Invertible Linear Transformations

Overview

Most of the theory of linear transformations concerns maps that run one way: inputs produce outputs and the traffic is one-directional. A transformation is invertible when that asymmetry disappears — when there is a companion transformation that returns every output to the input that produced it, and returns every input to itself after a round trip. Formally, $T \colon U \to V$ is invertible if some function $S \colon V \to U$ satisfies both $S \circ T = I_U$ and $T \circ S = I_V$ , where $I_U$ and $I_V$ are the identity transformations on the two spaces.

The requirement that both compositions give the identity is essential and is not redundant. One-sided conditions are strictly weaker in general: a transformation can be undone from the left without being undone from the right, or the reverse. Each one-sided condition corresponds to one of the two structural properties — a left inverse to injectivity, a right inverse to surjectivity — and requiring both is exactly what forces the transformation to be simultaneously one-to-one and onto.

Two facts make the definition workable. The inverse is unique: there is never a choice of which transformation undoes $T$ , so the notation $T^{-1}$ is unambiguous. And the inverse is automatically a linear transformation, even though the definition only demanded a function. Nothing extra needs to be imposed; linearity of $T$ propagates to its inverse through the composition identities. Consequently $T^{-1}$ is itself invertible, with inverse $T$ , and the relation between the two spaces is entirely symmetric.

In engineering terms, invertibility is the condition for exact reconstruction. If $T$ models an encoding, a sampling operation, a coordinate change or a measurement chain, then invertibility says the original object can be recovered from the recorded one with no residual and no ambiguity. Where the transformation is a matrix product, invertibility of the transformation is nonsingularity of the matrix, and $T^{-1}$ is multiplication by $A^{-1}$ — but the abstract formulation covers coordinate changes between polynomial, matrix and function spaces where no matrix is given in advance.

Definition

Identity Linear Transformation

IDLT

For a vector space $W$ , the identity transformation $I_W \colon W \to W$ is defined by $I_W(\vec{w}) = \vec{w}$ for every $\vec{w} \in W$ . It is a linear transformation, its kernel is trivial and its range is all of $W$ , so it is both injective and surjective.

The identity is the reference point against which invertibility is measured: a transformation is invertible when a round trip reproduces it exactly.

Invertible Linear Transformation

IVLT

Let $T \colon U \to V$ be a linear transformation. Then $T$ is invertible if there exists a function $S \colon V \to U$ satisfying both composition identities:

$S \circ T = I_U$ — applying $T$ then $S$ returns every input unchanged;
$T \circ S = I_V$ — applying $S$ then $T$ returns every output unchanged.

The function

S

is called the inverse of

T

and is written

T^{-1}

The definition asks only for a function. That the inverse turns out to be a linear transformation is a theorem, not part of the hypothesis.

Composition of Linear Transformations

LTC

For $T \colon U \to V$ and $S \colon V \to W$ , the composition $(S \circ T) \colon U \to W$ is defined by $(S \circ T)(\vec{u}) = S(T(\vec{u}))$ . The composition of two linear transformations is again a linear transformation.

Composition is associative but not commutative, and the order matters: $S \circ T$ means $T$ first. Both orders appear in the definition of an inverse, applied to different spaces.

Concepts

Why both composition identities are required

Consider $T \colon \mathbb{C}^2 \to \mathbb{C}^3$ with $T(x_1, x_2) = (x_1, x_2, x_1 + x_2)$ . The function $S(v_1, v_2, v_3) = (v_1, v_2)$ satisfies $S \circ T = I_{\mathbb{C}^2}$ , so $T$ has a left inverse; but $T \circ S$ is not the identity on $\mathbb{C}^3$ , since it maps $(0,0,1)$ to $(0,0,0)$ . Now consider $Q \colon \mathbb{C}^3 \to \mathbb{C}^2$ with $Q(x_1, x_2, x_3) = (x_1, x_2)$ . It has a right inverse but no left inverse. The first transformation is injective and not surjective; the second is surjective and not injective. Neither is invertible, and the two-sided requirement is precisely what excludes both cases.

The two ways an inverse fails to exist

Suppose an inverse $S$ is sought for a transformation that is not surjective. Choose $\vec{v}$ outside the range. The identity $T(S(\vec{v})) = \vec{v}$ would place $\vec{v}$ in the range, a contradiction, so no value can be assigned to $S(\vec{v})$ at all: the putative inverse is undefined on part of its domain. Now suppose the transformation is not injective, with $T(\vec{x}) = T(\vec{y}) = \vec{v}$ for distinct $\vec{x}, \vec{y}$ . The identity $S(T(\vec{u})) = \vec{u}$ demands both $S(\vec{v}) = \vec{x}$ and $S(\vec{v}) = \vec{y}$ : the putative inverse is multiply defined and is not a function. Since a non-trivial kernel is an entire subspace, in that case $S(\vec{v})$ would have to take a whole subspace's worth of values simultaneously.

The inverse is unique

Suppose $S_1$ and $S_2$ both satisfy the two composition identities for $T$ . Then, using associativity of composition, $S_1 = S_1 \circ I_V = S_1 \circ (T \circ S_2) = (S_1 \circ T) \circ S_2 = I_U \circ S_2 = S_2$ . Uniqueness is what licenses the notation $T^{-1}$ and the phrase ‘the inverse’. Note that the argument uses one identity from each side, which is another indication that the two-sided requirement is doing genuine work.

The inverse is automatically linear

The definition demands only a function, yet linearity comes free. For additivity, take $\vec{x}, \vec{y} \in V$ and write each as $T$ of its inverse image. Then $T^{-1}(\vec{x} + \vec{y}) = T^{-1}\left(T(T^{-1}(\vec{x})) + T(T^{-1}(\vec{y}))\right) = T^{-1}\left(T\left(T^{-1}(\vec{x}) + T^{-1}(\vec{y})\right)\right) = T^{-1}(\vec{x}) + T^{-1}(\vec{y})$ , where the middle step uses linearity of $T$ and the outer steps use the composition identities. The scalar case is identical with $\alpha T^{-1}(\vec{x})$ in place of the sum. So $T^{-1}$ satisfies both defining properties of a linear transformation.

Invertibility is symmetric

Because $T^{-1}$ is a linear transformation satisfying $T \circ T^{-1} = I_V$ and $T^{-1} \circ T = I_U$ , it meets the definition of invertibility with $T$ playing the role of its inverse. Hence $T^{-1}$ is invertible and $\left(T^{-1}\right)^{-1} = T$ . The relationship between $U$ and $V$ is therefore completely symmetric: neither space is privileged, and the pair $(T, T^{-1})$ identifies the two spaces with one another. That identification is what the term isomorphism names.

The criterion, and the matrix case

The definition is a statement about the existence of a companion function, which is awkward to check directly. The decisive result is that $T$ is invertible if and only if it is both injective and surjective, and that criterion converts the question into two computations already available. For $T(\vec{x}) = A\vec{x}$ on $\mathbb{C}^n$ , injectivity is full column rank and surjectivity is full row rank; for a square $A$ both reduce to nonsingularity, and then $T^{-1}(\vec{x}) = A^{-1}\vec{x}$ . Composition corresponds to matrix multiplication, and the reversal rule for inverses of compositions, $(S \circ T)^{-1} = T^{-1} \circ S^{-1}$ , is the transformation-level form of $(BA)^{-1} = A^{-1}B^{-1}$ .

Establishing that a transformation is invertible

Check the dimensionsInvertibility forces

\dim(U) = \dim(V)

, because injectivity requires one inequality and surjectivity the other. Unequal dimensions end the question.

Test injectivityShow the kernel is trivial, or equivalently that the coefficient matrix has full column rank.

Test surjectivityShow the range fills the codomain, or equivalently that the coefficient matrix has full row rank.

Construct the candidate inverseSolve

T(\vec{u}) = \vec{v}

for

\vec{u}

in terms of a symbolic

\vec{v}

; the resulting formula is

T^{-1}

Verify both compositionsConfirm

S \circ T = I_U

and

T \circ S = I_V

symbolically. One direction alone is not sufficient evidence.

Record the consequencesThe inverse is unique, is linear, and satisfies

\left(T^{-1}\right)^{-1} = T

Equations

Identity linear transformation

EQ-IVLT-01

I_{W} : W \to W, I_{W} (\vec{w}) = \vec{w} \forall \vec{w} \in W

The do-nothing transformation on a vector space, and the reference against which a round trip is compared.

Defining condition for invertibility

EQ-IVLT-02

S \circ T = I_{U} and T \circ S = I_{V}

Both identities are required, and they act on different spaces. Satisfying only one gives a one-sided inverse, which is strictly weaker.

Pointwise form of the two conditions

EQ-IVLT-03

S (T (\vec{u})) = \vec{u} \forall \vec{u} \in U, T (S (\vec{v})) = \vec{v} \forall \vec{v} \in V

The version used in practice: substitute the candidate formula and simplify until each side reduces to its own argument.

Uniqueness of the inverse

EQ-IVLT-04

S_{1} = S_{1} \circ (T \circ S_{2}) = (S_{1} \circ T) \circ S_{2} = S_{2}

Any two inverses coincide, which is what makes the notation $T^{-1}$ well defined. The argument uses one composition identity from each side.

Linearity of the inverse

EQ-IVLT-05

T^{- 1} (\vec{x} + \vec{y}) = T^{- 1} (\vec{x}) + T^{- 1} (\vec{y}), T^{- 1} (α \vec{x}) = α T^{- 1} (\vec{x})

The inverse inherits both defining properties of a linear transformation from $T$ , without any additional hypothesis.

Involution property

EQ-IVLT-06

(T^{- 1})^{- 1} = T

Inverting twice returns the original transformation, so the roles of the two spaces are interchangeable.

Invertibility criterion

EQ-IVLT-07

T invertible \Leftrightarrow T injective and T surjective

The workable test: injectivity makes the candidate inverse single-valued, surjectivity makes it defined everywhere on $V$ .

Variable Definitions

Symbols used on this page
Symbol	Name	Meaning	Domain / type
$T$	Linear transformation	The map $T \colon U \to V$ being inverted	function from U to V
$T^{- 1}$	Inverse transformation	The unique map $V \to U$ satisfying both composition identities	function from V to U
$S$	Candidate inverse	A proposed function $V \to U$ , before the two identities have been verified	function from V to U
$I_{U}$	Identity on the domain	The transformation reproduced by applying $T$ then $T^{-1}$	function from U to U
$I_{V}$	Identity on the codomain	The transformation reproduced by applying $T^{-1}$ then $T$	function from V to V
$U$	Domain	The space of original objects, recovered exactly when $T$ is invertible	vector space over C
$V$	Codomain	The space of encoded or measured objects	vector space over C
$A$	Matrix of the transformation	Coefficient matrix when $T(\vec{x}) = A\vec{x}$ ; invertibility of $T$ is nonsingularity of $A$	n x n matrix
$α$	Scalar	An arbitrary scalar used in the linearity checks	complex scalar

Worked Numerical Example

Problem statement

An acquisition channel is known to carry a quadratic signal $p(x) = a + bx + cx^{2}$ over a normalised window, and it is sampled at the three instants $x = 0, 1, 2$ . The sampling operation is the linear transformation $T \colon P_2 \to \mathbb{C}^{3}$ with $T(p) = (p(0), p(1), p(2))$ . Establish that the acquisition is invertible, construct the reconstruction formula, and verify both composition identities.

Write the sampling map explicitly
Evaluate the quadratic at the three sample instants and collect the results:
$T (a + b x + c x^{2}) = [\begin{matrix} a \\ a + b + c \\ a + 2 b + 4 c \end{matrix}]$
Confirm the dimensions match
The domain $P_2$ has dimension $3$ and the codomain $\mathbb{C}^3$ has dimension $3$ . Invertibility requires equality, so this is a necessary precondition and it is satisfied. Three samples for three unknown coefficients is exactly the balance point — one sample fewer would lose information, one more would be redundant.
Solve for the coefficients with a symbolic target
Let $\vec{v} = (v_1, v_2, v_3)$ be an arbitrary sample vector and solve $T(p) = \vec{v}$ . From the first equation $a = v_1$ . Subtracting gives $b + c = v_2 - v_1$ and $2b + 4c = v_3 - v_1$ , hence $b + 2c = \tfrac{1}{2}(v_3 - v_1)$ . Subtracting the two yields $c$ , and back-substitution yields $b$ :
$a = v_{1}, b = \frac{- 3 v_{1} + 4 v_{2} - v_{3}}{2}, c = \frac{v_{1} - 2 v_{2} + v_{3}}{2}$
Define the candidate inverse
The solution exists and is unique for every $\vec{v}$ , which already signals that $T$ is both surjective and injective. Package it as a function $S \colon \mathbb{C}^{3} \to P_2$ :
$S (\vec{v}) = v_{1} + \frac{- 3 v_{1} + 4 v_{2} - v_{3}}{2} x + \frac{v_{1} - 2 v_{2} + v_{3}}{2} x^{2}$
Verify the first composition identity
Apply $S$ to $T(a + bx + cx^{2}) = (a,\ a+b+c,\ a+2b+4c)$ . The constant term is $a$ directly. The linear coefficient is $\tfrac{1}{2}\left(-3a + 4(a+b+c) - (a+2b+4c)\right) = \tfrac{1}{2}(2b) = b$ . The quadratic coefficient is $\tfrac{1}{2}\left(a - 2(a+b+c) + (a+2b+4c)\right) = \tfrac{1}{2}(2c) = c$ . Every coefficient is returned unchanged.
$(S \circ T) (a + b x + c x^{2}) = a + b x + c x^{2} = I_{P_{2}} (a + b x + c x^{2})$
Verify the second composition identity
Now sample the reconstructed polynomial $q = S(\vec{v})$ at the three instants. At $x = 0$ , $q(0) = v_1$ . At $x = 1$ , $q(1) = v_1 + \tfrac{1}{2}(-3v_1 + 4v_2 - v_3) + \tfrac{1}{2}(v_1 - 2v_2 + v_3) = v_1 + \tfrac{1}{2}(-2v_1 + 2v_2) = v_2$ . At $x = 2$ , $q(2) = v_1 + (-3v_1 + 4v_2 - v_3) + 2(v_1 - 2v_2 + v_3) = v_3$ .
$(T \circ S) (\vec{v}) = \vec{v} = I_{ℂ^{3}} (\vec{v})$
Conclude and record the consequences
Both identities hold for arbitrary arguments, so $T$ is invertible and $T^{-1} = S$ . By uniqueness this is the only reconstruction rule that works, so no alternative interpolation formula on these three instants can differ from it. By the linearity theorem $T^{-1}$ is itself a linear transformation, and by the involution property its inverse is $T$ .
Spot-check with a numeric instance
Reconstruct from the sample vector $\vec{v} = (1, 0, 0)$ : the formula gives $q = 1 - \tfrac{3}{2}x + \tfrac{1}{2}x^{2}$ . Sampling back gives $q(0) = 1$ , $q(1) = 1 - \tfrac{3}{2} + \tfrac{1}{2} = 0$ and $q(2) = 1 - 3 + 2 = 0$ , recovering $\vec{v}$ exactly.
$Spot-check with a numeric instance$

Result

Three samples of a quadratic taken at distinct instants determine the signal completely, and the reconstruction is a linear transformation given in closed form. Because the inverse is unique, the reconstruction filter is not a design choice but a consequence of the sampling instants, and the three basis polynomials appearing as the columns of the reconstruction are the Lagrange interpolants for those instants. Adding a fourth sample instant would leave the map injective but destroy surjectivity onto the enlarged sample space; removing one would leave a one-parameter family of quadratics consistent with the data.

Applications & Industry Use

Signal processing

Exact reconstruction from samples

Sampling a band-limited or polynomial signal at enough distinct instants is an invertible linear transformation, and its inverse is the interpolation rule. Invertibility is what licenses the claim that the samples are a complete representation, with no information discarded between acquisition and reconstruction.

Computer graphics

Reversible coordinate frames

Model, view and projection transforms must be invertible for picking, ray casting and shadow mapping, all of which require mapping screen-space data back into world space. A degenerate transform breaks the round trip and is detected as failure of the two-sided composition test.

Cryptography

Invertible diffusion layers

The linear layer of a block cipher must be invertible over its finite field so decryption exists, and the inverse layer is derived once at design time. The reversal rule for compositions is what determines the order of the inverse round operations.

Control systems

State transformations and similarity

Changing state coordinates is an invertible linear transformation, and the inverse recovers the physical states from the transformed ones. Controller design carried out in a convenient coordinate frame is only meaningful because the change of basis can be undone exactly.

Data compression

Lossless linear transforms

Reversible transforms used in lossless coding must be invertible as linear maps, so that the decoder reproduces the original block bit for bit. Invertibility is verified by the two-sided composition identity rather than by round-trip testing on sample data.

Metrology

Calibration and de-calibration

A calibration matrix maps raw instrument readings to physical quantities. Invertibility guarantees that a physical target can be converted back into the raw reading that would produce it, which is required for closed-loop instrument verification and for generating synthetic test data.

Design Considerations

Verify both directions, not one

A candidate inverse that satisfies $S \circ T = I_U$ may fail $T \circ S = I_V$ , and vice versa. One-sided verification is the most common gap in a written argument. Check both identities symbolically on arbitrary arguments; a numeric round trip on sample data is a useful supplement but not a proof.

Equal dimensions are necessary, not sufficient

Invertibility forces $\dim(U) = \dim(V)$ , so unequal dimensions settle the question negatively at no cost. Equal dimensions settle nothing: the zero transformation between two spaces of the same dimension is not invertible. Follow the dimension check with injectivity or surjectivity, either of which suffices when the dimensions agree.

Prefer the criterion to the definition

Establishing invertibility directly requires producing a companion function and verifying two identities. Establishing injectivity and surjectivity requires two rank computations that are mechanical and mechanisable. Use the criterion to decide the question, and construct the inverse only when the formula itself is needed.

Exact invertibility is not numerical invertibility

A transformation can be invertible in exact arithmetic while its inverse amplifies noise catastrophically. The relevant engineering quantity is the condition number, not the yes-or-no verdict. Report both, and for reconstruction problems assess whether the required precision survives the amplification implied by the condition number.

Do not form the inverse when a solve will do

Constructing $T^{-1}$ explicitly and then applying it is usually more expensive and less accurate than solving $T(\vec{u}) = \vec{v}$ directly by factorisation. Build the inverse only when it is needed many times with the same transformation, or when the closed-form expression is itself the deliverable.

Respect the order reversal under composition

Inverting a chain of transformations reverses the order: $(S \circ T)^{-1} = T^{-1} \circ S^{-1}$ . Pipelines that undo their stages in the forward order will produce silently wrong results whenever the stages do not commute, which is the normal case.

Standards & Codes

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

Applicable standards, conventions and reference implementations
Reference	Title	Relevance to this topic
`ISO 80000-2`	Quantities and units — Part 2: Mathematics	Prescribes the notation for mappings, composition and inverse functions, including the superscript form $T^{-1}$ and the identity map, used throughout this page.
`IEEE 754-2019`	IEEE Standard for Floating-Point Arithmetic	Governs the arithmetic in which a numerical inverse is computed and applied, and explains why an exactly invertible transformation can still fail to reconstruct usefully when the condition number is large.
`LAPACK / BLAS reference`	Linear Algebra PACKage reference implementation	Provides `xGETRF` and `xGETRI` for explicit inversion, and `xGETRS` for solving without forming the inverse, together with condition estimation through `xGECON`.
`ISO/IEC 18033-3`	Information technology — Security techniques — Encryption algorithms — Block ciphers	Requires the linear diffusion layers of a cipher to be invertible over their field; the inverse layer used in decryption is exactly the $T^{-1}$ of this page.
`ISO/IEC 40314`	Mathematical Markup Language (MathML) Version 3.0	Encodes the composition identities and reconstruction formulae on this page as semantic markup rather than images, keeping them searchable and available to assistive technology.

Material Selection

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

Representation and precision selection
Representation	Select when	Trade-off
Exact rational arithmetic	Symbolic reconstruction formulae, interpolation rules and coordinate changes that must hold identically.	Produces an exact closed-form inverse suitable for publication, but the fractions grow and the approach does not scale to large systems.
IEEE 754 binary64 with LU factorisation and back-substitution	Numerical inversion of a moderate dense transformation, the default engineering choice.	Fast and accurate for well-conditioned problems; the factorisation is reusable across right-hand sides and avoids forming the inverse explicitly.
IEEE 754 binary64 with QR or SVD	Transformations that are invertible but ill conditioned, or where the condition number must be reported.	More stable and gives the singular values directly, at several times the cost of an LU factorisation.
Explicit inverse matrix	The same transformation is inverted very many times with different data, or the inverse itself is the deliverable.	Amortises well over many applications, but is less accurate than repeated solves and roughly triples the factorisation cost.
Fixed-point arithmetic	Embedded reconstruction with a fixed, known transformation and hard timing constraints.	Deterministic and cheap once the inverse coefficients are precomputed, but scaling must be designed so intermediate products cannot overflow.
Finite field arithmetic	Cipher diffusion layers and algebraic coding, where the inverse must be exact over the field in use.	Exact with bounded operand size and no conditioning concerns; invertibility over the field is a different question from invertibility over the rationals.

Manufacturing Notes

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

Constructing the inverse by hand

Solve $T(\vec{u}) = \vec{v}$ with a fully symbolic $\vec{v}$ and read the answer as a formula. For a matrix transformation this is the augmented reduction $[\, A \mid I_n \,] \to [\, I_n \mid A^{-1} \,]$ , which produces the inverse and simultaneously certifies nonsingularity: if the left block fails to reduce to the identity, the transformation is not invertible and the computation has already proved it.

Cost of inversion

An LU factorisation of an $n \times n$ matrix costs about $\tfrac{2}{3}n^3$ operations and each subsequent solve about $2n^2$ . Forming the explicit inverse costs roughly $2n^3$ , so inverting pays for itself only after order $n$ applications — and even then the repeated solve is usually the more accurate route.

Verification discipline

Verify both compositions symbolically on arbitrary arguments. A numeric check on random data is a valuable secondary test and catches transcription errors, but it can pass by coincidence on a low-dimensional sample and it never covers the whole space. Where a matrix is involved, checking $A A^{-1} = I$ and $A^{-1} A = I$ to within a tolerance is the numeric analogue.

Library routes

numpy.linalg.solve and scipy.linalg.lu_solve apply an inverse without forming it; numpy.linalg.inv forms it explicitly and is best avoided inside loops. numpy.linalg.cond reports the amplification factor. Symbolic systems expose Matrix.inv() over the rationals, which is the right tool when the closed-form reconstruction rule is what is wanted.

Composition and the reversal rule

When invertible transformations are chained, the composition is invertible and its inverse is the composition of the individual inverses in reverse order, $(S \circ T)^{-1} = T^{-1} \circ S^{-1}$ . Implementations that store a pipeline as an ordered list must therefore traverse it backwards when inverting, applying each stage's inverse in turn.

Failure Modes & Common Mistakes

Failure modes, root causes and prevention
Failure mode / mistake	Impact	Root cause	Prevention & detection
Verifying only one composition	high	Checking $S \circ T = I_U$ and assuming the other identity follows.	Verify both. One-sided inverses exist for merely injective or merely surjective transformations and do not establish invertibility.
Assuming equal dimensions imply invertibility	high	Treating $\dim(U) = \dim(V)$ as sufficient rather than necessary.	Equal dimensions permit invertibility. Confirm injectivity or surjectivity as well; when the dimensions agree, either one implies the other.
Writing $T^{-1}$ for a non-invertible transformation	high	Confusing the preimage notation $T^{-1}(\vec{v})$ , which always makes sense as a set, with an inverse function.	Reserve $T^{-1}$ as a function only after invertibility is established; use explicit preimage language otherwise.
Reversing a pipeline in the wrong order	medium	Applying stage inverses in forward order, ignoring the reversal rule for compositions.	Invert a chain back to front. Composition does not commute, so forward-order undoing is wrong except in special cases.
Forming an explicit inverse inside a loop	medium	Calling a matrix inversion routine each time a solve is required.	Factorise once and reuse the factors, or precompute the inverse outside the loop when it is genuinely reused.
Ignoring conditioning	high	Treating a non-zero determinant or a passing rank test as evidence that reconstruction is reliable.	Report the condition number. An invertible but ill-conditioned transformation amplifies input noise by that factor during reconstruction.
Testing invertibility by round-tripping sample data only	medium	Concluding invertibility from successful reconstruction of a few test vectors.	Sample data can round-trip through a non-invertible map if it happens to lie in the range. Verify the identities symbolically or by a rank argument.
Assuming a left and a right inverse must be equal without both identities	low	Applying the uniqueness argument when only one-sided conditions hold.	The uniqueness proof consumes one identity from each side. With only one-sided conditions, distinct one-sided inverses can coexist.

FAQs

Why does the definition require both compositions to be the identity?

Because each identity captures a different structural property. A left inverse exists exactly when the transformation is injective, and a right inverse exactly when it is surjective. Requiring both simultaneously is what forces the transformation to be one-to-one and onto, and only then is a genuine two-sided inverse available.

Is the inverse of a linear transformation automatically linear?

Yes, and this is a theorem rather than an assumption. The definition asks only for a function, but the composition identities combined with linearity of $T$ force $T^{-1}$ to preserve addition and scalar multiplication. No additional hypothesis is needed.

Can a transformation have more than one inverse?

No. If two functions both satisfy the two-sided condition, composing them through the identities shows they are equal. That uniqueness is what justifies the definite article in ‘the inverse’ and the notation $T^{-1}$ . One-sided inverses, by contrast, can genuinely be non-unique.

What is the relationship between invertibility and the matrix inverse?

For $T(\vec{x}) = A\vec{x}$ with $A$ square, $T$ is invertible exactly when $A$ is nonsingular, and $T^{-1}(\vec{x}) = A^{-1}\vec{x}$ . Composition of transformations corresponds to matrix multiplication, so the reversal rule $(S \circ T)^{-1} = T^{-1} \circ S^{-1}$ is the same statement as $(BA)^{-1} = A^{-1}B^{-1}$ .

Does the notation

T^{-1}(\vec{v})

presuppose that

T

is invertible?

No. The preimage $T^{-1}(\vec{v})$ is defined for any transformation as the set of inputs mapping to $\vec{v}$ , and it may be empty or infinite. The same symbol denotes the inverse function only when invertibility has been established, in which case every preimage is a single vector and the two readings agree.

My transformation is invertible but reconstruction is unusable. What went wrong?

Nothing algebraically. Invertibility is a yes-or-no property in exact arithmetic; usefulness depends on conditioning. If the condition number is $10^{k}$ , reconstruction can lose roughly $k$ significant digits, so an invertible transformation with a large condition number amplifies measurement noise beyond tolerance.

Must the domain and codomain have the same dimension?

Yes, for finite-dimensional spaces. Injectivity forces $\dim(U) \le \dim(V)$ and surjectivity forces $\dim(U) \ge \dim(V)$ , so invertibility forces equality. The converse fails: equal dimensions do not by themselves make a transformation invertible.

References

Beezer, R. A. A First Course in Linear Algebra, Version 0.70. University of Puget Sound, 2006. Section IVLT, subsection IVLT. Licensed under the GNU Free Documentation License v1.2.
ISO 80000-2:2019, Quantities and units — Part 2: Mathematics. International Organization for Standardization.
Halmos, P. R. Finite-Dimensional Vector Spaces, 2nd edition. Springer, 1974.
Higham, N. J. Accuracy and Stability of Numerical Algorithms, 2nd edition. Society for Industrial and Applied Mathematics, 2002.
Anderson, E. et al. LAPACK Users' Guide, 3rd edition. Society for Industrial and Applied Mathematics, 1999.

AI Suggested Questions

Give an explicit transformation with a left inverse but no right inverse, and show precisely which composition identity fails.
Derive the inverse of the map sending a cubic polynomial to its values at four distinct instants, and relate it to Lagrange interpolation.
Why is forming an explicit matrix inverse less accurate than solving with an LU factorisation, and by how much typically?
Show that the inverse of a composition reverses the order, and construct a case where forward-order undoing gives a wrong answer.
For a change of state coordinates with condition number 1e8, how much measurement precision survives the round trip?
Prove directly from the composition identities that the inverse of a linear transformation preserves scalar multiplication.

Related Calculators

Transformation Inverse Builder

Solve a linear transformation with a symbolic target to construct its inverse, and verify both composition identities automatically.

Matrix Inverse Calculator

Invert a square matrix by augmented reduction, with exact rational or floating-point arithmetic and a condition number estimate.

Invertibility Checker

Test a transformation for injectivity and surjectivity together and report the verdict with the supporting rank and conditioning data.