formulasrelated definitions
The Gram matrices ATAA^T A and AATA A^T have the same rank as AA. This identity underlies least squares: even when AA is rectangular, ATAA^T A has the same rank, and is invertible precisely when AA has full column rank.
Back to top

Rank-One Outer Product

A=uvT    rank(A)=1(u,v0)A = \mathbf{u}\mathbf{v}^T \implies \operatorname{rank}(A) = 1 \quad (\mathbf{u}, \mathbf{v} \neq \mathbf{0})
See details
Back to top
explanationconditionsrelated formulasrelated definitions
Every rank-one matrix is an outer product uvT\mathbf{u}\mathbf{v}^T of two nonzero vectors. Each column of AA is a scalar multiple of u\mathbf{u}, so the column space is the one-dimensional line through u\mathbf{u}. The rows are scalar multiples of vT\mathbf{v}^T. Rank-one matrices are the building blocks of the outer-product expansion of matrix multiplication and of low-rank approximation.
Back to top

Trace

(11 formulas)

Trace Definition

tr(A)=i=1naii=a11+a22++ann\operatorname{tr}(A) = \sum_{i=1}^{n} a_{ii} = a_{11} + a_{22} + \cdots + a_{nn}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The trace of a square matrix is the sum of its diagonal entries. Off-diagonal entries play no role. Defined only for square matrices, the trace encodes spectral information that is not obvious from its simple definition: it equals the sum of eigenvalues and is invariant under similarity.
Back to top

Trace Linearity

tr(cA+dB)=ctr(A)+dtr(B)\operatorname{tr}(cA + dB) = c\operatorname{tr}(A) + d\operatorname{tr}(B)
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The trace is a linear function from the space of n×nn \times n matrices to the scalars. Both additivity (tr(A+B)=tr(A)+tr(B)\operatorname{tr}(A + B) = \operatorname{tr}(A) + \operatorname{tr}(B)) and scalar homogeneity (tr(cA)=ctr(A)\operatorname{tr}(cA) = c\operatorname{tr}(A)) follow immediately from summing diagonal entries.
Back to top

Trace of Transpose

tr(AT)=tr(A)\operatorname{tr}(A^T) = \operatorname{tr}(A)
See details
Back to top
explanationrelated formulasrelated definitions
Transposition leaves the diagonal entries fixed, so the trace is unaffected.
Back to top

Trace Cyclic Property

tr(AB)=tr(BA)\operatorname{tr}(AB) = \operatorname{tr}(BA)
See details
Back to top
explanationderivationvariantsrelated formulasrelated definitions
The trace is invariant under cyclic permutations of a product. Notably, ABAB and BABA need not have the same dimensions — if AA is m×nm \times n and BB is n×mn \times m, the products are m×mm \times m and n×nn \times n respectively, yet share the same trace.
Back to top

Trace Sum of Eigenvalues

tr(A)=i=1nλi\operatorname{tr}(A) = \sum_{i=1}^{n} \lambda_i
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
The trace equals the sum of the eigenvalues, counted with algebraic multiplicity. This identity links a trivially computable quantity to spectral information that ordinarily requires solving a degree-nn polynomial. The companion identity det(A)=λi\det(A) = \prod \lambda_i relates the determinant to the product of eigenvalues.
Back to top

Trace Similarity Invariance

tr(P1AP)=tr(A)\operatorname{tr}(P^{-1}AP) = \operatorname{tr}(A)
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
Similar matrices have equal trace. The trace is a property of the linear transformation itself, not of any particular matrix representation — the value is independent of the chosen basis.
Back to top

Trace of Commutator

tr(ABBA)=0\operatorname{tr}(AB - BA) = 0
See details
Back to top
explanationderivationrelated formulasrelated definitions
The commutator [A,B]=ABBA[A, B] = AB - BA always has trace zero, regardless of the matrices involved. A consequence: the identity matrix II can never be written as a commutator over R\mathbb{R} or C\mathbb{C}, since tr(I)=n0\operatorname{tr}(I) = n \neq 0.
Back to top

Trace Symmetric Skew Product

tr(SK)=0(ST=S,  KT=K)\operatorname{tr}(SK) = 0 \quad (S^T = S, \; K^T = -K)
See details
Back to top
explanationderivationrelated formulasrelated definitions
The trace of a product of a symmetric matrix and a skew-symmetric matrix vanishes. The identity reflects an orthogonality between symmetric and skew-symmetric subspaces under the Frobenius inner product.
Back to top

Trace Orthonormal Basis

tr(A)=i=1nqiTAqi\operatorname{tr}(A) = \sum_{i=1}^{n} \mathbf{q}_i^T A \mathbf{q}_i
See details
Back to top
explanationconditionsrelated formulasrelated definitions
For any orthonormal basis {q1,,qn}\{\mathbf{q}_1, \ldots, \mathbf{q}_n\} of Rn\mathbb{R}^n, the trace can be computed as the sum of quadratic forms qiTAqi\mathbf{q}_i^T A \mathbf{q}_i. The result is independent of which orthonormal basis is used — another manifestation of similarity invariance, since change of orthonormal basis is an orthogonal similarity.
Back to top

Frobenius Inner Product

A,BF=tr(ATB)=i,jaijbij\langle A, B \rangle_F = \operatorname{tr}(A^T B) = \sum_{i,j} a_{ij} b_{ij}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The trace defines an inner product on the space of matrices. It is the dot product of AA and BB viewed as vectors of n2n^2 entries. It is symmetric, bilinear, and positive definite — bringing geometric concepts (angles, orthogonality, projections) to bear on matrices themselves.
Back to top

Frobenius Norm

AF=tr(ATA)=i,jaij2\|A\|_F = \sqrt{\operatorname{tr}(A^T A)} = \sqrt{\sum_{i,j} a_{ij}^2}
See details
Back to top
explanationrelated formulasrelated definitions
The Frobenius norm measures the total size of a matrix as the square root of the sum of squares of all entries — the matrix analogue of Euclidean length. Induced by the Frobenius inner product, it is one of several common matrix norms (alongside the operator norm and nuclear norm) and is the simplest to compute.
Back to top

Definitions

(4 formulas)

Determinant 2x2

det(abcd)=adbc\det\begin{pmatrix} a & b \\ c & d \end{pmatrix} = ad - bc
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
For a 2×22 \times 2 matrix, the determinant is the product of the main diagonal minus the product of the anti-diagonal. The matrix is invertible exactly when this number is nonzero.
Back to top

Determinant 3x3

det(A)=a11(a22a33a23a32)a12(a21a33a23a31)+a13(a21a32a22a31)\det(A) = a_{11}(a_{22}a_{33} - a_{23}a_{32}) - a_{12}(a_{21}a_{33} - a_{23}a_{31}) + a_{13}(a_{21}a_{32} - a_{22}a_{31})
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
The expansion along the first row: each entry a1ja_{1j} multiplies the 2×22 \times 2 determinant of the submatrix obtained by deleting row 11 and column jj. The signs alternate +,,++, -, +.
Back to top

Determinant Recursive Definition

det(A)={a11n=1j=1n(1)1+ja1jM1jn2\det(A) = \begin{cases} a_{11} & n = 1 \\ \displaystyle\sum_{j=1}^{n} (-1)^{1+j} \, a_{1j} \, M_{1j} & n \geq 2 \end{cases}
See details
Back to top
explanationconditionsnotationrelated formulasrelated definitions
The recursive definition: an n×nn \times n determinant reduces to nn determinants of size (n1)×(n1)(n-1) \times (n-1), each of which reduces further until reaching 1×11 \times 1 matrices. The expansion above uses row 11, but any row or column gives the same value.
Back to top

Determinant Permutation Formula

det(A)=σSnsgn(σ)aσ(1),1aσ(2),2aσ(n),n\det(A) = \sum_{\sigma \in S_n} \operatorname{sgn}(\sigma) \, a_{\sigma(1),1} \, a_{\sigma(2),2} \cdots a_{\sigma(n),n}
See details
Back to top
explanationnotationconditionsvariantsrelated formulasrelated definitions
The closed-form non-recursive definition: sum over all n!n! permutations of {1,2,,n}\{1, 2, \ldots, n\}, weighting each by the permutation's sign and the product of nn entries it selects (one per row and one per column).
Back to top

Cofactor Structure

(4 formulas)

Minor Definition

Mij=det ⁣(A(i,j))M_{ij} = \det\!\left(A^{(i,j)}\right)
See details
Back to top
explanationnotationconditionsrelated formulasrelated definitions
The (i,j)(i,j) minor of AA is the determinant of the (n1)×(n1)(n-1) \times (n-1) submatrix obtained by deleting row ii and column jj. The minor is itself a determinant — a scalar, not a matrix.
Back to top

Cofactor Definition

Cij=(1)i+jMijC_{ij} = (-1)^{i+j} \, M_{ij}
See details
Back to top
explanationrelated formulasrelated definitions
The (i,j)(i,j) cofactor is the signed minor. The factor (1)i+j(-1)^{i+j} produces a checkerboard pattern starting with ++ at position (1,1)(1,1) and alternating from there. The position alone determines the sign — the entries of AA play no role.
Back to top

Cofactor Matrix Definition

cof(A)=[Cij]n×n\operatorname{cof}(A) = \bigl[C_{ij}\bigr]_{n \times n}
See details
Back to top
explanationrelated formulasrelated definitions
The cofactor matrix has entry CijC_{ij} at position (i,j)(i,j). It is not the matrix of minors — the alternating sign factors are already incorporated. Each row of cof(A)\operatorname{cof}(A) contains the cofactors needed for Laplace expansion along the corresponding row of AA, and each column contains those needed for column expansion.
Back to top

Adjugate Definition

adj(A)=cof(A)T\operatorname{adj}(A) = \operatorname{cof}(A)^T
See details
Back to top
explanationrelated formulasrelated definitions
The adjugate (also called the classical adjoint) is the transpose of the cofactor matrix. Equivalently, [adj(A)]ij=Cji[\operatorname{adj}(A)]_{ij} = C_{ji} — the (i,j)(i,j) entry of the adjugate is the (j,i)(j,i) cofactor of AA.
Back to top

Cofactor Expansion

(2 formulas)

Laplace Row Expansion

det(A)=j=1naijCijfor any fixed row i\det(A) = \sum_{j=1}^{n} a_{ij} \, C_{ij} \qquad \text{for any fixed row } i
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
The determinant equals the sum of each entry in row ii multiplied by its cofactor, regardless of which row is chosen. The freedom to pick any row makes the formula practical: a row with many zeros eliminates entire sub-determinants from the sum.
Back to top

Laplace Column Expansion

det(A)=i=1naijCijfor any fixed column j\det(A) = \sum_{i=1}^{n} a_{ij} \, C_{ij} \qquad \text{for any fixed column } j
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
The determinant equals the sum of each entry in column jj multiplied by its cofactor, for any choice of column. Column expansion gives the same result as row expansion — a consequence of det(AT)=det(A)\det(A^T) = \det(A), which lets every column expansion be reinterpreted as a row expansion on the transpose.
Back to top

Row Operation Effects

(3 formulas)

Determinant Row Swap

det(B)=det(A)\det(B) = -\det(A)
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
Swapping two rows of a matrix flips the sign of its determinant. The same rule holds for column swaps, by transpose invariance.
Back to top

Determinant Row Scaling

det(B)=kdet(A)\det(B) = k \, \det(A)
See details
Back to top
explanationconditionsrelated formulasrelated definitions
Multiplying a single row by a scalar kk multiplies the determinant by kk. The same rule applies to columns. A common factor in any one row can be pulled out in front of the determinant — useful for hand simplification before further computation.
Back to top

Determinant Row Addition

det(B)=det(A)\det(B) = \det(A)
See details
Back to top
explanationconditionsrelated formulasrelated definitions
Adding a scalar multiple of one row to a different row leaves the determinant completely unchanged. This is the operation that does the heavy lifting in Gaussian elimination, and it costs nothing in determinant terms — making row reduction the practical method for computing determinants of large matrices.
Back to top

Algebraic Properties

(5 formulas)

Determinant of Transpose

det(AT)=det(A)\det(A^T) = \det(A)
See details
Back to top
explanationderivationrelated formulasrelated definitions
Transposing a matrix does not change its determinant. The practical consequence is that every row-based property of the determinant has a column-based counterpart: column swap flips sign, column scaling scales the determinant, column expansion equals row expansion.
Back to top

Determinant of Product

det(AB)=det(A)det(B)\det(AB) = \det(A) \, \det(B)
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
The determinant of a product equals the product of determinants — one of the most powerful structural facts about determinants. Geometrically, composing linear maps multiplies their volume-scaling factors; algebraically, this identity unlocks corollaries for inverses, powers, and similar matrices.
Back to top

Determinant of Scalar Multiple

det(kA)=kndet(A)\det(kA) = k^n \, \det(A)
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
Scaling the entire matrix by kk scales the determinant by knk^n, not by kk. The factor passes through each of the nn rows independently — a common error is to forget the exponent. Distinct from row scaling, which scales only one row and contributes a single factor of kk.
Back to top

Determinant of Power

det(Ak)=(det(A))k\det(A^k) = \bigl(\det(A)\bigr)^k
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The determinant of a matrix power is the corresponding power of the determinant. Direct corollary of the multiplicative property applied kk times.
Back to top

Determinant of Identity

det(In)=1\det(I_n) = 1
See details
Back to top
explanationrelated formulasrelated definitions
The identity matrix has determinant 11. Follows directly from the diagonal-product rule for triangular matrices: InI_n is diagonal with every diagonal entry equal to 11.
Back to top

Special Determinants

(2 formulas)

Block Triangular Determinant

det(A11A120A22)=det(A11)det(A22)\det\begin{pmatrix} A_{11} & A_{12} \\ 0 & A_{22} \end{pmatrix} = \det(A_{11}) \, \det(A_{22})
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
For a block upper triangular matrix with square diagonal blocks, the determinant factors as the product of the diagonal-block determinants. The off-diagonal block A12A_{12} contributes nothing — only the triangular placement of the zero block matters.
Back to top

Vandermonde Determinant

det(V)=1i<jn(xjxi)\det(V) = \prod_{1 \leq i < j \leq n} (x_j - x_i)
See details
Back to top
explanationnotationconditionsrelated formulasrelated definitions
The determinant of a Vandermonde matrix is the product of all pairwise differences of the nodes. It is nonzero precisely when all nodes are distinct — the algebraic foundation guaranteeing that a polynomial of degree at most n1n-1 is uniquely determined by its values at nn distinct points.
Back to top

Adjugate & Inverse

(1 formula)

Adjugate Identity

Aadj(A)=adj(A)A=det(A)IA \cdot \operatorname{adj}(A) = \operatorname{adj}(A) \cdot A = \det(A) \, I
See details
Back to top
explanationderivationrelated formulasrelated definitions
The product of AA with its adjugate equals the determinant times the identity. This is the structural identity that yields the explicit inverse formula A1=adj(A)/det(A)A^{-1} = \operatorname{adj}(A) / \det(A) whenever det(A)0\det(A) \neq 0, and it holds for every square matrix — invertible or not.
Back to top

Linear Systems

(1 formula)

Cramers Rule

xi=det(Ai)det(A)x_i = \frac{\det(A_i)}{\det(A)}
See details
Back to top
explanationnotationconditionsrelated formulasrelated definitions
For a square linear system Ax=bA\mathbf{x} = \mathbf{b} with nonzero determinant, each component of the solution is a ratio of determinants. The numerator uses a modified version of AA with column ii replaced by the right-hand side. Of theoretical importance — the solution is a rational function of the data — but computationally expensive: n+1n+1 determinant evaluations versus the O(n3)O(n^3) cost of Gaussian elimination.
Back to top

Geometric Interpretation

(5 formulas)

Determinant Signed Area 2D

signed area(u,v)=det(uv)=adbc\text{signed area}(\mathbf{u}, \mathbf{v}) = \det\begin{pmatrix} \mathbf{u} & \mathbf{v} \end{pmatrix} = ad - bc
See details
Back to top
explanationnotationconditionsrelated formulasrelated definitions
The determinant of a 2×22 \times 2 matrix equals the signed area of the parallelogram spanned by its columns. Positive value means the columns are counterclockwise-ordered, negative means clockwise, zero means parallel.
Back to top

Determinant Signed Volume 3D

signed volume(a,b,c)=det(abc)=a(b×c)\text{signed volume}(\mathbf{a}, \mathbf{b}, \mathbf{c}) = \det\begin{pmatrix} \mathbf{a} & \mathbf{b} & \mathbf{c} \end{pmatrix} = \mathbf{a} \cdot (\mathbf{b} \times \mathbf{c})
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The 3×33 \times 3 determinant equals the signed volume of the parallelepiped spanned by its column vectors, identical to the scalar triple product. Positive value means right-handed system, negative means left-handed, zero means coplanar.
Back to top

Determinant Volume Scaling Factor

vol(A(S))=det(A)vol(S)\operatorname{vol}\bigl(A(S)\bigr) = |\det(A)| \cdot \operatorname{vol}(S)
See details
Back to top
explanationnotationconditionsrelated formulasrelated definitions
The linear map xAx\mathbf{x} \mapsto A\mathbf{x} scales every nn-dimensional region by the factor det(A)|\det(A)|. When det(A)>1|\det(A)| > 1 volumes expand; when 0<det(A)<10 < |\det(A)| < 1 they compress; when det(A)=1|\det(A)| = 1 they are preserved (rotations and reflections); when det(A)=0\det(A) = 0 the image collapses to a lower-dimensional subspace.
Back to top

Triangle Area via Determinant

Area=12det(x2x1x3x1y2y1y3y1)\text{Area} = \frac{1}{2} \left| \det\begin{pmatrix} x_2 - x_1 & x_3 - x_1 \\ y_2 - y_1 & y_3 - y_1 \end{pmatrix} \right|
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The area of a triangle with vertices (x1,y1)(x_1, y_1), (x2,y2)(x_2, y_2), (x3,y3)(x_3, y_3) is half the absolute value of the determinant whose columns are the edge vectors from vertex 11 to the other two. The triangle occupies exactly half the parallelogram spanned by these edges.
Back to top

Tetrahedron Volume via Determinant

V=16det(e1e2e3)V = \frac{1}{6} \left| \det\begin{pmatrix} \mathbf{e}_1 & \mathbf{e}_2 & \mathbf{e}_3 \end{pmatrix} \right|
See details
Back to top
explanationnotationconditionsrelated formulasrelated definitions
The volume of a tetrahedron in R3\mathbb{R}^3 equals one-sixth the absolute value of the determinant whose columns are the three edge vectors emanating from any single chosen vertex. The factor 16\frac{1}{6} arises because a tetrahedron occupies one-sixth of the parallelepiped spanned by its three edges.
Back to top

Eigenvalue Connection

(1 formula)

Determinant Product of Eigenvalues

det(A)=λ1λ2λn\det(A) = \lambda_1 \, \lambda_2 \cdots \lambda_n
See details
Back to top
explanationderivationconditionsrelated formulasrelated definitions
The determinant equals the product of all eigenvalues, counted with algebraic multiplicity. Combined with the trace identity tr(A)=λi\operatorname{tr}(A) = \sum \lambda_i, this links the determinant and trace to the eigenvalue spectrum and gives an immediate invertibility criterion: AA is invertible if and only if no eigenvalue is zero.
Back to top

Standard Forms

(4 formulas)

Linear Equation Standard Form

a1x1+a2x2++anxn=ba_1 x_1 + a_2 x_2 + \cdots + a_n x_n = b
See details
Back to top
explanationnotationconditionsrelated formulasrelated definitions
A single linear equation in nn unknowns. Each unknown xix_i appears to the first power and is multiplied by a scalar coefficient aia_i. The right-hand side bb is a constant. No products of unknowns, no powers above one, no transcendental functions.
Back to top

Linear System Matrix Form

Ax=bA\mathbf{x} = \mathbf{b}
See details
Back to top
explanationnotationvariantsrelated formulasrelated definitions
Compresses an entire system of mm equations in nn unknowns into a single matrix equation. Each row of AA encodes one equation; each column corresponds to one unknown. Asking whether the system has a solution becomes asking whether b\mathbf{b} lies in the column space of AA.
Back to top

Vector Equation Form

x1a1+x2a2++xnan=bx_1 \mathbf{a}_1 + x_2 \mathbf{a}_2 + \cdots + x_n \mathbf{a}_n = \mathbf{b}
See details
Back to top
explanationnotationrelated formulasrelated definitions
Recasts the system as a linear combination of the columns of AA, weighted by the entries of x\mathbf{x}. Each aj\mathbf{a}_j is the jj-th column of AA. The system has a solution exactly when b\mathbf{b} can be assembled from the columns — that is, when b\mathbf{b} lies in their span.
Back to top

Augmented Matrix Construction

[Ab]=(a11a12a1nb1a21a22a2nb2am1am2amnbm)[A \mid \mathbf{b}] = \left(\begin{array}{cccc|c} a_{11} & a_{12} & \cdots & a_{1n} & b_1 \\ a_{21} & a_{22} & \cdots & a_{2n} & b_2 \\ \vdots & \vdots & \ddots & \vdots & \vdots \\ a_{m1} & a_{m2} & \cdots & a_{mn} & b_m \end{array}\right)
See details
Back to top
explanationconditionsrelated formulasrelated definitions
Packages the coefficient matrix AA and the right-hand side b\mathbf{b} into a single m×(n+1)m \times (n+1) matrix. The vertical bar is purely notational. Row operations performed on [Ab][A \mid \mathbf{b}] correspond directly to legal manipulations of the underlying equations.
Back to top

Echelon Forms

(4 formulas)

Row Echelon Form Definition

REF: (000000000)\text{REF: } \begin{pmatrix} \boxed{\ast} & \bullet & \bullet & \bullet & \bullet \\ 0 & \boxed{\ast} & \bullet & \bullet & \bullet \\ 0 & 0 & 0 & \boxed{\ast} & \bullet \\ 0 & 0 & 0 & 0 & 0 \end{pmatrix}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
A matrix is in row echelon form if (1) every zero row sits at the bottom, (2) the leading nonzero entry of each row (the pivot, \boxed{\ast}) is strictly to the right of the pivot in the row above, and (3) every entry below a pivot is zero. Bullets \bullet denote arbitrary entries.
Back to top

Reduced Row Echelon Form Definition

RREF: (100010000100000)\text{RREF: } \begin{pmatrix} \boxed{1} & 0 & \bullet & 0 & \bullet \\ 0 & \boxed{1} & \bullet & 0 & \bullet \\ 0 & 0 & 0 & \boxed{1} & \bullet \\ 0 & 0 & 0 & 0 & 0 \end{pmatrix}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
RREF satisfies all REF conditions plus two more: every pivot equals 11, and every pivot is the only nonzero entry in its column (zeros above and below). Each pivot column becomes a unit vector. Free columns (bullets) can contain anything.
Back to top

RREF Uniqueness

RREF(A) is unique\text{RREF}(A) \text{ is unique}
See details
Back to top
explanationconditionsnotationrelated formulasrelated definitions
Every matrix has exactly one reduced row echelon form. No matter which sequence of row operations is used to reach RREF, the result is identical. This makes pivot positions, rank, and free-variable structure intrinsic properties of the matrix.
Back to top

Pivot Definition

pivot=leading nonzero entry of a row in echelon form\text{pivot} = \text{leading nonzero entry of a row in echelon form}
See details
Back to top
explanationnotationrelated formulasrelated definitions
The pivot of a nonzero row in echelon form is its leftmost nonzero entry. The column containing a pivot is a pivot column; all other columns are free columns. The number of pivots equals the rank of the matrix.
Back to top

Elementary Row Operations

(2 formulas)

Elementary Row Operations

RiRj(swap)kRiRi,k0(scaling)Ri+cRjRi(addition)\begin{aligned} R_i &\leftrightarrow R_j \quad \text{(swap)} \\ kR_i &\to R_i, \quad k \neq 0 \quad \text{(scaling)} \\ R_i + cR_j &\to R_i \quad \text{(addition)} \end{aligned}
See details
Back to top
explanationconditionsnotationrelated formulasrelated definitions
The three operations that transform an augmented matrix without altering the solution set. Row swap reorders equations. Row scaling rescales an equation by a nonzero factor. Row addition replaces a row with itself plus a multiple of another row — this is the operation that performs elimination.
Back to top

Row Equivalence Preserves Solutions

[Ab][Ab]    Sol(Ax=b)=Sol(Ax=b)[A \mid \mathbf{b}] \sim [A' \mid \mathbf{b}'] \;\Longrightarrow\; \text{Sol}(A\mathbf{x} = \mathbf{b}) = \text{Sol}(A'\mathbf{x} = \mathbf{b}')
See details
Back to top
explanationconditionsnotationrelated formulasrelated definitions
If two augmented matrices are row-equivalent (one is reachable from the other by a finite sequence of elementary row operations), their associated linear systems have identical solution sets. This is what justifies Gaussian elimination as a solution method — every step preserves the answer.
Back to top

Solvability

(3 formulas)

Free Variables Count

(number of free variables)=nrank(A)\text{(number of free variables)} = n - \text{rank}(A)
See details
Back to top
explanationnotationrelated formulasrelated definitions
In the echelon form of an m×nm \times n coefficient matrix, the number of pivot columns equals rank(A)\text{rank}(A). The remaining nrank(A)n - \text{rank}(A) columns are free, and each contributes one free parameter to the solution. When this count is zero, the solution (if it exists) is unique; when positive, the solution set is infinite.
Back to top

Solvability Rank Criterion

Ax=b is consistent    rank(A)=rank([Ab])A\mathbf{x} = \mathbf{b} \text{ is consistent} \iff \text{rank}(A) = \text{rank}([A \mid \mathbf{b}])
See details
Back to top
explanationconditionsrelated formulasrelated definitions
A linear system has at least one solution if and only if appending b\mathbf{b} as a column to the coefficient matrix does not increase the rank. Equivalently, b\mathbf{b} must lie in the column space of AA. Also known as the Rouché-Capelli theorem (or Kronecker-Capelli theorem).
Back to top

Solution Structure Decomposition

x=xp+xh,xhNull(A)\mathbf{x} = \mathbf{x}_p + \mathbf{x}_h, \quad \mathbf{x}_h \in \text{Null}(A)
See details
Back to top
explanationconditionsnotationrelated formulasrelated definitions
Every solution to a consistent non-homogeneous system Ax=bA\mathbf{x} = \mathbf{b} decomposes into a particular solution xp\mathbf{x}_p plus a homogeneous component xh\mathbf{x}_h from the null space of AA. The particular solution accounts for b\mathbf{b}; the null-space component accounts for the freedom. The full solution set is an affine subspace — the null space translated by xp\mathbf{x}_p.
Back to top

Homogeneous Systems

(2 formulas)

Homogeneous Solution Space Dimension

dimNull(A)=nrank(A)\dim \text{Null}(A) = n - \text{rank}(A)
See details
Back to top
explanationconditionsnotationrelated formulasrelated definitions
The solution set of a homogeneous system Ax=0A\mathbf{x} = \mathbf{0} equals the null space of AA — a subspace of Rn\mathbb{R}^n. Its dimension (the nullity of AA) is nn minus the rank. When the nullity is zero, only the trivial solution exists; when positive, the solution set is a flat through the origin of that dimension.
Back to top

Underdetermined Homogeneous Has Nontrivial

n>m    Ax=0 has a nontrivial solutionn > m \;\Longrightarrow\; A\mathbf{x} = \mathbf{0} \text{ has a nontrivial solution}
See details
Back to top
explanationconditionsnotationrelated formulasrelated definitions
A homogeneous system with more unknowns than equations always has a nonzero solution. The rank of an m×nm \times n matrix cannot exceed mm, so when n>mn > m the rank is strictly less than nn, leaving at least nmn - m free variables.
Back to top

Definition & Properties

(3 formulas)

Linear Transformation Definition

T(cu+dv)=cT(u)+dT(v)T(c\mathbf{u} + d\mathbf{v}) = cT(\mathbf{u}) + dT(\mathbf{v})
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
A function T:VWT: V \to W between vector spaces is linear when it preserves both vector addition and scalar multiplication. The single combined condition above packages both — setting c=d=1c = d = 1 recovers additivity T(u+v)=T(u)+T(v)T(\mathbf{u} + \mathbf{v}) = T(\mathbf{u}) + T(\mathbf{v}), setting d=0d = 0 recovers homogeneity T(cv)=cT(v)T(c\mathbf{v}) = cT(\mathbf{v}). Linearity extends to arbitrary linear combinations: T(civi)=ciT(vi)T(\sum c_i \mathbf{v}_i) = \sum c_i T(\mathbf{v}_i).
Back to top

Zero Vector Preservation

T(0V)=0WT(\mathbf{0}_V) = \mathbf{0}_W
See details
Back to top
explanationconditionsderivationvariantsrelated formulasrelated definitions
Every linear transformation sends the zero vector of the domain to the zero vector of the codomain. This is the fastest necessary check for linearity: if T(0)0T(\mathbf{0}) \neq \mathbf{0}, the function cannot be linear. Translations T(v)=v+bT(\mathbf{v}) = \mathbf{v} + \mathbf{b} with b0\mathbf{b} \neq \mathbf{0} fail this test immediately.
Back to top

Composition Is Linear

(ST)(u)=S(T(u)),[ST]=[S][T](S \circ T)(\mathbf{u}) = S(T(\mathbf{u})), \qquad [S \circ T] = [S]\,[T]
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
The composition of two linear transformations is itself linear. If T:UVT: U \to V has matrix AA and S:VWS: V \to W has matrix BB, then ST:UWS \circ T: U \to W has matrix BABA. The matrix-multiplication order matches the composition order: SS acts after TT, so BB multiplies on the left. This is the structural reason matrix multiplication is defined as it is — it encodes function composition exactly.
Back to top

Matrix Representation

(3 formulas)

Standard Matrix

A=[T(e1)    T(e2)        T(en)]A = \bigl[\,T(\mathbf{e}_1) \;\; T(\mathbf{e}_2) \;\; \cdots \;\; T(\mathbf{e}_n)\,\bigr]
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
For a linear transformation T:RnRmT: \mathbb{R}^n \to \mathbb{R}^m, the standard matrix has the images of the standard basis vectors as its columns. Column jj is T(ej)T(\mathbf{e}_j) — the image of the jj-th standard basis vector. Together with Linear Map as Matrix Multiplication, this gives a one-to-one correspondence between linear maps RnRm\mathbb{R}^n \to \mathbb{R}^m and m×nm \times n matrices.
Back to top

Linear Map as Matrix Multiplication

T(x)=Axfor every xRnT(\mathbf{x}) = A\mathbf{x} \quad \text{for every } \mathbf{x} \in \mathbb{R}^n
See details
Back to top
explanationconditionsnotationrelated formulasrelated definitions
Every linear transformation T:RnRmT: \mathbb{R}^n \to \mathbb{R}^m is matrix multiplication by a unique m×nm \times n matrix AA. This is not an optional representation — it is forced by linearity. Conversely, every m×nm \times n matrix defines a linear transformation. Linear maps and matrices are the same objects viewed from two perspectives.
Back to top

Matrix Representation Abstract Bases

[T]CB=[[T(v1)]C        [T(vn)]C][T]_{\mathcal{C} \leftarrow \mathcal{B}} = \bigl[\,[T(\mathbf{v}_1)]_{\mathcal{C}} \;\; \cdots \;\; [T(\mathbf{v}_n)]_{\mathcal{C}}\,\bigr]
See details
Back to top
explanationconditionsnotationrelated formulasrelated definitions
For a linear transformation T:VWT: V \to W between abstract vector spaces, the matrix depends on a choice of basis B={v1,,vn}\mathcal{B} = \{\mathbf{v}_1, \ldots, \mathbf{v}_n\} for VV and basis C\mathcal{C} for WW. Column jj is the C\mathcal{C}-coordinate vector of T(vj)T(\mathbf{v}_j) — the scalars that express T(vj)T(\mathbf{v}_j) as a linear combination of C\mathcal{C}-basis vectors. The standard matrix is the special case where both bases are standard.
Back to top

Image & Kernel

(5 formulas)

Image Definition

Im(T)={T(v):vV}W\text{Im}(T) = \{T(\mathbf{v}) : \mathbf{v} \in V\} \subseteq W
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
The image (or range) of a linear transformation is the set of all output vectors. It is a subspace of the codomain WW. For matrix transformations T(x)=AxT(\mathbf{x}) = A\mathbf{x}, the image is the column space of AA, and its dimension is rank(A)\text{rank}(A). The image answers the reachability question: wIm(T)\mathbf{w} \in \text{Im}(T) iff Ax=wA\mathbf{x} = \mathbf{w} has a solution.
Back to top

Kernel Definition

ker(T)={vV:T(v)=0}V\ker(T) = \{\mathbf{v} \in V : T(\mathbf{v}) = \mathbf{0}\} \subseteq V
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
The kernel (or null space) of a linear transformation is the set of inputs that map to zero. It is a subspace of the domain VV. For matrix transformations T(x)=AxT(\mathbf{x}) = A\mathbf{x}, the kernel is the null space of AA — all solutions to the homogeneous system Ax=0A\mathbf{x} = \mathbf{0}. The kernel measures the information lost by TT.
Back to top

Injectivity Kernel Criterion

T injective    ker(T)={0}T \text{ injective} \iff \ker(T) = \{\mathbf{0}\}
See details
Back to top
explanationderivationvariantsrelated formulasrelated definitions
For linear transformations, injectivity reduces to a single check on the kernel. Equivalently for matrix transformations: T(x)=AxT(\mathbf{x}) = A\mathbf{x} is injective iff AA has full column rank, iff every column is a pivot column, iff the columns are linearly independent.
Back to top

Rank-Nullity for Maps

dimIm(T)+dimker(T)=dimV\dim\text{Im}(T) + \dim\ker(T) = \dim V
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
For a linear transformation T:VWT: V \to W with VV finite-dimensional, the dimensions of the image and kernel sum to the dimension of the domain. The dimV\dim V degrees of freedom split between what survives the map (image) and what is annihilated (kernel). This is the abstract version of the matrix rank-nullity theorem.
Back to top

Bijectivity Equal Dim Case

dimV=dimW(T injective    T surjective    T bijective)\dim V = \dim W \Rightarrow \bigl(T \text{ injective} \iff T \text{ surjective} \iff T \text{ bijective}\bigr)
See details
Back to top
explanationderivationvariantsrelated formulasrelated definitions
When the domain and codomain have the same finite dimension, the three conditions collapse — verifying any one establishes the others. A bijective linear transformation is an isomorphism; the two spaces are structurally identical as vector spaces. For square matrices, bijectivity corresponds exactly to invertibility.
Back to top

Similarity & Basis Change

(3 formulas)

Similarity Relation

A=P1APA' = P^{-1} A P
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
If a linear operator T:VVT: V \to V has matrix AA in basis B\mathcal{B} and matrix AA' in basis C\mathcal{C}, then AA and AA' are related by similarity via the change-of-basis matrix P=PCBP = P_{\mathcal{C} \leftarrow \mathcal{B}}. Two matrices satisfying this relation for some invertible PP are called similar — they represent the same linear transformation in different coordinate systems.
Back to top

Similarity Invariants

A=P1AP{det(A)=det(A)tr(A)=tr(A)rank(A)=rank(A)eigenvalues(A)=eigenvalues(A)A' = P^{-1}AP \Rightarrow \begin{cases} \det(A') = \det(A) \\ \text{tr}(A') = \text{tr}(A) \\ \text{rank}(A') = \text{rank}(A) \\ \text{eigenvalues}(A') = \text{eigenvalues}(A) \end{cases}
See details
Back to top
explanationderivationvariantsrelated formulasrelated definitions
Similar matrices share every property intrinsic to the underlying linear transformation, but not properties tied to a specific coordinate representation. Determinant, trace, rank, eigenvalues (with multiplicities), and the characteristic polynomial are all preserved. Individual matrix entries, symmetry, and sparsity are generally not preserved — a symmetric AA can become non-symmetric under P1APP^{-1}AP unless PP is orthogonal.
Back to top

Diagonalization Formula

A=PDP1,D=diag(λ1,,λn),P=[v1    vn]A = P D P^{-1}, \quad D = \operatorname{diag}(\lambda_1, \ldots, \lambda_n), \quad P = [\mathbf{v}_1 \;\cdots\; \mathbf{v}_n]
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
When a linear operator has nn linearly independent eigenvectors, using them as a basis makes the operator's matrix diagonal: T(vi)=λiviT(\mathbf{v}_i) = \lambda_i \mathbf{v}_i. The matrix PP has the eigenvectors as columns, DD is the diagonal matrix of corresponding eigenvalues, and similarity gives A=PDP1A = PDP^{-1}. Diagonalization reduces matrix powers and exponentials to trivial diagonal operations.
Back to top

Geometric Transformations

(7 formulas)

Rotation Matrix 2D

Rθ=(cosθsinθsinθcosθ)R_\theta = \begin{pmatrix} \cos\theta & -\sin\theta \\ \sin\theta & \cos\theta \end{pmatrix}
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
Counterclockwise rotation by angle θ\theta about the origin in R2\mathbb{R}^2. The first column (cosθ,sinθ)(\cos\theta, \sin\theta) is the image of e1\mathbf{e}_1 — the point on the unit circle at angle θ\theta. The second column is the image of e2\mathbf{e}_2 — the point at angle θ+90°\theta + 90°. Determinant cos2θ+sin2θ=1\cos^2\theta + \sin^2\theta = 1 confirms orientation- and area-preserving.
Back to top

Rotation Matrices 3D

Rx(θ)=(1000cosθsinθ0sinθcosθ),    Ry(θ)=(cosθ0sinθ010sinθ0cosθ),    Rz(θ)=(cosθsinθ0sinθcosθ0001)R_x(\theta) = \begin{pmatrix} 1 & 0 & 0 \\ 0 & \cos\theta & -\sin\theta \\ 0 & \sin\theta & \cos\theta \end{pmatrix}, \;\; R_y(\theta) = \begin{pmatrix} \cos\theta & 0 & \sin\theta \\ 0 & 1 & 0 \\ -\sin\theta & 0 & \cos\theta \end{pmatrix}, \;\; R_z(\theta) = \begin{pmatrix} \cos\theta & -\sin\theta & 0 \\ \sin\theta & \cos\theta & 0 \\ 0 & 0 & 1 \end{pmatrix}
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
Rotations in R3\mathbb{R}^3 about each coordinate axis. The fixed axis appears as a 11 on the diagonal; the other two coordinates form a 2×22 \times 2 rotation block. The axis of rotation is the eigenvector with eigenvalue 11 — the direction left unchanged. Every 3×33 \times 3 rotation is orthogonal with determinant +1+1.
Back to top

Reflection Across Line 2D

Hα=(cos2αsin2αsin2αcos2α)H_\alpha = \begin{pmatrix} \cos 2\alpha & \sin 2\alpha \\ \sin 2\alpha & -\cos 2\alpha \end{pmatrix}
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
Reflection across the line through the origin at angle α\alpha from the positive xx-axis. Determinant cos22αsin22α=1-\cos^2 2\alpha - \sin^2 2\alpha = -1 (orientation-reversing). The matrix is orthogonal and involutory: Hα2=IH_\alpha^2 = I — reflecting twice returns every vector. Eigenvalues are +1+1 (vectors along the mirror line) and 1-1 (vectors perpendicular to it).
Back to top

Householder Reflection

H=I2nnTH = I - 2\,\mathbf{n}\mathbf{n}^T
See details
Back to top
explanationconditionsderivationvariantsrelated formulasrelated definitions
Reflection across the hyperplane through the origin with unit normal n\mathbf{n}. The matrix subtracts twice the component of each vector in the direction of n\mathbf{n}, effectively mirroring across the perpendicular plane. Householder reflections work in any dimension and are the building blocks of QR decomposition.
Back to top

Projection onto Line

P=uuTuTuP = \frac{\mathbf{u}\mathbf{u}^T}{\mathbf{u}^T\mathbf{u}}
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
Orthogonal projection onto the line through the origin in direction u\mathbf{u}. The matrix has rank 11 (image is the line spanned by u\mathbf{u}). The outer product uuT\mathbf{u}\mathbf{u}^T is divided by uTu=u2\mathbf{u}^T\mathbf{u} = \|\mathbf{u}\|^2 to normalize. When u\mathbf{u} is a unit vector, the formula simplifies to P=uuTP = \mathbf{u}\mathbf{u}^T.
Back to top

Projection onto Plane

P=InnTnTnP = I - \frac{\mathbf{n}\mathbf{n}^T}{\mathbf{n}^T\mathbf{n}}
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
Orthogonal projection onto the plane through the origin with normal vector n\mathbf{n}. The formula subtracts the component along n\mathbf{n} from each input, leaving only the perpendicular component (which lies in the plane). Closely related to the Householder reflection — projection subtracts the n\mathbf{n}-component once; reflection subtracts it twice.
Back to top

Shear Matrix

Shearx=(1k01),Sheary=(10k1)\text{Shear}_x = \begin{pmatrix} 1 & k \\ 0 & 1 \end{pmatrix}, \qquad \text{Shear}_y = \begin{pmatrix} 1 & 0 \\ k & 1 \end{pmatrix}
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
A shear displaces each point in proportion to its distance from a fixed line. The horizontal shear Shearx\text{Shear}_x shifts the xx-coordinate by kk times the yy-coordinate; the vertical shear Sheary\text{Shear}_y shifts the yy-coordinate by kk times the xx-coordinate. Both are triangular with determinant 11 — area-preserving and orientation-preserving — but not orthogonal: angles are distorted.
Back to top

Foundation

(3 formulas)

Eigenvalue Definition

Av=λv,v0A\mathbf{v} = \lambda\mathbf{v}, \quad \mathbf{v} \neq \mathbf{0}
See details
Back to top
explanationconditionsnotationrelated formulasrelated definitions
For a square matrix AA, a nonzero vector v\mathbf{v} is an eigenvector if AvA\mathbf{v} is a scalar multiple of v\mathbf{v}. The scalar λ\lambda is the corresponding eigenvalue. Geometrically, eigenvectors are the directions that the linear transformation xAx\mathbf{x} \mapsto A\mathbf{x} preserves — it stretches, compresses, or reverses along these directions without deflecting them.
Back to top

Characteristic Equation

det(AλI)=0\det(A - \lambda I) = 0
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
The eigenvalue equation Av=λvA\mathbf{v} = \lambda\mathbf{v} rearranges to (AλI)v=0(A - \lambda I)\mathbf{v} = \mathbf{0}, a homogeneous system. Nontrivial solutions exist iff AλIA - \lambda I is singular — iff its determinant vanishes. This converts the geometric eigenvalue question ("which directions are preserved?") into the algebraic one ("which λ\lambda make this determinant zero?"). The eigenvalues are exactly the roots.
Back to top

Eigenspace

Eλ=Null(AλI)={v:Av=λv}E_\lambda = \text{Null}(A - \lambda I) = \{\mathbf{v} : A\mathbf{v} = \lambda\mathbf{v}\}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The eigenspace of eigenvalue λ\lambda is the set of all vectors v\mathbf{v} satisfying Av=λvA\mathbf{v} = \lambda\mathbf{v}, including the zero vector. It equals the null space of AλIA - \lambda I and is a subspace of Rn\mathbb{R}^n. Any linear combination of eigenvectors sharing the same eigenvalue is again an eigenvector for that eigenvalue (or zero). The dimension of EλE_\lambda is the geometric multiplicity mg(λ)m_g(\lambda).
Back to top

Characteristic Polynomial

(3 formulas)

Characteristic Polynomial

p(λ)=det(AλI)p(\lambda) = \det(A - \lambda I)
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
The characteristic polynomial of an n×nn \times n matrix is a polynomial of degree nn in λ\lambda. Its roots are exactly the eigenvalues. The polynomial packages the entire eigenvalue spectrum into a single algebraic expression — its leading term is (1)nλn(-1)^n \lambda^n, its constant term is p(0)=det(A)p(0) = \det(A), and the coefficient of λn1\lambda^{n-1} is (1)n1tr(A)(-1)^{n-1}\text{tr}(A).
Back to top

Characteristic Polynomial 2x2

p(λ)=λ2tr(A)λ+det(A)p(\lambda) = \lambda^2 - \text{tr}(A)\,\lambda + \det(A)
See details
Back to top
explanationderivationrelated formulasrelated definitions
For a 2×22 \times 2 matrix, the characteristic polynomial collapses to a quadratic in trace and determinant. Eigenvalues follow from the quadratic formula: λ=(tr(A)±tr(A)24det(A))/2\lambda = \bigl(\text{tr}(A) \pm \sqrt{\text{tr}(A)^2 - 4\det(A)}\bigr)/2. The discriminant classifies the eigenvalue type — see Discriminant Classification 2x2.
Back to top

Cayley-Hamilton

p(A)=Op(A) = O
See details
Back to top
explanationvariantsderivationrelated formulasrelated definitions
Every square matrix satisfies its own characteristic polynomial. Substituting AA for λ\lambda in p(λ)p(\lambda) (with constants multiplied by II) yields the zero matrix. The theorem provides a recurrence reducing any power AkA^k with knk \geq n to a polynomial in AA of degree at most n1n-1, and expresses A1A^{-1} (when it exists) as a polynomial in AA.
Back to top

Multiplicities

(2 formulas)

Multiplicity Inequality

1mg(λ)ma(λ)1 \leq m_g(\lambda) \leq m_a(\lambda)
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
Every eigenvalue has two multiplicities: algebraic (mam_a, the root multiplicity in the characteristic polynomial) and geometric (mgm_g, the dimension of the eigenspace). The geometric multiplicity is at least 11 (eigenspaces contain a nonzero eigenvector by definition) and at most the algebraic multiplicity. Equality across all eigenvalues is the diagonalizability condition.
Back to top

Independence of Distinct Eigenvectors

λ1,,λk distinct{v1,,vk} linearly independent\lambda_1, \ldots, \lambda_k \text{ distinct} \Rightarrow \{\mathbf{v}_1, \ldots, \mathbf{v}_k\} \text{ linearly independent}
See details
Back to top
explanationderivationrelated formulasrelated definitions
Eigenvectors corresponding to distinct eigenvalues are always linearly independent. This is the key structural fact making distinct-eigenvalue matrices automatically diagonalizable. The proof is by induction: from civi=0\sum c_i \mathbf{v}_i = \mathbf{0}, multiply by AA and subtract λk\lambda_k times the original to eliminate vk\mathbf{v}_k, then apply the induction hypothesis.
Back to top

Eigenvalue Algebra

(4 formulas)

Eigenvalue of Power

Av=λvAkv=λkvA\mathbf{v} = \lambda\mathbf{v} \Rightarrow A^k\mathbf{v} = \lambda^k\mathbf{v}
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
Raising AA to a power raises every eigenvalue to that power, while preserving the eigenvectors. The eigenvector basis is invariant under power operations — only the scaling factors change. Combined with diagonalization, this gives the cheap-power formula Ak=PDkP1A^k = PD^kP^{-1}.
Back to top

Eigenvalue of Inverse

Av=λvA1v=1λvA\mathbf{v} = \lambda\mathbf{v} \Rightarrow A^{-1}\mathbf{v} = \frac{1}{\lambda}\mathbf{v}
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
The eigenvalues of A1A^{-1} are the reciprocals of the eigenvalues of AA, with the same eigenvectors. Invertibility of AA guarantees λ0\lambda \neq 0, so reciprocation is always defined.
Back to top

Eigenvalue of Polynomial

q(A)v=q(λ)vq(A)\mathbf{v} = q(\lambda)\mathbf{v}
See details
Back to top
explanationconditionsderivationvariantsrelated formulasrelated definitions
For any polynomial q(t)=c0+c1t++cmtmq(t) = c_0 + c_1 t + \cdots + c_m t^m, the matrix polynomial q(A)=c0I+c1A++cmAmq(A) = c_0 I + c_1 A + \cdots + c_m A^m has eigenvalues q(λi)q(\lambda_i) with the same eigenvectors. This generalizes the power and shift identities into a single law: eigenvalues transform by qq while eigenvectors stay fixed.
Back to top

Eigenvalue Shift

Av=λv(A+cI)v=(λ+c)vA\mathbf{v} = \lambda\mathbf{v} \Rightarrow (A + cI)\mathbf{v} = (\lambda + c)\mathbf{v}
See details
Back to top
explanationderivationvariantsrelated formulasrelated definitions
Adding cIcI to AA shifts every eigenvalue by cc while leaving eigenvectors unchanged. Useful for making a matrix positive definite (shifting all eigenvalues positive) or for shifting a known eigenvalue to zero (the eigenvalue equation (Aλ0I)v=0(A - \lambda_0 I)\mathbf{v} = \mathbf{0} is exactly this construction).
Back to top

Diagonalizability

(2 formulas)

Diagonalizability Condition

A diagonalizable    mg(λ)=ma(λ) for every eigenvalue λA \text{ diagonalizable} \iff m_g(\lambda) = m_a(\lambda) \text{ for every eigenvalue } \lambda
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
A matrix is diagonalizable iff its eigenvectors span Rn\mathbb{R}^n — equivalently, iff each eigenvalue's geometric multiplicity matches its algebraic multiplicity. When the geometric multiplicity falls short for any eigenvalue, the matrix is defective: there are not enough eigenvectors to form a basis, and the best achievable form is the Jordan normal form rather than a diagonal matrix.
Back to top

Distinct Eigenvalues Imply Diagonalizable

A has n distinct eigenvaluesA is diagonalizableA \text{ has } n \text{ distinct eigenvalues} \Rightarrow A \text{ is diagonalizable}
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
An n×nn \times n matrix with nn distinct eigenvalues is automatically diagonalizable. By Independence of Distinct Eigenvectors, the nn eigenvectors are linearly independent, providing exactly enough vectors for an eigenvector basis. The converse fails — diagonalizable matrices can have repeated eigenvalues (e.g., cIcI).
Back to top

Special Spectra

(1 formula)

Special Matrix Eigenvalue Restrictions

symmetric (real):  λRskew-symmetric (real):  λ=0 or λiRorthogonal:  λ=1idempotent:  λ{0,1}nilpotent:  λ=0involutory:  λ{1,+1}positive definite:  λ>0\begin{aligned} \text{symmetric (real)}: \;& \lambda \in \mathbb{R} \\ \text{skew-symmetric (real)}: \;& \lambda = 0 \text{ or } \lambda \in i\mathbb{R} \\ \text{orthogonal}: \;& |\lambda| = 1 \\ \text{idempotent}: \;& \lambda \in \{0, 1\} \\ \text{nilpotent}: \;& \lambda = 0 \\ \text{involutory}: \;& \lambda \in \{-1, +1\} \\ \text{positive definite}: \;& \lambda > 0 \end{aligned}
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
Structural properties of a matrix restrict its eigenvalue spectrum. Many follow directly from the defining equation: if A2=AA^2 = A then λ2=λ\lambda^2 = \lambda, forcing λ{0,1}\lambda \in \{0, 1\}; if Ak=OA^k = O then λk=0\lambda^k = 0, forcing λ=0\lambda = 0; if A2=IA^2 = I then λ2=1\lambda^2 = 1, forcing λ=±1\lambda = \pm 1. For symmetric matrices, the real-eigenvalue property is deeper (Spectral Theorem). For orthogonal matrices, Qv=v\|Q\mathbf{v}\| = \|\mathbf{v}\| forces λ=1|\lambda| = 1.
Back to top

Spectral

(2 formulas)

Spectral Theorem

A=ATA=QDQT,QTQ=I,D=diag(λ1,,λn)Rn×nA = A^T \Rightarrow A = Q D Q^T, \quad Q^T Q = I, \quad D = \text{diag}(\lambda_1, \ldots, \lambda_n) \in \mathbb{R}^{n \times n}
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
Every real symmetric matrix is orthogonally diagonalizable: the diagonalizing matrix QQ can be chosen with orthonormal columns of eigenvectors, and the diagonal DD contains real eigenvalues. This is stronger than ordinary diagonalizability — it guarantees real eigenvalues, mutually orthogonal eigenvectors, and a numerically stable diagonalization (since Q1=QTQ^{-1} = Q^T).
Back to top

Spectral Decomposition

A=i=1nλiqiqiTA = \sum_{i=1}^{n} \lambda_i \, \mathbf{q}_i \mathbf{q}_i^T
See details
Back to top
explanationconditionsderivationvariantsrelated formulasrelated definitions
The spectral theorem A=QDQTA = QDQ^T expands into a sum of rank-one projections: each qiqiT\mathbf{q}_i\mathbf{q}_i^T is the orthogonal projection onto the eigenspace direction qi\mathbf{q}_i, weighted by eigenvalue λi\lambda_i. This decomposition makes the matrix's action transparent — every input is projected onto each eigendirection, scaled by the corresponding eigenvalue, and summed.
Back to top

Applications

(1 formula)

Matrix Exponential

eAt=PeDtP1=Pdiag(eλ1t,,eλnt)P1e^{At} = P\, e^{Dt}\, P^{-1} = P \,\text{diag}(e^{\lambda_1 t}, \ldots, e^{\lambda_n t})\, P^{-1}
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
For a diagonalizable matrix, the matrix exponential reduces to scalar exponentials of the eigenvalues. The solution to x=Ax\mathbf{x}' = A\mathbf{x} with x(0)=x0\mathbf{x}(0) = \mathbf{x}_0 is x(t)=eAtx0\mathbf{x}(t) = e^{At}\mathbf{x}_0, generalizing the scalar x(t)=eatx0x(t) = e^{at}x_0. The general defining series eAt=k=0(At)k/k!e^{At} = \sum_{k=0}^\infty (At)^k/k! converges for any square matrix, but the explicit formula above only applies in the diagonalizable case.
Back to top

Complex

(3 formulas)

Complex Conjugate Pairs

ARn×n,    Av=λvAvˉ=λˉvˉA \in \mathbb{R}^{n \times n}, \;\; A\mathbf{v} = \lambda\mathbf{v} \Rightarrow A\bar{\mathbf{v}} = \bar{\lambda}\bar{\mathbf{v}}
See details
Back to top
explanationderivationvariantsrelated formulasrelated definitions
Complex eigenvalues of a real matrix always come in conjugate pairs, with conjugate eigenvectors. The proof leverages that the characteristic polynomial has real coefficients: complex roots of a real polynomial pair with their conjugates. One consequence: every odd-dimensional real matrix has at least one real eigenvalue.
Back to top

Discriminant Classification 2x2

Δ=tr(A)24det(A){Δ>0:two distinct real eigenvaluesΔ=0:one repeated real eigenvalueΔ<0:complex conjugate pair\Delta = \text{tr}(A)^2 - 4\det(A) \quad \begin{cases} \Delta > 0: & \text{two distinct real eigenvalues} \\ \Delta = 0: & \text{one repeated real eigenvalue} \\ \Delta < 0: & \text{complex conjugate pair} \end{cases}
See details
Back to top
explanationderivationvariantsrelated formulasrelated definitions
For a 2×22 \times 2 real matrix, the discriminant of the characteristic polynomial λ2tr(A)λ+det(A)\lambda^2 - \text{tr}(A)\lambda + \det(A) classifies the eigenvalue structure. Negative discriminant means no real direction is preserved — the transformation rotates rather than purely stretching.
Back to top

Real Canonical Form 2x2

λ=a±biP1AP=(abba)=r(cosθsinθsinθcosθ)\lambda = a \pm bi \Rightarrow P^{-1}AP = \begin{pmatrix} a & -b \\ b & a \end{pmatrix} = r\begin{pmatrix} \cos\theta & -\sin\theta \\ \sin\theta & \cos\theta \end{pmatrix}
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
A 2×22 \times 2 real matrix with complex eigenvalues a±bia \pm bi is similar (over R\mathbb{R}) to a rotation-scaling matrix. The complex eigenvector v=u+iw\mathbf{v} = \mathbf{u} + i\mathbf{w} contributes its real and imaginary parts as columns of P=[u    w]P = [\mathbf{u} \;\; \mathbf{w}]. The transformation rotates by θ=arctan(b/a)\theta = \arctan(b/a) and scales by r=a2+b2r = \sqrt{a^2 + b^2}.
Back to top

Inner Product

(5 formulas)

Cauchy-Schwarz Inequality (General)

uvuv|\mathbf{u} \cdot \mathbf{v}| \leq \|\mathbf{u}\| \, \|\mathbf{v}\|
See details
Back to top
explanationderivationvariantsrelated formulasrelated definitions
The absolute value of the dot product never exceeds the product of the norms. Equality holds iff the vectors are parallel (one is a scalar multiple of the other). The inequality is what makes the angle formula cosθ=(uv)/(uv)\cos\theta = (\mathbf{u}\cdot\mathbf{v})/(\|\mathbf{u}\|\|\mathbf{v}\|) legitimate — it guarantees the right-hand side stays in [1,1][-1, 1].
Back to top

Triangle Inequality (Inner Product)

u+vu+v\|\mathbf{u} + \mathbf{v}\| \leq \|\mathbf{u}\| + \|\mathbf{v}\|
See details
Back to top
explanationderivationvariantsrelated formulasrelated definitions
The length of a sum is bounded by the sum of lengths — one side of a triangle never exceeds the sum of the other two. Equality holds iff u\mathbf{u} and v\mathbf{v} point in the same direction (one is a non-negative scalar multiple of the other). The inequality is what makes the distance function a valid metric.
Back to top

Pythagorean Theorem

uv=0u+v2=u2+v2\mathbf{u} \cdot \mathbf{v} = 0 \Rightarrow \|\mathbf{u} + \mathbf{v}\|^2 = \|\mathbf{u}\|^2 + \|\mathbf{v}\|^2
See details
Back to top
explanationderivationvariantsrelated formulasrelated definitions
When two vectors are orthogonal, the squared length of their sum decomposes into the sum of squared lengths. This generalizes the elementary plane-geometry theorem to Rn\mathbb{R}^n and to any inner product space. It is the underlying reason orthogonal decompositions are computationally clean: lengths split additively across perpendicular components.
Back to top

Inner Product Axioms

Symmetry:  u,v=v,uLinearity:  cu+dw,v=cu,v+dw,vPositive definite:  v,v>0 for v0\begin{aligned} \text{Symmetry:} \;& \langle\mathbf{u},\mathbf{v}\rangle = \langle\mathbf{v},\mathbf{u}\rangle \\ \text{Linearity:} \;& \langle c\mathbf{u}+d\mathbf{w},\mathbf{v}\rangle = c\langle\mathbf{u},\mathbf{v}\rangle + d\langle\mathbf{w},\mathbf{v}\rangle \\ \text{Positive definite:} \;& \langle\mathbf{v},\mathbf{v}\rangle > 0 \text{ for } \mathbf{v} \neq \mathbf{0} \end{aligned}
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
An inner product on a vector space VV is any function ,:V×VR\langle\cdot,\cdot\rangle: V \times V \to \mathbb{R} satisfying these three axioms. The standard dot product is one example; weighted inner products u,v=uTWv\langle\mathbf{u},\mathbf{v}\rangle = \mathbf{u}^T W \mathbf{v} (with WW symmetric positive definite), L2L^2 function integrals, and the Frobenius matrix inner product are others. Every inner product induces a norm, distance, and notion of orthogonality.
Back to top

Distance Formula (Inner Product)

d(u,v)=uv=i=1n(uivi)2d(\mathbf{u}, \mathbf{v}) = \|\mathbf{u} - \mathbf{v}\| = \sqrt{\sum_{i=1}^{n} (u_i - v_i)^2}
See details
Back to top
explanationvariantsrelated formulasrelated definitions
The Euclidean distance between two vectors is the length of their difference. The function dd satisfies the metric axioms: non-negativity with equality iff u=v\mathbf{u} = \mathbf{v}, symmetry d(u,v)=d(v,u)d(\mathbf{u},\mathbf{v}) = d(\mathbf{v},\mathbf{u}), and the triangle inequality d(u,w)d(u,v)+d(v,w)d(\mathbf{u},\mathbf{w}) \leq d(\mathbf{u},\mathbf{v}) + d(\mathbf{v},\mathbf{w}).
Back to top

Orthogonal Complement

(3 formulas)

Orthogonal Complement Definition

W={vRn:vw=0 for all wW}W^\perp = \{\mathbf{v} \in \mathbb{R}^n : \mathbf{v} \cdot \mathbf{w} = 0 \text{ for all } \mathbf{w} \in W\}
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
The orthogonal complement of a subspace WRnW \subseteq \mathbb{R}^n is the set of all vectors perpendicular to every vector in WW. It is itself a subspace of Rn\mathbb{R}^n. Taking the complement twice returns the original: (W)=W(W^\perp)^\perp = W. The complement structure underlies projection, least squares, and the four fundamental subspaces.
Back to top

Complement Dimension Sum

dim(W)+dim(W)=n\dim(W) + \dim(W^\perp) = n
See details
Back to top
explanationvariantsrelated formulasrelated definitions
For any subspace WW of Rn\mathbb{R}^n, the dimensions of WW and its orthogonal complement add to the ambient dimension. Together they span all of Rn\mathbb{R}^n as a direct sum Rn=WW\mathbb{R}^n = W \oplus W^\perp — every vector decomposes uniquely into a WW-component and a WW^\perp-component.
Back to top

Orthogonal Decomposition (Subspace)

v=v^+z,v^W,zW\mathbf{v} = \hat{\mathbf{v}} + \mathbf{z}, \quad \hat{\mathbf{v}} \in W, \quad \mathbf{z} \in W^\perp
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
Every vector decomposes uniquely into a component in a chosen subspace WW and a component in its orthogonal complement. The WW-component is the orthogonal projection v^=projWv\hat{\mathbf{v}} = \text{proj}_W\mathbf{v}, the closest point in WW to v\mathbf{v}. The residual z\mathbf{z} is perpendicular to all of WW and equals v\mathbf{v} minus its projection.
Back to top

Orthogonal Sets

(4 formulas)

Orthogonal Set Independence

vivj=0    (ij),    vi0    {v1,,vk} linearly independent\mathbf{v}_i \cdot \mathbf{v}_j = 0 \;\; (i \neq j), \;\; \mathbf{v}_i \neq \mathbf{0} \;\Rightarrow\; \{\mathbf{v}_1, \ldots, \mathbf{v}_k\} \text{ linearly independent}
See details
Back to top
explanationderivationvariantsrelated formulasrelated definitions
Any orthogonal set of nonzero vectors is automatically linearly independent — no separate independence check is needed. This is one of the central reasons orthogonality simplifies linear algebra: orthogonal sets come pre-equipped with the independence property that general sets require effort to verify.
Back to top

Orthonormal Set

qiqj=δij={1i=j0ij\mathbf{q}_i \cdot \mathbf{q}_j = \delta_{ij} = \begin{cases} 1 & i = j \\ 0 & i \neq j \end{cases}
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
An orthonormal set is an orthogonal set of unit vectors. The Kronecker delta δij\delta_{ij} packages both conditions: pairwise orthogonality (iji \neq j entries vanish) and unit length (i=ji = j entries equal 11). Any orthogonal set normalizes to orthonormal by dividing each vector by its length.
Back to top

Coordinates via Orthonormal Basis

x=i=1n(xqi)qi,ci=xqi\mathbf{x} = \sum_{i=1}^{n} (\mathbf{x} \cdot \mathbf{q}_i)\,\mathbf{q}_i, \qquad c_i = \mathbf{x} \cdot \mathbf{q}_i
See details
Back to top
explanationderivationvariantsrelated formulasrelated definitions
For an orthonormal basis, the coordinate of x\mathbf{x} along each basis vector is a single dot product. No system to solve, no matrix to invert — each cic_i is computed independently. This is the defining computational advantage of orthonormal bases.
Back to top

Parseval Identity

x2=i=1n(xqi)2=i=1nci2\|\mathbf{x}\|^2 = \sum_{i=1}^{n} (\mathbf{x} \cdot \mathbf{q}_i)^2 = \sum_{i=1}^{n} c_i^2
See details
Back to top
explanationderivationvariantsrelated formulasrelated definitions
In an orthonormal basis, the squared length of a vector equals the sum of squared coordinates. This is the Pythagorean theorem applied to the orthonormal expansion x=ciqi\mathbf{x} = \sum c_i \mathbf{q}_i — orthogonal components contribute additively to squared length, with no cross-terms.
Back to top

Projection

(5 formulas)

Projection onto Subspace

b^=A(ATA)1ATb,P=A(ATA)1AT\hat{\mathbf{b}} = A(A^TA)^{-1}A^T\,\mathbf{b}, \qquad P = A(A^TA)^{-1}A^T
See details
Back to top
explanationconditionsderivationvariantsrelated formulasrelated definitions
When the columns of AA form a basis for a subspace WW, the orthogonal projection of b\mathbf{b} onto WW is computed by the formula above. The matrix P=A(ATA)1ATP = A(A^TA)^{-1}A^T is the projection matrix — it maps any vector to its projection. This is the general formula that works regardless of whether the basis is orthogonal.
Back to top

Projection onto Orthonormal Basis

projWb=i=1k(qib)qi\text{proj}_W\,\mathbf{b} = \sum_{i=1}^{k} (\mathbf{q}_i \cdot \mathbf{b})\,\mathbf{q}_i
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
When the basis {q1,,qk}\{\mathbf{q}_1, \ldots, \mathbf{q}_k\} for WW is orthonormal, projection decomposes into kk independent single-vector projections. Orthogonality eliminates cross-talk: each component is computed by one dot product, with no interference between basis vectors. This is the cleanest projection formula in linear algebra.
Back to top

Orthonormal Columns Projection

P=QQT,QTQ=IkP = QQ^T, \qquad Q^TQ = I_k
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
When the columns of QQ form an orthonormal basis for WW, the projection matrix onto WW collapses to the outer product QQTQQ^T. No inversion is required — ATAA^TA becomes IkI_k and disappears from the general formula. This is the most numerically stable form of projection.
Back to top

Projection Matrix Properties

P2=P,PT=PP^2 = P, \qquad P^T = P
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
An orthogonal projection matrix is idempotent and symmetric. Idempotence (P2=PP^2 = P) reflects that projecting twice gives the same result as projecting once — vectors already in WW are fixed by PP. Symmetry (PT=PP^T = P) is what makes the projection orthogonal rather than oblique: it forces the residual perpendicular to WW, not merely transverse to it. Any matrix satisfying both conditions is an orthogonal projection onto some subspace.
Back to top

Complementary Projection

PW=IPW,PWb+(IPW)b=bP_{W^\perp} = I - P_W, \qquad P_W\mathbf{b} + (I - P_W)\mathbf{b} = \mathbf{b}
See details
Back to top
explanationderivationvariantsrelated formulasrelated definitions
If PP projects onto WW, then IPI - P projects onto WW^\perp. The two projections together decompose every vector into its WW-component and its perpendicular residual. The complementary projection inherits both defining properties: (IP)2=IP(I - P)^2 = I - P and (IP)T=IP(I - P)^T = I - P.
Back to top

Gram-Schmidt

(1 formula)

Gram-Schmidt Process

u1=v1,uj=vji=1j1uivjuiuiui,qj=ujuj\mathbf{u}_1 = \mathbf{v}_1, \qquad \mathbf{u}_j = \mathbf{v}_j - \sum_{i=1}^{j-1} \frac{\mathbf{u}_i \cdot \mathbf{v}_j}{\mathbf{u}_i \cdot \mathbf{u}_i}\,\mathbf{u}_i, \qquad \mathbf{q}_j = \frac{\mathbf{u}_j}{\|\mathbf{u}_j\|}
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
The Gram-Schmidt process converts any linearly independent set {v1,,vk}\{\mathbf{v}_1, \ldots, \mathbf{v}_k\} into an orthogonal set {u1,,uk}\{\mathbf{u}_1, \ldots, \mathbf{u}_k\} spanning the same subspace. At each step jj, vj\mathbf{v}_j has its projections onto all previously computed orthogonal vectors subtracted, leaving only the component perpendicular to Span{u1,,uj1}\text{Span}\{\mathbf{u}_1, \ldots, \mathbf{u}_{j-1}\}. Optionally each uj\mathbf{u}_j is normalized to produce an orthonormal set.
Back to top

Least Squares

(3 formulas)

Normal Equations

ATAx^=ATbA^TA\,\hat{\mathbf{x}} = A^T\mathbf{b}
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
When Ax=bA\mathbf{x} = \mathbf{b} has no exact solution, the least-squares solution x^\hat{\mathbf{x}} — the minimizer of Axb2\|A\mathbf{x} - \mathbf{b}\|^2 — satisfies this square n×nn \times n system regardless of AA's shape. The equations express the orthogonality condition: the residual bAx^\mathbf{b} - A\hat{\mathbf{x}} is perpendicular to every column of AA.
Back to top

Least-Squares Solution

x^=(ATA)1ATb,A+=(ATA)1AT\hat{\mathbf{x}} = (A^TA)^{-1} A^T \mathbf{b}, \qquad A^+ = (A^TA)^{-1}A^T
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
When AA has full column rank, the normal equations have the unique closed-form solution above. The matrix A+=(ATA)1ATA^+ = (A^TA)^{-1}A^T is the left pseudoinverse of AA — it satisfies A+A=InA^+A = I_n. The projection of b\mathbf{b} onto the column space is Ax^=AA+b=PbA\hat{\mathbf{x}} = AA^+\mathbf{b} = P\mathbf{b}.
Back to top

Least-Squares via QR

A=QR    Rx^=QTbA = QR \;\Rightarrow\; R\,\hat{\mathbf{x}} = Q^T\mathbf{b}
See details
Back to top
explanationconditionsderivationrelated formulasrelated definitions
Using the QR factorization A=QRA = QR, the normal equations reduce to an upper triangular system, solved by back substitution. This avoids forming ATAA^TA explicitly, which is numerically critical: the condition number of ATAA^TA is the square of the condition number of AA, so direct normal-equations approaches amplify rounding errors. QR-based least squares is the standard algorithm in numerical software (LAPACK, NumPy, MATLAB).
Back to top

LU

(4 formulas)

LU Decomposition

A=LUA = LU
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
Factors a square matrix AA into a unit lower triangular LL (ones on the diagonal, multipliers below) and an upper triangular UU (the row echelon form). UU records the result of Gaussian elimination; LL records the multipliers used to produce it. The factorization captures the elimination process in reusable form: once computed, any system Ax=bA\mathbf{x} = \mathbf{b} reduces to two triangular solves.
Back to top

PA LU Partial Pivoting

PA=LUPA = LU
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
When zero or near-zero pivots appear during elimination, row swaps are needed. Partial pivoting selects the largest absolute value in the current pivot column as the pivot at each step. The permutation matrix PP records all swaps. This factorization exists for every invertible matrix and is the numerically stable default in software.
Back to top

Determinant via LU

det(A)=(1)si=1nuii\det(A) = (-1)^s \prod_{i=1}^{n} u_{ii}
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
The determinant is a free byproduct of LU factorization: multiply the diagonal entries of UU (the pivots) and account for the sign of the row permutation. Since det(L)=1\det(L) = 1 (unit diagonal) and det(U)=uii\det(U) = \prod u_{ii} (triangular), det(PA)=det(L)det(U)=uii\det(PA) = \det(L)\det(U) = \prod u_{ii}, so det(A)=(1)suii\det(A) = (-1)^s \prod u_{ii} where ss counts row swaps.
Back to top

LU Solve Steps

Ax=b    {Ly=Pb(forward sub)Ux=y(back sub)A\mathbf{x} = \mathbf{b} \;\Longleftrightarrow\; \begin{cases} L\mathbf{y} = P\mathbf{b} & (\text{forward sub}) \\ U\mathbf{x} = \mathbf{y} & (\text{back sub}) \end{cases}
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
Given PA=LUPA = LU, solving Ax=bA\mathbf{x} = \mathbf{b} reduces to two triangular solves: forward substitution down through LL, then back substitution up through UU. Each solve costs O(n2)O(n^2). Factor once at 23n3\frac{2}{3}n^3, then amortize: kk systems with the same coefficient matrix cost 23n3+2kn2\frac{2}{3}n^3 + 2kn^2.
Back to top

Cholesky

(3 formulas)

Cholesky Decomposition

A=LLTA = L L^T
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
Factors a symmetric positive definite matrix AA into a lower triangular LL with strictly positive diagonal entries, times its own transpose. LL is the unique Cholesky factor — the matrix "square root" of AA in the sense A=LLTA = LL^T. Exploits symmetry to halve the cost of LU and requires no pivoting since positive definiteness guarantees positive pivots throughout.
Back to top

Cholesky Diagonal Formula

ljj=ajjk=1j1ljk2l_{jj} = \sqrt{a_{jj} - \sum_{k=1}^{j-1} l_{jk}^2}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The diagonal entries of the Cholesky factor are computed left-to-right. The jj-th diagonal involves subtracting the squared entries already placed in row jj from the diagonal entry of AA, then taking the positive square root. Positive definiteness guarantees the argument under the root is strictly positive at every step.
Back to top

Cholesky Off-Diagonal Formula

lij=1ljj(aijk=1j1likljk),i>jl_{ij} = \frac{1}{l_{jj}}\left( a_{ij} - \sum_{k=1}^{j-1} l_{ik} l_{jk} \right), \qquad i > j
See details
Back to top
explanationconditionsrelated formulasrelated definitions
After computing the diagonal ljjl_{jj}, the entries below it in column jj are computed by subtracting cross-terms involving previously computed factors and dividing by ljjl_{jj}. This fills column jj before moving to column j+1j+1. Together with the diagonal formula, this completes the Cholesky algorithm.
Back to top

QR

(3 formulas)

QR Decomposition

A=QRA = Q R
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
Factors an m×nm \times n matrix AA (with mnm \geq n and linearly independent columns) into QQ with orthonormal columns and RR upper triangular with positive diagonal. The columns of QQ are an orthonormal basis for Col(A)\text{Col}(A); RR records the coefficients expressing each column of AA in that basis. Produced by Gram-Schmidt, Householder reflections, or Givens rotations.
Back to top

QR Gram-Schmidt R Entries

Rij=qiaj    (ij),Rij=0    (i>j),Rjj=ujR_{ij} = \mathbf{q}_i \cdot \mathbf{a}_j \;\; (i \leq j), \qquad R_{ij} = 0 \;\; (i > j), \qquad R_{jj} = \|\mathbf{u}_j\|
See details
Back to top
explanationderivationrelated formulasrelated definitions
When Gram-Schmidt is applied to the columns of AA, the entries of RR are the dot products computed along the way. The upper triangular structure reflects the sequential nature of orthogonalization: aj\mathbf{a}_j's projection onto qi\mathbf{q}_i is zero for i>ji > j because that direction has not yet been introduced. The diagonal entry Rjj=ujR_{jj} = \|\mathbf{u}_j\| is the norm of the unnormalized Gram-Schmidt vector — always positive, making RR unique.
Back to top

QR Algorithm for Eigenvalues

Ak=QkRk,Ak+1=RkQkA_k = Q_k R_k, \qquad A_{k+1} = R_k Q_k
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
The standard algorithm for computing eigenvalues of general matrices. Starting from A0=AA_0 = A, each iteration factors Ak=QkRkA_k = Q_k R_k and forms Ak+1=RkQkA_{k+1} = R_k Q_k. Since Ak+1=QkTAkQkA_{k+1} = Q_k^T A_k Q_k, each step is a similarity transformation preserving eigenvalues. Under mild conditions, AkA_k converges to an upper triangular matrix with eigenvalues on the diagonal — without ever forming the characteristic polynomial.
Back to top

SVD

(10 formulas)

SVD

A=UΣVTA = U \Sigma V^T
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
The most general matrix factorization. Every m×nm \times n matrix — any shape, any rank — factors as the product of an m×mm \times m orthogonal UU, an m×nm \times n diagonal Σ\Sigma with non-negative entries, and an n×nn \times n orthogonal VV (transposed). Geometrically: every linear transformation is a rotation, followed by axis-aligned scaling, followed by another rotation.
Back to top

Singular Values

σi=λi(ATA)=λi(AAT)\sigma_i = \sqrt{\lambda_i(A^TA)} = \sqrt{\lambda_i(AA^T)}
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
The singular values of AA are the square roots of the eigenvalues of ATAA^TA (equivalently AATAA^T). Since ATAA^TA is symmetric positive semi-definite, its eigenvalues are non-negative, making the singular values real and non-negative. They measure the stretching factors of the linear transformation along its principal axes: σ1=maxx=1Ax\sigma_1 = \max_{\|\mathbf{x}\|=1}\|A\mathbf{x}\| is the maximum stretching.
Back to top

SVD Rank

rank(A)=#{i:σi>0}\text{rank}(A) = \#\{i : \sigma_i > 0\}
See details
Back to top
explanationvariantsrelated formulasrelated definitions
The rank of AA equals the number of nonzero singular values. This is the most numerically stable rank determination method: row reduction can miscount rank when small numerical errors push true zeros to small nonzeros, but SVD with a tolerance gives a robust effective rank. Standard practice: count singular values above a tolerance ϵσ1\epsilon \sigma_1.
Back to top

SVD Outer Product Form

A=i=1rσiuiviTA = \sum_{i=1}^{r} \sigma_i \, \mathbf{u}_i \mathbf{v}_i^T
See details
Back to top
explanationderivationvariantsrelated formulasrelated definitions
The SVD expands as a sum of r=rank(A)r = \text{rank}(A) rank-one matrices, each weighted by a singular value. Terms are naturally ordered by importance — the largest σi\sigma_i contributes most. Truncating at kk terms gives the best rank-kk approximation (Eckart-Young). This form underlies image compression, noise reduction, latent semantic analysis, and most matrix approximation methods.
Back to top

Moore-Penrose Pseudoinverse

A+=VΣ+UTA^+ = V \Sigma^+ U^T
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
The Moore-Penrose pseudoinverse generalizes matrix inversion to any matrix, including rectangular and rank-deficient ones. Σ+\Sigma^+ is formed by reciprocating each nonzero singular value and transposing the shape. The pseudoinverse satisfies four defining (Penrose) conditions: AA+A=AAA^+A = A, A+AA+=A+A^+AA^+ = A^+, (AA+)T=AA+(AA^+)^T = AA^+, (A+A)T=A+A(A^+A)^T = A^+A. These uniquely determine A+A^+.
Back to top

Eckart-Young Low-Rank Approximation

Ak=i=1kσiuiviT,AAk2=σk+1,AAkF=i=k+1rσi2A_k = \sum_{i=1}^{k} \sigma_i \mathbf{u}_i \mathbf{v}_i^T, \qquad \|A - A_k\|_2 = \sigma_{k+1}, \quad \|A - A_k\|_F = \sqrt{\sum_{i=k+1}^{r}\sigma_i^2}
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
The Eckart-Young-Mirsky theorem: among all matrices of rank at most kk, the truncated SVD AkA_k is closest to AA in both the operator norm and the Frobenius norm. The approximation error equals the first discarded singular value (operator norm) or the root-sum-of-squares of discarded singular values (Frobenius norm). This is the mathematical foundation of dimensionality reduction.
Back to top

Condition Number

κ(A)=σ1σr\kappa(A) = \frac{\sigma_1}{\sigma_r}
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
The condition number measures how sensitive the linear system Ax=bA\mathbf{x} = \mathbf{b} is to perturbations in b\mathbf{b}. A matrix with κ(A)=10k\kappa(A) = 10^k can lose roughly kk digits of accuracy in floating-point arithmetic. κ=1\kappa = 1 characterizes orthogonal matrices (perfectly conditioned); κ=\kappa = \infty means singular. The ratio of largest to smallest nonzero singular value quantifies the geometric distortion of the transformation.
Back to top

Operator Norm

A2=σ1=maxx=1Ax\|A\|_2 = \sigma_1 = \max_{\|\mathbf{x}\|=1} \|A\mathbf{x}\|
See details
Back to top
explanationconditionsvariantsrelated formulasrelated definitions
The operator (spectral, 2\ell_2) norm of a matrix is its largest singular value. Geometrically, it is the maximum stretching factor: the largest length AA can produce from a unit input. Equivalently, the largest eigenvalue of ATAA^TA in absolute value.
Back to top

Frobenius Norm via Singular Values

AF=i=1rσi2\|A\|_F = \sqrt{\sum_{i=1}^{r} \sigma_i^2}
See details
Back to top
explanationderivationvariantsrelated formulasrelated definitions
The Frobenius norm — equal to i,jaij2=tr(ATA)\sqrt{\sum_{i,j}|a_{ij}|^2} = \sqrt{\text{tr}(A^TA)} — has a clean SVD characterization as the root-sum-of-squares of singular values. This connects the entrywise "total energy" of a matrix to its spectral content.
Back to top

SVD Four Fundamental Subspaces

Col(A)=Span{u1,,ur}Null(AT)=Span{ur+1,,um}Row(A)=Span{v1,,vr}Null(A)=Span{vr+1,,vn}\begin{aligned} \text{Col}(A) &= \text{Span}\{\mathbf{u}_1, \ldots, \mathbf{u}_r\} \\ \text{Null}(A^T) &= \text{Span}\{\mathbf{u}_{r+1}, \ldots, \mathbf{u}_m\} \\ \text{Row}(A) &= \text{Span}\{\mathbf{v}_1, \ldots, \mathbf{v}_r\} \\ \text{Null}(A) &= \text{Span}\{\mathbf{v}_{r+1}, \ldots, \mathbf{v}_n\} \end{aligned}
See details
Back to top
explanationconditionsrelated formulasrelated definitions
The SVD simultaneously delivers orthonormal bases for all four fundamental subspaces of any matrix. The first rr left singular vectors (columns of UU) span the column space; the remaining mrm-r span its orthogonal complement (left null space). The first rr right singular vectors (columns of VV) span the row space; the remaining nrn-r span the null space. No other factorization provides all four bases at once, and all four are guaranteed orthonormal.
Back to top

Cross-Decomposition

(1 formula)

Quadratic Form Diagonalization

xTAx=yTDy=i=1nλiyi2,x=Qy\mathbf{x}^T A \mathbf{x} = \mathbf{y}^T D \mathbf{y} = \sum_{i=1}^{n} \lambda_i y_i^2, \qquad \mathbf{x} = Q\mathbf{y}
See details
Back to top
explanationderivationvariantsrelated formulasrelated definitions
For symmetric AA with spectral decomposition A=QDQTA = QDQ^T, the change of variables x=Qy\mathbf{x} = Q\mathbf{y} diagonalizes the quadratic form xTAx\mathbf{x}^TA\mathbf{x} into a sum of independent squared terms weighted by eigenvalues. The eigenvectors define the principal axes of the quadratic surface; eigenvalue signs classify the form: all λi>0\lambda_i > 0 (positive definite, ellipsoid), all 0\geq 0 (positive semi-definite), mixed signs (indefinite, hyperboloid).
Back to top
Representation
Vector Component FormStandard Basis DecompositionDirection CosinesDirection Cosines Identity
Operations
Vector AdditionVector SubtractionScalar Multiplication of VectorsLinear CombinationSpanMatrix EqualityMatrix AdditionMatrix SubtractionScalar Multiplication of MatricesMatrix MultiplicationMatrix-Vector Product (Column Form)Matrix Multiplication AssociativityMatrix Multiplication DistributivityMatrix Multiplication Non-CommutativityMatrix Power
Norm & Distance
Euclidean NormDistance FormulaVector NormalizationNorm Scaling PropertyTriangle InequalityCauchy-Schwarz Inequality
Dot Product
Dot Product (Algebraic)Dot Product (Geometric)Angle Between VectorsSelf Dot ProductOrthogonality ConditionScalar ProjectionVector ProjectionOrthogonal Decomposition
Cross Product
Cross Product (Component Form)Cross Product (Determinant Form)Cross Product MagnitudeStandard Basis Cross ProductsCross Product Anti-CommutativityParallelism Test (Cross Product)Vector Triple ProductLagrange Identity
Triple Products
Scalar Triple ProductParallelogram AreaParallelepiped VolumePyramid Volume
Vector Space
Vector Space AxiomsScalar-Zero PropertyNegative One Scalar
Subspaces
Subspace TestSubspace Test Combined
Span
Span (Set Definition)Span Membership CriterionSpan Is Smallest Subspace
Linear Independence
Linear Independence EquationLinear Independence Matrix TestLinear Independence Determinant TestMax Independent Set SizeWronskian Test
Basis & Coordinates
Basis DefinitionUnique Basis RepresentationCoordinate VectorStandard Basis (Rn)Change of Basis FormulaChange of Basis InverseCoordinate Map Linearity
Dimension & Rank
Dimension DefinitionSubspace Dimension InequalityDimension Sum FormulaDirect Sum CriterionDirect Sum DimensionRank-Nullity Theorem (Matrix Form)Four Fundamental Subspaces DimensionsRow Rank Equals Column Rank
Transpose & Symmetry
Transpose DefinitionTranspose InvolutionTranspose of SumTranspose of Scalar MultipleTranspose of ProductSymmetric Matrix DefinitionSkew-Symmetric Matrix DefinitionSymmetric Skew DecompositionGram Matrix Symmetry
Special Matrix Types
Identity Matrix DefinitionIdentity Matrix PropertyDiagonal Matrix DefinitionDiagonal Matrix PowerDiagonal Matrix DeterminantTriangular Matrix DeterminantOrthogonal Matrix DefinitionOrthogonal Matrix DeterminantIdempotent Matrix DefinitionIdempotent Rank TraceNilpotent Matrix DefinitionNeumann Series NilpotentInvolutory Matrix DefinitionCross Product Skew Matrix
Inverse
Inverse DefinitionInverse 2x2 FormulaInverse via AdjugateInverse via Row ReductionInverse InvolutionInverse of ProductInverse of TransposeInverse of Scalar MultipleInverse of PowerDeterminant of InverseDiagonal Matrix InverseOrthogonal Matrix InverseSolve System via InverseInvertible Matrix TheoremSingular Matrix Definition
Rank
Rank BoundsRank of TransposeRank Product InequalitySylvester Rank InequalityRank Sum InequalityRank Invariance InvertibleGram Rank IdentityRank-One Outer Product
Trace
Trace DefinitionTrace LinearityTrace of TransposeTrace Cyclic PropertyTrace Sum of EigenvaluesTrace Similarity InvarianceTrace of CommutatorTrace Symmetric Skew ProductTrace Orthonormal BasisFrobenius Inner ProductFrobenius Norm
Definitions
Determinant 2x2Determinant 3x3Determinant Recursive DefinitionDeterminant Permutation Formula
Cofactor Structure
Minor DefinitionCofactor DefinitionCofactor Matrix DefinitionAdjugate Definition
Cofactor Expansion
Laplace Row ExpansionLaplace Column Expansion
Row Operation Effects
Determinant Row SwapDeterminant Row ScalingDeterminant Row Addition
Algebraic Properties
Determinant of TransposeDeterminant of ProductDeterminant of Scalar MultipleDeterminant of PowerDeterminant of Identity
Special Determinants
Block Triangular DeterminantVandermonde Determinant
Adjugate & Inverse
Adjugate Identity
Linear Systems
Cramers Rule
Geometric Interpretation
Determinant Signed Area 2DDeterminant Signed Volume 3DDeterminant Volume Scaling FactorTriangle Area via DeterminantTetrahedron Volume via Determinant
Eigenvalue Connection
Determinant Product of Eigenvalues
Standard Forms
Linear Equation Standard FormLinear System Matrix FormVector Equation FormAugmented Matrix Construction
Echelon Forms
Row Echelon Form DefinitionReduced Row Echelon Form DefinitionRREF UniquenessPivot Definition
Elementary Row Operations
Elementary Row OperationsRow Equivalence Preserves Solutions
Solvability
Free Variables CountSolvability Rank CriterionSolution Structure Decomposition
Homogeneous Systems
Homogeneous Solution Space DimensionUnderdetermined Homogeneous Has Nontrivial
Definition & Properties
Linear Transformation DefinitionZero Vector PreservationComposition Is Linear
Matrix Representation
Standard MatrixLinear Map as Matrix MultiplicationMatrix Representation Abstract Bases
Image & Kernel
Image DefinitionKernel DefinitionInjectivity Kernel CriterionRank-Nullity for MapsBijectivity Equal Dim Case
Similarity & Basis Change
Similarity RelationSimilarity InvariantsDiagonalization Formula
Geometric Transformations
Rotation Matrix 2DRotation Matrices 3DReflection Across Line 2DHouseholder ReflectionProjection onto LineProjection onto PlaneShear Matrix
Foundation
Eigenvalue DefinitionCharacteristic EquationEigenspace
Characteristic Polynomial
Characteristic PolynomialCharacteristic Polynomial 2x2Cayley-Hamilton
Multiplicities
Multiplicity InequalityIndependence of Distinct Eigenvectors
Eigenvalue Algebra
Eigenvalue of PowerEigenvalue of InverseEigenvalue of PolynomialEigenvalue Shift
Diagonalizability
Diagonalizability ConditionDistinct Eigenvalues Imply Diagonalizable
Special Spectra
Special Matrix Eigenvalue Restrictions
Spectral
Spectral TheoremSpectral Decomposition
Applications
Matrix Exponential
Complex
Complex Conjugate PairsDiscriminant Classification 2x2Real Canonical Form 2x2
Inner Product
Cauchy-Schwarz Inequality (General)Triangle Inequality (Inner Product)Pythagorean TheoremInner Product AxiomsDistance Formula (Inner Product)
Orthogonal Complement
Orthogonal Complement DefinitionComplement Dimension SumOrthogonal Decomposition (Subspace)
Orthogonal Sets
Orthogonal Set IndependenceOrthonormal SetCoordinates via Orthonormal BasisParseval Identity
Projection
Projection onto SubspaceProjection onto Orthonormal BasisOrthonormal Columns ProjectionProjection Matrix PropertiesComplementary Projection
Gram-Schmidt
Gram-Schmidt Process
Least Squares
Normal EquationsLeast-Squares SolutionLeast-Squares via QR
LU
LU DecompositionPA LU Partial PivotingDeterminant via LULU Solve Steps
Cholesky
Cholesky DecompositionCholesky Diagonal FormulaCholesky Off-Diagonal Formula
QR
QR DecompositionQR Gram-Schmidt R EntriesQR Algorithm for Eigenvalues
SVD
SVDSingular ValuesSVD RankSVD Outer Product FormMoore-Penrose PseudoinverseEckart-Young Low-Rank ApproximationCondition NumberOperator NormFrobenius Norm via Singular ValuesSVD Four Fundamental Subspaces
Cross-Decomposition
Quadratic Form Diagonalization