Conjugates of Complex Numbers | KEVOS® Mathematics

Overview

Conjugation reflects a complex number across the real axis: the real part is untouched and the imaginary part changes sign. Trivial as an operation, it is structurally central, because it is the only non-identity map from $\mathbb{C}$ to itself that preserves both arithmetic operations and fixes every real number. That single fact is what makes the conjugate the right tool wherever a complex computation must be made to yield a real result.

Three theorems carry the load. Conjugation respects addition, $\overline{c + d} = \overline{c} + \overline{d}$ ; it respects multiplication, $\overline{cd} = \overline{c}\,\overline{d}$ ; and applying it twice returns the original number, $\overline{\overline{c}} = c$ . Each is proved by writing the numbers in the form $a + bi$ and comparing parts, and together they mean that conjugating an entire expression is the same as conjugating each of its constituents. Any identity in complex numbers can therefore be conjugated wholesale to yield a second, generally distinct, identity for free.

The immediate consequences are the ones used daily. The product $c\overline{c} = a^{2} + b^{2}$ is real and non-negative, which is what makes complex division possible and what defines the modulus. The sum $c + \overline{c} = 2a$ and difference $c - \overline{c} = 2bi$ recover the two parts. And $\overline{c} = c$ holds exactly when $b = 0$ , so conjugation doubles as a test for reality — a test that generalises directly to matrices, where fixing the conjugate transpose is the definition of Hermitian.

In linear algebra proper, conjugation appears wherever a complex object must behave like its real counterpart. The inner product conjugates one of its arguments so that the inner product of a vector with itself is a non-negative real number and a norm can be defined. The adjoint conjugates as well as transposes, so that unitary matrices generalise orthogonal ones. And because a real polynomial is unchanged by conjugating its coefficients, the complex roots of a real characteristic polynomial arrive in conjugate pairs, with conjugate eigenvectors to match.

Definition

Conjugate of a Complex Number

CCN

For a complex number $c = a + bi$ with $a$ and $b$ real, the conjugate of $c$ is $\overline{c} = a - bi$ . The real part is unchanged and the imaginary part is negated.

Written $\overline{c}$ in mathematics and frequently $c^{*}$ in physics and engineering. The overline is used throughout this library, reserving the star for the matrix adjoint $A^{*}$ .

Conjugation as an Involution

CCT

Conjugation applied twice is the identity: $\overline{\overline{c}} = c$ for every complex number $c$ . An operation with this property is called an involution, and it follows that conjugation is a bijection of $\mathbb{C}$ onto itself which is its own inverse.

The involution property is what allows a conjugated equation to be conjugated back, and it is why conjugating both sides of an identity loses no information.

Real Number Test

CCN-RT

A complex number $c$ satisfies $\overline{c} = c$ if and only if $c$ is real. Equivalently, $c - \overline{c} = 0$ precisely when the imaginary part vanishes.

Every subsequent reality criterion in the subject is a version of this one: a matrix is Hermitian when $A^{*} = A$ , and a Hermitian matrix has real eigenvalues for exactly this reason.

Concepts

Conjugation respects addition and multiplication

Writing $c = a + bi$ and $d = r + si$ , the sum is $(a + r) + (b + s)i$ , whose conjugate is $(a + r) - (b + s)i = (a - bi) + (r - si)$ , which is $\overline{c} + \overline{d}$ . The product is $(ar - bs) + (as + br)i$ , whose conjugate is $(ar - bs) - (as + br)i$ ; expanding $\overline{c}\,\overline{d} = (a - bi)(r - si)$ gives $(ar - bs) - (as + br)i$ as well. Both proofs are pure bookkeeping, but the conclusion is strong: conjugation is a field automorphism of $\mathbb{C}$ . It commutes with every finite sum, product, power and quotient, so it may be pushed through an arbitrary rational expression term by term.

The three derived identities that get used

From the definition alone, $c + \overline{c} = 2a$ recovers twice the real part, $c - \overline{c} = 2bi$ recovers twice the imaginary part times $i$ , and $c\overline{c} = a^{2} + b^{2}$ is real, non-negative and zero only for $c = 0$ . The third is by far the most used: it converts a complex denominator into a real one, which is the entire mechanism of complex division, and its square root is the definition of the modulus. It also explains why conjugation, and not negation of the whole number, is the operation needed — $c \cdot (-c) = -c^{2}$ is not real in general.

Conjugation is conjugate-linear, not linear

Because $\overline{\alpha c} = \overline{\alpha}\,\overline{c}$ , conjugation preserves sums but pulls a conjugate out of a scalar multiple. It is therefore not a linear transformation of $\mathbb{C}$ viewed as a one-dimensional complex vector space; it is conjugate-linear, sometimes called antilinear. It is a linear transformation of $\mathbb{C}$ viewed as a two-dimensional real vector space, where it acts as reflection in the real axis. Confusing the two viewpoints is a common source of sign errors in proofs about inner products, which inherit the same asymmetry.

Conjugate roots of real polynomials

Let $p(x) = c_n x^{n} + \cdots + c_1 x + c_0$ have real coefficients and suppose $p(\lambda) = 0$ . Conjugating the whole equation and pushing the conjugate through every sum and product gives $\overline{c_n}\,\overline{\lambda}^{\,n} + \cdots + \overline{c_0} = 0$ . Since each $c_k$ is real, $\overline{c_k} = c_k$ , so $p(\overline{\lambda}) = 0$ as well. Non-real roots of a real polynomial therefore occur in conjugate pairs of equal multiplicity. Applied to a characteristic polynomial this says that a real matrix has complex eigenvalues in matched pairs, which is why oscillatory modes always come two at a time.

Conjugate eigenvectors accompany conjugate eigenvalues

The pairing extends to the eigenvectors. If $A$ has real entries and $A\vec{x} = \lambda \vec{x}$ , then conjugating every entry of both sides gives $\overline{A}\,\overline{\vec{x}} = \overline{\lambda}\,\overline{\vec{x}}$ , and $\overline{A} = A$ because the entries are real. So $\overline{\vec{x}}$ is an eigenvector for $\overline{\lambda}$ . The practical consequence is that only half the eigenvector computations need to be performed for a real matrix: the remainder are obtained by changing signs. The real and imaginary parts of $\vec{x}$ span a two-dimensional real subspace that $A$ maps into itself, which is the real substitute for a pair of complex eigendirections.

Why the inner product conjugates one argument

If the inner product on $\mathbb{C}^m$ were defined without conjugation, then a vector such as $\begin{bmatrix} 1 \\ i \end{bmatrix}$ would satisfy $\langle \vec{u}, \vec{u} \rangle = 1 + i^{2} = 0$ despite being non-zero, and no norm could be built from it. Conjugating one argument replaces each term $[\vec{u}]_k^{2}$ by $\overline{[\vec{u}]_k}\,[\vec{u}]_k = |[\vec{u}]_k|^{2}$ , restoring a sum of non-negative reals that vanishes only for the zero vector. The price is that the inner product is linear in one argument and conjugate-linear in the other, and that swapping the arguments conjugates the result.

Decision path: when to reach for the conjugate

Is there a complex number in a denominator?Multiply above and below by the conjugate of that denominator. The denominator becomes the real number

a^{2} + b^{2}

Does a quantity need to be real?Pair each factor with its conjugate. The product

c\overline{c}

is real and non-negative; this is how moduli, norms and inner products are constructed.

Do you need to test a number for reality?Compare

c

with

\overline{c}

. Equality is exactly the condition

b = 0

, and the same test defines Hermitian matrices.

Is a second root or eigenvector required?If the coefficients or matrix entries are real, conjugate the one you already have rather than repeating the computation.

Are you transposing a complex matrix?Decide deliberately between the plain transpose

A^{t}

and the adjoint

A^{*}

, which conjugates as well. Only the adjoint preserves inner-product structure.

Extract a real part or imaginary partUse

c + \overline{c} = 2a

and

c - \overline{c} = 2bi

when the parts are needed as separate real quantities.

Equations

Definition of the conjugate

EQ-CCN-01

\bar{c} = \bar{a + b i} = a - b i, a, b \in ℝ

The real part is fixed and the imaginary part is negated. Geometrically, reflection of the plane in the real axis.

Conjugation respects addition

EQ-CCN-02

\bar{c + d} = \bar{c} + \bar{d}

Extends to any finite sum by induction, which is what permits conjugating a polynomial or a matrix product term by term.

Conjugation respects multiplication

EQ-CCN-03

\bar{c d} = \bar{c} \bar{d}, \bar{(\frac{c}{d})} = \frac{\bar{c}}{\bar{d}} (d \neq 0)

Together with the addition rule this makes conjugation an automorphism of the field $\mathbb{C}$ fixing $\mathbb{R}$ pointwise.

Involution

EQ-CCN-04

\bar{\bar{c}} = c

Conjugation is its own inverse, so an equation may be conjugated and conjugated back with nothing lost.

Recovering the parts and forming a real product

EQ-CCN-05

c + \bar{c} = 2 a, c - \bar{c} = 2 b i, c \bar{c} = a^{2} + b^{2}

The third identity is the workhorse: real, non-negative, and zero only when $c$ is zero. Its square root is the modulus.

Test for a real number

EQ-CCN-06

\bar{c} = c \Leftrightarrow c \in ℝ

The scalar prototype of the Hermitian condition $A^{*} = A$ for matrices and of self-adjointness for operators.

Conjugate root pairing

EQ-CCN-07

p (x) = \sum_{k}^{=} c_{k} x^{k}, c_{k} \in ℝ, p (λ) = 0 \Rightarrow p (\bar{λ}) = 0

Non-real roots of a real polynomial occur in conjugate pairs of equal multiplicity, so a real matrix has paired complex eigenvalues.

Conjugation inside the inner product

EQ-CCN-08

⟨ \vec{u}, \vec{v} ⟩ = \sum_{k}^{=} \bar{[\vec{u}]_{k}} [\vec{v}]_{k}, ⟨ \vec{u}, \vec{u} ⟩ \geq 0

Conjugating the first argument is exactly what forces the self-inner-product to be a non-negative real number and permits a norm.

Variable Definitions

Symbols used on this page
Symbol	Name	Meaning	Domain / type
$c$	Complex number	A general element $a + bi$ of $\mathbb{C}$	complex scalar
$\bar{c}$	Conjugate	The reflection $a - bi$ of $c$ in the real axis	complex scalar
$a$	Real part	Unchanged by conjugation; recoverable as $(c + \overline{c})/2$	real number
$b$	Imaginary part	Negated by conjugation; recoverable as $(c - \overline{c})/(2i)$	real number
$λ$	Eigenvalue or polynomial root	Paired with $\overline{\lambda}$ whenever the coefficients are real	complex scalar
$A^{*}$	Adjoint of a matrix	Conjugate of the transpose, $[A^{*}]_{ij} = \overline{[A]_{ji}}$	n x m complex matrix
$\vec{x}$	Eigenvector	Complex column vector; $\overline{\vec{x}}$ is the eigenvector for $\overline{\lambda}$ when $A$ is real	vector in C^n
$p (x)$	Polynomial	Real-coefficient polynomial, typically a characteristic polynomial	polynomial over R

Worked Numerical Example

Problem statement

A two-state model of a damped mechanical resonator has state matrix $A = \begin{bmatrix} -3 & -2 \\ 2 & -3 \end{bmatrix}$ , with real entries. Find its eigenvalues and eigenvectors, obtaining the second pair by conjugation rather than by a second computation, and identify the real invariant plane.

Form the characteristic polynomial
The trace is $-6$ and the determinant is $(-3)(-3) - (-2)(2) = 9 + 4 = 13$ , so the characteristic polynomial is $x^{2} - (\mathrm{tr}\,A)x + \det A$ . Every coefficient is real, which is the hypothesis the conjugate root pairing requires.
$p_{A} (x) = \det (A - x I_{2}) = x^{2} + 6 x + 13$
Solve for the eigenvalues
The discriminant is $36 - 52 = -16$ , negative, so there are no real roots. Over $\mathbb{C}$ the roots exist and, as the pairing theorem predicts, they are conjugates of one another.
$λ = \frac{- 6 \pm \sqrt{- 16}}{2} = - 3 \pm 2 i, λ_{2} = \bar{λ_{1}}$
Compute one eigenvector
For $\lambda_1 = -3 + 2i$ , form $A - \lambda_1 I_2$ . Dividing the first row by $-2$ gives $i x_1 + x_2 = 0$ , so $x_2 = -i x_1$ . Taking $x_1 = 1$ yields the eigenvector below; the second row is then automatically satisfied.
$A - λ_{1} I_{2} = [\begin{matrix} - 2 i & - 2 \\ 2 & - 2 i \end{matrix}], {\vec{x}}_{1} = [\begin{matrix} 1 \\ - i \end{matrix}]$
Obtain the second eigenvector by conjugation
Conjugate every entry of $A\vec{x}_1 = \lambda_1 \vec{x}_1$ . Conjugation respects both addition and multiplication, so it passes through the matrix-vector product entry by entry; and $\overline{A} = A$ because $A$ is real. No second null space computation is needed.
$A \bar{{\vec{x}}_{1}} = \bar{A {\vec{x}}_{1}} = \bar{λ_{1} {\vec{x}}_{1}} = \bar{λ_{1}} \bar{{\vec{x}}_{1}}, \bar{{\vec{x}}_{1}} = [\begin{matrix} 1 \\ i \end{matrix}]$
Verify both pairs directly
Multiply out to confirm. For the first pair, $A\vec{x}_1 = (3 + 2i,\ 2 - 3i)$ and $\lambda_1 \vec{x}_1 = (-3+2i)(1,\ -i) = (-3 + 2i,\ 2 + 3i)$ . A discrepancy would appear here immediately, so the check is worth doing explicitly.
$A [\begin{matrix} 1 \\ - i \end{matrix}] = [\begin{matrix} - 3 + 2 i \\ 2 + 3 i \end{matrix}] = (- 3 + 2 i) [\begin{matrix} 1 \\ - i \end{matrix}]$
Extract the real invariant plane
Split the eigenvector into real and imaginary parts, $\vec{x}_1 = \vec{p} + i\vec{q}$ with $\vec{p} = (1, 0)$ and $\vec{q} = (0, -1)$ . Because $\vec{x}_1$ and $\overline{\vec{x}_1}$ span the same subspace as $\vec{p}$ and $\vec{q}$ , and that span is closed under $A$ , the pair $\{\vec{p}, \vec{q}\}$ is a real basis of an $A$ -invariant plane.
$\vec{p} = \frac{{\vec{x}}_{1} + \bar{{\vec{x}}_{1}}}{2}, \vec{q} = \frac{{\vec{x}}_{1} - \bar{{\vec{x}}_{1}}}{2 i}$

Result

The resonator has a single conjugate pair of eigenvalues, $-3 \pm 2i$ : the real part $-3$ is the decay rate and the imaginary part $2$ is the damped angular frequency, so the response is an oscillation at $2\ \mathrm{rad/s}$ whose amplitude decays with time constant $1/3\ \mathrm{s}$ . Half the eigen-analysis was obtained free of charge by conjugation, and the real and imaginary parts of the single computed eigenvector give a real coordinate frame in which the motion is a decaying rotation — the form an engineer needs when the model must stay in real arithmetic.

Applications & Industry Use

Mechanical and structural dynamics

Damped modes as conjugate pairs

The state matrix of a damped structure is real, so its complex eigenvalues occur in conjugate pairs. Each pair corresponds to one physical mode, with the real part giving the decay rate and the imaginary part the damped frequency. Modal software exploits the pairing to store and report half the spectrum, and to reconstruct real mode shapes from the real and imaginary parts of a single complex eigenvector.

Digital signal processing

Hermitian symmetry of real signal spectra

The discrete Fourier transform of a real sequence satisfies $X[N-k] = \overline{X[k]}$ , a direct consequence of conjugation passing through sums and products. Real-input transform routines exploit this to compute and store only half the spectrum, roughly halving both the operation count and the memory footprint of a spectral analysis pipeline.

Control systems

Pole pairs and second-order behaviour

A real transfer function has poles that are either real or in conjugate pairs. Each pair contributes a second-order factor with real coefficients, which is why controller design is organised around second-order sections. Root-locus and pole-placement tools plot only the upper half-plane because the lower half is determined by conjugation.

Quantum and computational physics

Self-adjoint operators and real observables

Physical observables are modelled by operators satisfying $A^{*} = A$ , the matrix generalisation of the scalar test $\overline{c} = c$ . That single condition forces every eigenvalue to be real, which is precisely the requirement that a measured quantity be a real number, and it forces eigenvectors for distinct eigenvalues to be orthogonal.

Communications engineering

Matched filtering and correlation

The matched filter for a complex baseband waveform is its time-reversed conjugate, and correlation of complex sequences conjugates one argument for exactly the reason the inner product does: without it the self-correlation at zero lag would not be a real, non-negative energy. Hardware correlators implement the conjugation as a sign inversion on the quadrature path.

Numerical linear algebra

Choosing between transpose and adjoint

Library routines expose both an unconjugated and a conjugated form of essentially every operation — dot products, matrix transposes, factorisations. Selecting the unconjugated form where the adjoint was intended produces results that are silently wrong for complex data and identical for real data, so the defect survives every real-valued test case.

Design Considerations

Fix the notation and never mix it

The overline and the star both denote conjugation of a scalar, but the star is also used for the matrix adjoint and, in some literature, for the plain transpose. Within a single document choose $\overline{c}$ for scalars and $A^{*}$ for the matrix adjoint, and state the choice. Ambiguity here produces sign errors that survive review because both readings look plausible.

Decide transpose versus adjoint at the point of definition

For real data the two coincide, so the choice appears not to matter and is often made carelessly. For complex data only the adjoint preserves inner products, makes $A^{*}A$ positive semi-definite and generalises orthogonality to unitarity. Define every complex operation in terms of the adjoint unless there is a specific reason to use the plain transpose, and document that reason where it occurs.

Exploit the pairing to halve the work

For a real matrix, half of the complex eigenvalues and eigenvectors are determined by the other half. Computing both independently wastes effort and, worse, invites the two halves to disagree at the last significant digit, producing an apparent asymmetry that then has to be explained. Compute one member of each pair and conjugate.

Keep a real formulation available

Many downstream tools — embedded targets, real-only solvers, plotting pipelines — cannot accept complex data. The real and imaginary parts of a complex eigenvector span a real invariant plane, so a real basis is always recoverable. Plan the conversion at the interface rather than discovering it after the fact.

Treat conjugate-linearity as a design constraint

Conjugation is additive but pulls a conjugate out of a scalar multiple, so any construction built on it inherits an asymmetry. The inner product is linear in one argument and conjugate-linear in the other; swapping the argument order conjugates the answer. Fix the convention for which argument is conjugated once, at the top of a library, because reversing it changes signs throughout.

Assert reality rather than discarding it

Quantities that must be real by construction — a modulus squared, the diagonal of a Hermitian matrix, total energy — will carry a small imaginary residue in floating point. Silently taking the real part hides genuine errors. Assert that the residue is below a tolerance scaled to the magnitude, then discard it.

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	Specifies the overline as the notation for the complex conjugate and distinguishes it from the star used for adjoints, together with the upright setting of $\mathrm{Re}$ and $\mathrm{Im}$ .
`BLAS Level 1`	Basic Linear Algebra Subprograms, vector operations	Provides `zdotc`, which conjugates its first argument, alongside `zdotu`, which does not. Choosing the wrong one is undetectable on real test data and is the classic conjugation defect in numerical code.
`ISO/IEC 9899 (C)`	Programming languages — C, complex.h	Defines `conj`, `conjf` and `conjl` and specifies that conjugation is exact, being a sign change on the imaginary component with no rounding.
`IEEE 754-2019`	IEEE Standard for Floating-Point Arithmetic	Governs the signed-zero behaviour that makes the conjugate of $a + 0i$ distinguishable from $a - 0i$ at the bit level, which matters for branch-cut consistency in complex elementary functions.
`ISO/IEC 40314`	Mathematical Markup Language (MathML) Version 3.0	Encodes the overline as a semantic accent rather than a drawn rule, so that a screen reader announces a conjugate as such instead of describing a line above a symbol.

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 Gaussian rationals	Symbolic verification that a conjugate pair is exact, or certification of a Hermitian structure.	Conjugation is exact and free; the limitation is the usual growth of rational numerators and denominators in the surrounding computation.
IEEE 754 binary64 pair	General engineering computation with complex eigenvalues, spectra or impedances.	Conjugation costs one sign flip and introduces no rounding at all, so identities involving only conjugation hold exactly; identities involving arithmetic do not.
IEEE 754 binary32 pair	Large spectral transforms and GPU pipelines where bandwidth dominates.	Same exactness for conjugation itself, but the conjugate-symmetry checks used to validate a real-input transform must be applied with a tolerance around seven digits rather than sixteen.
Half-spectrum storage using Hermitian symmetry	Transforms of real-valued data, where $X[N-k] = \overline{X[k]}$ holds by construction.	Halves memory and operation count, but the packed layout differs between libraries and an incorrect unpacking produces a plausible-looking but wrong spectrum.
Packed Hermitian matrix storage	Large Hermitian systems where only one triangle need be stored, the other implied by conjugation.	Nearly halves memory and guarantees the Hermitian property exactly, at the cost of less regular memory access and reduced cache efficiency.
Real two-by-two block representation	The toolchain accepts only real data, or the deliverable must be a real invariant subspace.	Conjugation becomes a transpose of the block, keeping everything real, but the dimension doubles and complex structure is no longer visible to the solver.

Manufacturing Notes

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

Cost and exactness

Conjugation is a single sign inversion on the imaginary component. It performs no arithmetic, introduces no rounding and cannot overflow, so in floating point it is one of very few operations that is exact for every input. Identities such as $\overline{\overline{c}} = c$ therefore hold bit for bit, which makes conjugation useful as a zero-cost consistency check inside longer computations.

Signed zero and branch cuts

Under IEEE 754 the values $+0$ and $-0$ are distinguishable, so $\overline{a + 0i}$ is $a - 0i$ and not $a + 0i$ . This matters for complex elementary functions, whose branch cuts are defined so that the sign of a zero imaginary part selects which side of the cut is taken. Code that normalises negative zero away can silently move a result to the wrong branch.

The conjugated versus unconjugated dot product

BLAS exposes zdotc for $\sum \overline{u_k} v_k$ and zdotu for $\sum u_k v_k$ . NumPy's numpy.dot does not conjugate, whereas numpy.vdot conjugates its first argument; MATLAB's ' operator conjugates while .' does not. These defaults differ by ecosystem and are the single most productive source of complex-arithmetic defects, because every real-valued test passes.

Verifying complex work with conjugate identities

Three cheap assertions catch most errors. First, $c\overline{c}$ must have zero imaginary part. Second, for a real input matrix the computed spectrum must be closed under conjugation to within tolerance. Third, for a Hermitian matrix the computed diagonal must be real. Each is a one-line check that fails loudly on a mis-selected transpose or a dropped conjugate.

Hand computation guidance

When conjugating a compound expression, push the bar inwards one operation at a time rather than rewriting the whole expression at once: the conjugate of a quotient is the quotient of the conjugates, the conjugate of a product is the product of the conjugates, and only at the leaves is the definition applied. Attempting to conjugate a large expression in a single step is where terms get missed.

Failure Modes & Common Mistakes

Failure modes, root causes and prevention
Failure mode / mistake	Impact	Root cause	Prevention & detection
Conjugating only some factors of a product	high	Applying the bar to part of an expression while leaving another part untouched, particularly inside a matrix product.	Push the conjugate inwards one operation at a time. Every factor and every term must receive it, since conjugation distributes over both operations.
Using the transpose where the adjoint is required	high	Writing $A^{t}$ out of habit for complex data, so the conjugation is omitted.	Default to the adjoint $A^{*}$ for complex matrices. Test with a matrix having genuinely complex entries; real test data cannot distinguish the two.
Calling the unconjugated dot product	high	Using `zdotu`, `numpy.dot` or `.'` where an inner product with conjugation was intended.	Choose the conjugating variant explicitly at every call site and include one complex-valued unit test whose expected answer differs between the two.
Treating conjugation as a linear map over $\mathbb{C}$	medium	Assuming $\overline{\alpha c} = \alpha \overline{c}$ , forgetting that the scalar is conjugated too.	Remember conjugate-linearity: scalars come out conjugated. This is inherited by the inner product and by every construction built on it.
Confusing conjugation with negation	medium	Negating the whole number, giving $-a - bi$ , rather than only the imaginary part.	Check against the defining property: $c\overline{c}$ must be real and non-negative. The product $c(-c)$ is not.
Computing both members of a conjugate pair independently	low	Repeating a null space calculation for $\overline{\lambda}$ when the matrix has real entries.	Conjugate the eigenvector already found. Independent computation wastes effort and can produce two answers that disagree in the last digits.
Assuming conjugate pairing for a complex matrix	medium	Applying the conjugate root theorem when the coefficients or matrix entries are themselves complex, so the hypothesis fails.	Check that every coefficient is real before invoking the pairing. A genuinely complex matrix can have a single isolated non-real eigenvalue with no partner.
Silently discarding an imaginary residue	medium	Taking the real part of a quantity that ought to be real, without checking how large the discarded part was.	Assert that the imaginary part is below a tolerance scaled to the magnitude before discarding it; a large residue indicates an upstream error, not rounding.

FAQs

Why is the conjugate the right tool for making a denominator real?

Because $c\overline{c} = a^{2} + b^{2}$ is always real, always non-negative, and zero only when $c$ is zero. No other simple companion to $c$ has all three properties: the negative $-c$ gives $-c^{2}$ , which is generally complex, and the reciprocal begs the question. Multiplying above and below by the conjugate is therefore the unique elementary way to clear a complex denominator.

\overline{c}

the same as

c^{*}

For a scalar, yes — the two notations denote the same operation, with the overline favoured in mathematics and the star in physics and engineering. For matrices they diverge: $A^{*}$ almost always means the conjugate transpose, whereas $\overline{A}$ means entrywise conjugation without transposition. This library uses the overline for scalars and entrywise conjugation, and the star for the adjoint.

Do complex eigenvalues always come in conjugate pairs?

Only when the matrix has real entries, because the pairing follows from the characteristic polynomial having real coefficients. A matrix with genuinely complex entries can have an isolated non-real eigenvalue with no conjugate partner in its spectrum. The hypothesis is easy to lose sight of because most engineering matrices are real.

Why does the inner product conjugate one argument and not both?

Conjugating both would leave the self-inner-product still complex in general, and conjugating neither allows a non-zero vector to have zero self-inner-product. Conjugating exactly one turns each term into a squared modulus, so the self-inner-product is a sum of non-negative reals. The cost is an asymmetry: the form is linear in one argument and conjugate-linear in the other.

What does it mean that conjugation is an automorphism of

\mathbb{C}

It means conjugation is a bijection of $\mathbb{C}$ onto itself that preserves both addition and multiplication, and fixes every real number. Consequently any true statement about complex numbers, expressed purely in terms of arithmetic, remains true after conjugating every quantity in it. That is the licence behind the conjugate root theorem and the conjugate eigenvector result.

Is conjugation exact in floating-point arithmetic?

Yes. It flips one sign bit and performs no arithmetic, so it cannot round, overflow or underflow. Identities involving conjugation alone hold bit for bit. The only subtlety is signed zero: the conjugate of $a + 0i$ is $a - 0i$ , which matters for the branch cuts of complex elementary functions.

How do I turn a complex conjugate pair of eigenvectors into something real?

Take the real and imaginary parts of one eigenvector separately. Those two real vectors span the same subspace as the complex pair, and that subspace is mapped into itself by the matrix. In that real basis the transformation acts as a scaling combined with a rotation, with the rate and frequency read from the real and imaginary parts of the eigenvalue.

References

Beezer, R. A. A First Course in Linear Algebra, Version 0.70. University of Puget Sound, 2006. Section CNO, Subsection CCN. Licensed under the GNU Free Documentation License v1.2.
ISO 80000-2:2019, Quantities and units — Part 2: Mathematics. International Organization for Standardization.
Lawson, C. L., Hanson, R. J., Kincaid, D. R. and Krogh, F. T. Basic Linear Algebra Subprograms for Fortran Usage. ACM Transactions on Mathematical Software, volume 5, 1979.
Horn, R. A. and Johnson, C. R. Matrix Analysis, 2nd edition. Cambridge University Press, 2013. Adjoints, Hermitian matrices and conjugate symmetry.
Kahan, W. Branch Cuts for Complex Elementary Functions. In The State of the Art in Numerical Analysis, Oxford University Press, 1987. Signed zero and conjugation.

AI Suggested Questions

Show me a complex matrix with an isolated non-real eigenvalue whose conjugate is not in the spectrum, and explain which hypothesis fails.
Derive the Hermitian symmetry of the discrete Fourier transform of a real sequence directly from the conjugation theorems.
What exactly goes wrong if the inner product on complex vectors is defined without conjugating either argument?
Given a complex eigenvector of a real matrix, construct the real two-dimensional invariant subspace and the real matrix that acts on it.
Compare zdotc against zdotu on a worked complex example where the two answers differ, and identify which one a norm calculation needs.
How does conjugate symmetry let a real-input FFT halve both its operation count and its storage?

Related Calculators

Complex Conjugate Calculator

Compute conjugates, real and imaginary parts, and the products $c\overline{c}$ , $c + \overline{c}$ and $c - \overline{c}$ for any complex input.

Conjugate Root Finder

Factor a real-coefficient polynomial and group its roots into real roots and conjugate pairs with their multiplicities.

Adjoint & Transpose Calculator

Compute $A^{t}$ , $\overline{A}$ and $A^{*}$ side by side and test whether a matrix is symmetric, Hermitian or unitary.