Inner Products of Vectors | KEVOS® Mathematics

Overview

Almost every operation so far has taken vectors to vectors. The inner product is different: it consumes two vectors and produces a single scalar. That change of type is what makes it the gateway to geometry. Length, distance, angle, orthogonality, projection and least squares are all defined from it, and none of them is available without it.

For real vectors the construction is the familiar dot product, the sum of products of matching entries. For complex vectors that formula fails badly: applied to $\vec{u} = (1, i)$ it returns $1 + i^2 = 0$ , so a non-zero vector would have zero length. The repair is to conjugate the entries of one argument before multiplying, which turns each term into a squared modulus when the two arguments coincide.

The price of the repair is a loss of symmetry. Swapping the arguments conjugates the result rather than leaving it unchanged, and pulling a scalar out of the second argument conjugates the scalar. The inner product is therefore sesquilinear — linear in one slot, conjugate-linear in the other — and the accounting must be done carefully. Over $\mathbb{R}$ all of this collapses into ordinary bilinearity and symmetry.

Which argument carries the conjugate is a convention, and it differs across texts, libraries and disciplines. This library conjugates the second argument, so $\langle \cdot, \cdot \rangle$ is linear in its first slot. The mathematics is identical either way, but mixing conventions inside one derivation or one codebase produces conjugates in the wrong places, and the resulting errors are invisible on real data.

Definition

Inner Product

For $\vec{u}, \vec{v} \in \mathbb{C}^m$ , the inner product of $\vec{u}$ and $\vec{v}$ is the complex scalar

$\langle \vec{u}, \vec{v} \rangle = [\vec{u}]_1 \overline{[\vec{v}]_1} + [\vec{u}]_2 \overline{[\vec{v}]_2} + \cdots + [\vec{u}]_m \overline{[\vec{v}]_m}$

equivalently

\langle \vec{u}, \vec{v} \rangle = \sum_{i=1}^{m} [\vec{u}]_i \overline{[\vec{v}]_i}

. Both vectors must have the same size; the result is a scalar, not a vector.

When every entry is real the conjugates do nothing and the definition reduces exactly to the dot product of school and calculus courses.

Sesquilinear Form

SLF

A scalar-valued function of two vector arguments that is additive in each argument, linear under scalar multiplication in one argument, and conjugate-linear in the other. The prefix means "one and a half": one full linearity plus one twisted one. The complex inner product is the archetype.

Concepts

Additivity in each argument

The inner product distributes over vector addition on both sides: $\langle \vec{u} + \vec{v}, \vec{w} \rangle = \langle \vec{u}, \vec{w} \rangle + \langle \vec{v}, \vec{w} \rangle$ and $\langle \vec{u}, \vec{v} + \vec{w} \rangle = \langle \vec{u}, \vec{v} \rangle + \langle \vec{u}, \vec{w} \rangle$ . Both proofs are entrywise: expand the definition, use the entrywise definition of vector addition, and appeal to distributivity in $\mathbb{C}$ .

The second identity carries an extra step. The sum sits inside a conjugate, so the additivity of conjugation must be invoked before distributivity applies. Additivity of conjugation is unconditional, so the conclusion has the same clean form as the first — unlike the scalar rule below.

Scalars, and where the conjugate appears

A scalar pulled out of the first argument emerges unchanged, $\langle \alpha\vec{u}, \vec{v} \rangle = \alpha \langle \vec{u}, \vec{v} \rangle$ , because nothing conjugates it. A scalar pulled out of the second emerges conjugated, $\langle \vec{u}, \alpha\vec{v} \rangle = \overline{\alpha} \langle \vec{u}, \vec{v} \rangle$ , because it passes under the bar in every term.

Combining additivity with these rules gives the working statement: the inner product is linear in the first argument and conjugate-linear in the second. When a whole linear combination sits in the second slot, every coefficient conjugates as it comes out. This single asymmetry accounts for the majority of sign and phase errors in complex derivations.

Conjugate symmetry

Interchanging the arguments conjugates the value: $\langle \vec{u}, \vec{v} \rangle = \overline{\langle \vec{v}, \vec{u} \rangle}$ . The proof conjugates the defining sum, uses that conjugation distributes over sums and products, and uses that double conjugation is the identity, leaving the terms of the opposite ordering.

The identity has two immediate consequences. First, orthogonality is symmetric: if $\langle \vec{u}, \vec{v} \rangle = 0$ then $\langle \vec{v}, \vec{u} \rangle = \overline{0} = 0$ , so the order of testing never matters. Second, taking $\vec{v} = \vec{u}$ gives $\langle \vec{u}, \vec{u} \rangle = \overline{\langle \vec{u}, \vec{u} \rangle}$ , which forces that quantity to be real — a fact needed before it can be compared with zero.

Positivity of a vector with itself

Setting both arguments equal collapses each term to $[\vec{u}]_i \overline{[\vec{u}]_i} = |[\vec{u}]_i|^2$ , a non-negative real number. Hence $\langle \vec{u}, \vec{u} \rangle$ is a sum of squared moduli: real, non-negative, and zero only when every modulus is zero, which happens only for $\vec{u} = \vec{0}$ .

This property, called positive definiteness, is what licenses the definition of a norm as $\sqrt{\langle \vec{u}, \vec{u} \rangle}$ . Without the conjugate in the definition it fails outright: the naive sum of squares of $(1, i)$ is zero for a non-zero vector, and any "length" built on it would be useless.

Matrix form and the adjoint

The inner product can be written as a matrix product, $\langle \vec{u}, \vec{v} \rangle = \vec{v}^{*} \vec{u}$ , where $\vec{v}^{*} = \overline{\vec{v}}^{\,t}$ is the adjoint of $\vec{v}$ , a $1 \times m$ row. The result is a $1 \times 1$ matrix identified with its single entry.

This form is how inner products are actually implemented and how they generalise. It shows immediately that $\langle A\vec{u}, \vec{v} \rangle = \langle \vec{u}, A^{*}\vec{v} \rangle$ , the defining property of the adjoint, and it is the reason unitary matrices — those with $A^{*}A = I$ — are exactly the matrices that preserve every inner product.

Relation to the dot product, and to physical quantities

For real vectors the conjugates vanish and $\langle \vec{u}, \vec{v} \rangle$ is the dot product $\sum u_i v_i$ , so all the geometry familiar from two and three dimensions extends unchanged to $\mathbb{R}^m$ . The complex case is a strict generalisation, not a different object.

In engineering the inner product is rarely abstract. It is the correlation between two sampled signals, the modal participation of a mode shape in a load, the complex power delivered by a voltage-current phasor pair, and the amplitude of one quantum state in another. In each case the conjugate is what makes the self-inner-product an energy rather than a meaningless complex number.

Procedure: evaluate and sanity-check an inner product

Confirm the sizes matchBoth vectors must lie in the same

\mathbb{C}^m

. An inner product between vectors of different sizes is undefined, not zero.

Fix the conventionDecide, and state, which argument carries the conjugate. This library conjugates the second, so the form is linear in the first.

Conjugate the second vectorFlip the sign of every imaginary part of

\vec{v}

. This step costs nothing and is exact in floating point.

Accumulate the productsMultiply matching entries and sum. Use a compensated or extended-precision accumulator when

m

is large or the entries vary widely in magnitude.

Check the invariantsRecompute with the arguments swapped and confirm the answer is the conjugate. Compute

\langle \vec{u}, \vec{u} \rangle

and confirm it is real and non-negative.

Interpret the valueZero means the vectors are orthogonal. A non-zero value gives the correlation, participation or projection weight, depending on the application.

Equations

Definition of the inner product

EQ-IP-01

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

The conjugate falls on the second argument. Both vectors have $m$ entries and the result is a single complex scalar.

Additivity in each argument

EQ-IP-02

⟨ \vec{u} + \vec{v}, \vec{w} ⟩ = ⟨ \vec{u}, \vec{w} ⟩ + ⟨ \vec{v}, \vec{w} ⟩, ⟨ \vec{u}, \vec{v} + \vec{w} ⟩ = ⟨ \vec{u}, \vec{v} ⟩ + ⟨ \vec{u}, \vec{w} ⟩

Both hold without any conjugation appearing, because conjugation is itself additive.

Scalar multiplication rules

EQ-IP-03

⟨ α \vec{u}, \vec{v} ⟩ = α ⟨ \vec{u}, \vec{v} ⟩, ⟨ \vec{u}, α \vec{v} ⟩ = \bar{α} ⟨ \vec{u}, \vec{v} ⟩

Linear in the first slot, conjugate-linear in the second. Together with additivity this is sesquilinearity.

Conjugate symmetry

EQ-IP-04

⟨ \vec{u}, \vec{v} ⟩ = \bar{⟨ \vec{v}, \vec{u} ⟩}

Swapping the arguments conjugates the value. Consequently orthogonality is a symmetric relation and $\langle \vec{u}, \vec{u} \rangle$ must be real.

Positive definiteness

EQ-IP-05

⟨ \vec{u}, \vec{u} ⟩ = \sum_{i}^{=} | [\vec{u}]_{i} |^{2} \geq 0, ⟨ \vec{u}, \vec{u} ⟩ = 0 \Leftrightarrow \vec{u} = \vec{0}

A sum of squared moduli. This is the property that allows a norm to be defined and is destroyed if the conjugate is omitted.

Matrix form via the adjoint

EQ-IP-06

⟨ \vec{u}, \vec{v} ⟩ = {\vec{v}}^{*} \vec{u} = {\bar{\vec{v}}}^{t} \vec{u}

The implementation form. It exposes the adjoint relation $\langle A\vec{u}, \vec{v} \rangle = \langle \vec{u}, A^{*}\vec{v} \rangle$ directly.

Real case reduces to the dot product

EQ-IP-07

\vec{u}, \vec{v} \in ℝ^{m} \Rightarrow ⟨ \vec{u}, \vec{v} ⟩ = \sum_{i}^{=} u_{i} v_{i} = \vec{u} \cdot \vec{v}

Every conjugate is inert on real data, so the complex definition is a strict extension of the familiar one.

Variable Definitions

Symbols used on this page
Symbol	Name	Meaning	Domain / type
$\vec{u}$	First argument	The vector in the linear slot of the inner product	vector in C^m
$\vec{v}$	Second argument	The vector in the conjugate-linear slot; its entries are conjugated	vector in C^m
$⟨ \vec{u}, \vec{v} ⟩$	Inner product	The complex scalar produced from the two vectors	element of C
$[\vec{u}]_{i}$	Vector entry	The $i$ -th component of $\vec{u}$	complex scalar
$\bar{[\vec{v}]_{i}}$	Conjugated entry	The complex conjugate of the $i$ -th component of $\vec{v}$	complex scalar
$α$	Scalar	A multiplier; conjugated when extracted from the second argument	element of C
$m$	Vector size	Common number of entries of the two vectors	positive integer
${\vec{v}}^{*}$	Adjoint of a vector	The $1 \times m$ row obtained by conjugating and transposing $\vec{v}$	1 x m matrix over C

Worked Numerical Example

Problem statement

Two receiver channels sample a three-element antenna array, giving complex baseband vectors. Compute their inner product, verify conjugate symmetry and the two scalar rules, and evaluate the self-inner-product that represents channel energy.

State the two channel vectors
Each entry is one antenna element's complex baseband sample.
$\vec{u} = [\begin{matrix} 2 + i \\ 3 - 2 i \\ - 1 + 4 i \end{matrix}], \vec{v} = [\begin{matrix} 1 - i \\ 2 + 2 i \\ 3 + i \end{matrix}]$
Conjugate the second vector and multiply
The conjugated entries are $1 + i$ , $2 - 2i$ and $3 - i$ . Term by term: $(2+i)(1+i) = 1 + 3i$ ; $(3-2i)(2-2i) = 6 - 6i - 4i + 4i^2 = 2 - 10i$ ; $(-1+4i)(3-i) = -3 + i + 12i - 4i^2 = 1 + 13i$ .
$⟨ \vec{u}, \vec{v} ⟩ = (1 + 3 i) + (2 - 10 i) + (1 + 13 i) = 4 + 6 i$
Verify conjugate symmetry
Now conjugate $\vec{u}$ instead: $(1-i)(2-i) = 1 - 3i$ ; $(2+2i)(3+2i) = 6 + 4i + 6i + 4i^2 = 2 + 10i$ ; $(3+i)(-1-4i) = -3 - 12i - i - 4i^2 = 1 - 13i$ . The sum is the conjugate of the previous answer, as required.
$⟨ \vec{v}, \vec{u} ⟩ = (1 - 3 i) + (2 + 10 i) + (1 - 13 i) = 4 - 6 i = \bar{⟨ \vec{u}, \vec{v} ⟩}$
Scale the first argument
Take $\alpha = 2 + i$ as a receiver gain applied to channel $\vec{u}$ . The rule predicts the inner product is multiplied by $\alpha$ unchanged: $(2+i)(4+6i) = 8 + 12i + 4i + 6i^2 = 2 + 16i$ . Direct computation on $\alpha\vec{u}$ gives the same value.
$⟨ α \vec{u}, \vec{v} ⟩ = (2 + i) (4 + 6 i) = 2 + 16 i$
Scale the second argument
Applying the same gain to channel $\vec{v}$ instead brings out the conjugate $\overline{\alpha} = 2 - i$ : $(2-i)(4+6i) = 8 + 12i - 4i - 6i^2 = 14 + 8i$ . The two results differ, which is the observable content of conjugate-linearity.
$⟨ \vec{u}, α \vec{v} ⟩ = (2 - i) (4 + 6 i) = 14 + 8 i$
Compute the channel energy
Setting both arguments to $\vec{u}$ turns every term into a squared modulus: $|2+i|^2 = 5$ , $|3-2i|^2 = 13$ , $|-1+4i|^2 = 17$ . The total is real and positive, as positive definiteness requires.
$⟨ \vec{u}, \vec{u} ⟩ = 5 + 13 + 17 = 35$

Result

The channels are correlated with complex value $4 + 6i$ , so they are not orthogonal; the magnitude $\sqrt{52} \approx 7.21$ measures how strongly they overlap and the argument gives the relative phase of that overlap. Applying a complex gain to the reference channel rather than the measured one changes the correlation to its conjugate-weighted value, which is why matched-filter weights must be conjugated before use. The self-inner-product $35$ is the total energy across the three elements, real and positive by construction.

Applications & Industry Use

Radar & array processing

Matched filtering and beamforming

A matched filter correlates a received vector against a reference by inner product, and the optimal beamforming weight vector is the conjugate of the steering vector. Conjugate-linearity is exactly why the conjugate appears: it aligns the phases across elements so the contributions add coherently rather than cancelling.

Electrical power engineering

Complex power from phasor pairs

Apparent power at a bus is computed as the product of a voltage phasor with the conjugate of a current phasor, and aggregated over a network this is an inner product of phasor vectors. Real and reactive power appear as the real and imaginary parts, so the conjugate placement determines the sign convention for reactive flow.

Structural dynamics

Modal participation factors

The contribution of a mode shape to a given load distribution is an inner product of the two vectors, weighted by mass. Orthogonality of modes — a zero inner product under that weighting — is what decouples the equations of motion and makes modal superposition tractable.

Quantum engineering

Transition amplitudes and probabilities

The amplitude for one state to be observed in another is their inner product, and the observed probability is its squared modulus. Conjugate symmetry guarantees that the two directions of a transition have equal probability, while positive definiteness makes normalisation possible.

Machine learning

Similarity and kernel evaluation

Cosine similarity, linear kernels and attention scores are all inner products between feature vectors, followed by a normalisation. Where features are complex — as in spectral or phase-aware models — the conjugate is essential, otherwise the self-similarity of a vector need not even be real.

Communications

Symbol detection by correlation

A coherent receiver decides which symbol was sent by correlating the received vector with each candidate waveform. Positive definiteness ensures the energy terms in the resulting decision metric are real, and conjugate symmetry ensures the correlation matrix of the candidate set is Hermitian.

Design Considerations

State the conjugation convention explicitly

Mathematics texts often conjugate the first argument; physics conventions almost always do, following the bra-ket notation. This library conjugates the second. Neither is more correct, but a derivation that silently switches midway is wrong, and the mistake produces a conjugated answer rather than an obviously broken one. Write the convention down at the top of any derivation or module.

Never omit the conjugate for convenience

The unconjugated bilinear form $\sum u_i v_i$ is a legitimate object and appears in some engineering formulas, but it is not an inner product: it fails positive definiteness, so it defines no norm and no meaningful orthogonality. If a formula genuinely calls for it, name it separately and never route it through inner-product machinery.

Watch accumulation error for long vectors

An inner product of length $m$ has a worst-case relative error growing with $m$ in naive summation, and cancellation between large terms of opposite sign can leave few correct digits. Use pairwise or compensated summation, or accumulate in a wider precision than the operands, whenever $m$ is large or the entries span many orders of magnitude.

Weighted inner products change the geometry

Many applications use $\langle \vec{u}, \vec{v} \rangle_M = \vec{v}^{*} M \vec{u}$ for a Hermitian positive definite $M$ — mass-weighted modal analysis is the standard example. All the properties on this page carry over, but orthogonality and length are now measured with respect to $M$ , and results are not comparable to those from the unweighted form.

Zero has a precise meaning

An inner product of exactly zero means orthogonality. A small non-zero value means near-orthogonality, and how small counts as small depends on the magnitudes of the two vectors. Always compare $|\langle \vec{u}, \vec{v} \rangle|$ against a scale such as the product of the norms rather than against an absolute constant.

Prefer the library kernel over a hand-written loop

Vendor implementations of the inner product exploit vectorisation, fused multiply-add and blocked accumulation, and are typically several times faster and more accurate than a naive loop. They also make the conjugation choice explicit in the routine name, which documents the convention at every call site.

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 angle-bracket notation for scalar products, the overline for conjugation and the asterisk for the adjoint, keeping the three operations typographically distinct.
`BLAS Level 1`	Basic Linear Algebra Subprograms, vector operations	Defines `xDOT` for real vectors and the pair `CDOTU` and `CDOTC` for complex ones, which differ exactly in whether the first argument is conjugated.
`IEEE 754-2019`	IEEE Standard for Floating-Point Arithmetic	Defines the rounding of each multiply and add in the accumulation, and the fused multiply-add that modern inner-product kernels rely on for accuracy.
`ISO/IEC 14882 (C++)`	Programming languages — C++, numeric algorithms	Standardises `std::inner_product` and `std::transform_reduce`, whose default operations do not conjugate, making the convention a caller responsibility.
`ISO/IEC 40314`	Mathematical Markup Language (MathML) Version 3.0	Encodes angle brackets, overlines and summation limits as structured markup so the placement of a conjugate bar is unambiguous to software and 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
IEEE 754 binary64 complex	General numerical work where accuracy matters more than throughput.	About sixteen significant digits per component, at four real multiplies and two adds per term.
IEEE 754 binary32 complex	High-throughput array processing, GPU kernels and large-scale spectral work.	Halves memory traffic, but the roughly seven-digit precision makes cancellation in long correlations a genuine risk.
Extended or compensated accumulation	Very long vectors, or entries spanning many orders of magnitude with sign changes.	Substantially reduces summation error for modest extra cost, but complicates vectorisation and is not bit-reproducible across implementations.
Exact Gaussian rational arithmetic	Verifying orthogonality claims symbolically, or validating a numerical routine against ground truth.	Gives an exact zero for exactly orthogonal vectors, removing all tolerance questions, but is far too slow for production sizes.
Real-only arithmetic on real vectors	The data are genuinely real, as in most static structural and geometric work.	Roughly a quarter of the multiply cost and half the memory, but any later extension to complex data requires reinstating conjugates throughout.
Fixed-point complex (Q-format)	Embedded correlators and beamformers on DSP or FPGA targets.	Deterministic latency and low power, but requires explicit headroom analysis since an inner product accumulates $m$ products before any scaling.

Manufacturing Notes

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

Operation count

A complex inner product of length $m$ costs $m$ complex multiplications and $m-1$ complex additions, which unrolls to roughly $4m$ real multiplications and $4m-2$ real additions with the schoolbook product. It is memory-bandwidth bound for large $m$ , so fusing it into a surrounding loop usually beats calling it as a separate pass.

Hand computation discipline

Conjugate the second vector first, in a separate written column, before multiplying anything. Conjugating on the fly while multiplying is where sign errors originate. Then check the result against conjugate symmetry by recomputing with the arguments swapped — a full independent check for the cost of one extra pass.

Library behaviour and conjugation traps

NumPy's numpy.dot does not conjugate, while numpy.vdot conjugates its first argument, and numpy.inner conjugates neither. MATLAB's dot(u,v) conjugates the first argument. BLAS offers CDOTC (conjugates the first) alongside CDOTU (conjugates neither). Every one of these is a documented, deliberate choice; assume nothing and check the reference page before use.

Verification technique

Four checks cover the definition completely: conjugate symmetry under argument swap; linearity in the first argument with a complex scalar; conjugate-linearity in the second with the same scalar; and $\langle \vec{u}, \vec{u} \rangle$ real and non-negative. Only the second and third distinguish a correct implementation from one with the conjugate on the wrong side, and only when the test scalar is not real.

Numerical stability of the accumulation

The computed inner product satisfies a backward error bound proportional to $m$ times the unit roundoff times the sum of the products of the moduli. When the true result is much smaller than that sum — the near-orthogonal case — almost all significant digits can be lost. Testing orthogonality to high confidence therefore needs either extended precision or a bound scaled by the norms.

Failure Modes & Common Mistakes

Failure modes, root causes and prevention
Failure mode / mistake	Impact	Root cause	Prevention & detection
Conjugating the wrong argument	high	Mixing a text or library that conjugates the first argument with one that conjugates the second.	State the convention at the top of every derivation and module, and verify against a case with genuinely complex entries where the two answers differ.
Omitting the conjugate entirely	high	Extending the real dot product to complex vectors by direct analogy.	Test that $\langle \vec{u}, \vec{u} \rangle$ is real and positive; for $\vec{u} = (1, i)$ the unconjugated form returns zero, exposing the error immediately.
Comparing an inner product with zero for orthogonality in floating point	high	Testing for exact equality with zero on rounded data, so nearly orthogonal vectors are reported as non-orthogonal.	Test $\|\langle \vec{u}, \vec{v} \rangle\|$ against a tolerance scaled by the product of the norms and the vector length.
Assuming the inner product is symmetric	medium	Writing $\langle \vec{u}, \vec{v} \rangle = \langle \vec{v}, \vec{u} \rangle$ from the real case, which is only true up to conjugation.	Use the conjugate-symmetry identity. Note that orthogonality itself is symmetric, which can mask the error in tests that only check for zeros.
Treating $\langle \vec{u}, \vec{u} \rangle$ as complex	low	Carrying a residual imaginary part of size $10^{-17}$ through subsequent real-valued computations, then failing a type or domain check.	Take the real part explicitly after computing a self-inner-product, and assert the discarded imaginary part is within tolerance.
Mismatched vector sizes	medium	Silently broadcasting or truncating when the two arguments have different lengths.	Assert equal sizes before accumulation; an inner product between vectors of different sizes is undefined, not zero.
Catastrophic cancellation in a long correlation	medium	Naive summation of many large terms whose true sum is nearly zero, leaving no correct digits.	Use compensated or pairwise summation, or accumulate in wider precision, and scale the orthogonality tolerance by the sum of the term magnitudes.
Applying an unweighted inner product where a weighted one is required	medium	Testing modal orthogonality without the mass matrix, so genuinely orthogonal modes appear correlated.	Identify the weighting matrix the application defines and use $\vec{v}^{*} M \vec{u}$ consistently throughout.

FAQs

Why is one argument conjugated at all?

So that $\langle \vec{u}, \vec{u} \rangle$ is a sum of squared moduli, hence real and non-negative, and zero only for the zero vector. Without the conjugate the vector $(1, i)$ would have a self-product of $1 + i^2 = 0$ , so any length function built on it would assign zero length to a non-zero vector.

Which argument should carry the conjugate?

It is a convention, not a mathematical fact. This library conjugates the second argument, making the form linear in the first; most physics literature conjugates the first. Everything derivable in one convention is derivable in the other with conjugates relocated, so the only real requirement is consistency within a derivation or codebase.

Is the inner product linear?

In one argument only. It is additive in both and linear in the first under scalar multiplication, but a scalar extracted from the second argument comes out conjugated. That mixture is called sesquilinearity. For real vectors both slots are genuinely linear and the distinction disappears.

How does the inner product relate to the dot product?

It is the same operation whenever the entries are real, because conjugation does nothing to a real number. The complex inner product is a strict extension chosen so that all the geometric consequences of the dot product — length, angle, orthogonality, projection — survive over $\mathbb{C}^m$ .

What does a complex-valued inner product mean physically?

The magnitude measures how strongly the two vectors overlap and the argument gives the relative phase of that overlap. In array processing it is the correlation and phase offset between two channels; in power engineering the real and imaginary parts separate real from reactive power.

Why is

\langle \vec{u}, \vec{u} \rangle

guaranteed to be real?

Conjugate symmetry gives $\langle \vec{u}, \vec{u} \rangle = \overline{\langle \vec{u}, \vec{u} \rangle}$ , and the only complex numbers equal to their own conjugate are the real ones. The explicit computation confirms it directly, since each term is a squared modulus.

How small does an inner product have to be to call two vectors orthogonal numerically?

There is no absolute threshold. Compare the magnitude of the inner product against the product of the two norms, scaled by the vector length and the unit roundoff. A vector pair whose normalised correlation is at the level of $m$ times machine epsilon is orthogonal to the limit of the arithmetic used.

References

Beezer, R. A. A First Course in Linear Algebra, Version 0.70. University of Puget Sound, 2006. Section O, Subsection IP. Licensed under the GNU Free Documentation License v1.2.
ISO 80000-2:2019, Quantities and units — Part 2: Mathematics. International Organization for Standardization.
Higham, N. J. Accuracy and Stability of Numerical Algorithms, 2nd edition. Society for Industrial and Applied Mathematics, 2002.
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, 1979.
Trefethen, L. N. and Bau, D. Numerical Linear Algebra. Society for Industrial and Applied Mathematics, 1997.

AI Suggested Questions

Show a concrete pair of complex vectors for which conjugating the first argument instead of the second gives a different answer, and identify which convention each library uses.
Derive the adjoint identity that the inner product of Au with v equals the inner product of u with the adjoint of A applied to v.
How large can the rounding error in a length-one-million inner product be, and how much does compensated summation help?
Explain why the unconjugated bilinear form fails to define a norm, using a two-entry complex vector as the counterexample.
Derive the Cauchy-Schwarz inequality from the properties of the complex inner product listed here.
How does a mass-weighted inner product change the orthogonality of structural mode shapes compared with the unweighted one?

Related Calculators

Complex Inner Product Calculator

Evaluate the inner product of two complex vectors under either conjugation convention and display the intermediate terms.

Orthogonality Checker

Test a pair or a set of vectors for orthogonality with a norm-scaled numerical tolerance.

Vector Norm Calculator

Compute the norm from the self-inner-product and return the corresponding unit vector.

Overview

Definition

Inner Product

Sesquilinear Form

Concepts

Additivity in each argument

Scalars, and where the conjugate appears

Conjugate symmetry

Positivity of a vector with itself

Matrix form and the adjoint

Relation to the dot product, and to physical quantities

Procedure: evaluate and sanity-check an inner product

Equations

Variable Definitions

Worked Numerical Example

Problem statement

State the two channel vectors

Conjugate the second vector and multiply

Verify conjugate symmetry

Scale the first argument

Scale the second argument

Compute the channel energy

Applications &amp; Industry Use

Matched filtering and beamforming

Complex power from phasor pairs

Modal participation factors

Transition amplitudes and probabilities

Similarity and kernel evaluation

Symbol detection by correlation

Design Considerations

State the conjugation convention explicitly

Never omit the conjugate for convenience

Watch accumulation error for long vectors

Weighted inner products change the geometry

Zero has a precise meaning

Prefer the library kernel over a hand-written loop

Standards &amp; Codes

Material Selection

Manufacturing Notes

Operation count

Hand computation discipline

Library behaviour and conjugation traps

Verification technique

Numerical stability of the accumulation

Failure Modes &amp; Common Mistakes

FAQs

Related Engineering Topics

References

AI Suggested Questions

Related Calculators

Continue learning

Applications & Industry Use

Standards & Codes

Failure Modes & Common Mistakes