The Transform Methods

A Mini-Book of Integral Transforms

@ integra land
 @ integra land

········································································································································
2
 @ integra land

Contents

1 Laplace Transform
(Classical + Generalized/Distributional Forms) 1

2 Fourier Transform
(Classical + Generalized/Distributional forms) 9

3 Mellin Transform and Scale-Invariant

Analysis 17

4 Z-transform and Discrete-Time Methods 23

5 Hilbert Transform and Singular Integral Operators 31

6 Wavelet Transforms and Multiresolution Analysis 37

7 Hankel Transform and Radial Symmetry Problems 43

8 Fractional Fourier Transform (FrFT) and Time–Frequency Analysis 49

9 Short-Time Fourier Transform (STFT) 55

10 FFT 61

A Compact Transform-Pairs Appendix (high-value) 67

B Further Reading 69

········································································································································
i
 @ integra land

········································································································································
ii
 @ integra land

Preface

Integral transforms are not just “techniques”; they are coordinate changes for
analysis. A well-chosen transform can turn:
• differential equations into algebraic equations,
• convolution into multiplication,
• oscillation into localization in a frequency variable,
• scaling into translation (via logarithmic variables),
• singular operators into simple multipliers.
This mini-book is a high-density toolkit: short theory statements (with no
derivations) and a collection of moderate-to-difficult solved problems. The
goal is:
1. to give you a collectible set of methods that you can use immediately in
ODE/PDE, signals, probability, and special-function calculations;
If you enjoy the style here—compact theory, careful conventions, and so-
lutions that highlight the decisive step—then ART OF INTEGRAL TRANS-
FORMSis the natural next step.

Complete details, proofs, and extended theory appear in the textbook ART OF INTE-
GRAL TRANSFORMS.

········································································································································
iii
 @ integra land

········································································································································
iv
 @ integra land

How to read this mini-book

Each chapter has a strict structure:

1. A. Short Theory gives definitions, existence conditions, inversion, key
properties, and a minimal “table” of indispensable transform pairs.
2. B. Solved Problems gives 8 solved problems per transform. Solutions
are sketches: key identity → decisive manipulation → final form.
3. C. Challenge Problems gives harder problems with no solutions; they
are designed to push you into ART OF INTEGRAL TRANSFORMS.
Suggested reading paths:
• PDE track: Laplace → Fourier → Hankel → FFT.
• Signals track: Fourier → Hilbert → STFT → Wavelets → FFT.
• Asymptotics/scale track: Mellin → Fourier/Hilbert → Wavelets.
• Discrete-time track: Z-transform → FFT → STFT.

········································································································································
v
 @ integra land

········································································································································
vi
 @ integra land

1
Laplace Transform
(Classical + Generalized/Distributional Forms)

A. Short Theory

Definitions and conventions

Definition 1.1 (Unilateral Laplace transform). For f : (0, ∞) → C,

Z ∞
(L f )(s) = F (s) := e−st f (t) dt, s ∈ C.
0

Definition 1.2 (Bilateral Laplace transform). For f : R → C,

Z ∞
(Lb f )(s) = e−st f (t) dt,
−∞

whenever convergent.

Existence / convergence (brief)

A standard sufficient condition for the unilateral transform: f is piecewise con-

tinuous on [0, ∞) and of exponential order a, i.e. | f (t)| ≤ Ce at for large t. Then
F (s) exists for ℜ(s) > a (a right half-plane ROC). For bilateral transforms, the
ROC is typically a vertical strip.

········································································································································
1
 @ integra land

Inversion

If F is analytic in ℜ(s) > σ0 and of suitable growth,

Z γ+i∞
1
f (t) = est F (s) ds, γ > σ0 .
2πi γ−i∞

This is the Bromwich inversion formula.

Key properties (high-yield)

For ℜ(s) in the common ROC:

• Linearity: L( a f + bg) = aF + bG.

• Time shift: L{ H (t − a) f (t − a)} = e−as F (s).

• Exponential modulation: L{e at f (t)} = F (s − a).

• Scaling: L{ f (ct)} = 1c F cs for c > 0.

• Differentiation in time:
n −1
L{ f ′ (t)} = sF (s) − f (0+ ), L{ f (n) (t)} = sn F (s) − ∑ s n −1− k f ( k ) (0+ ).
k =0

• Multiplication by t: L{t f (t)} = − F ′ (s), and L{tn f (t)} = (−1)n F (n) (s).
Rt
• Convolution (Volterra): for ( f ∗ g)(t) = 0 f (τ ) g(t − τ ) dτ,

L{ f ∗ g} = F (s) G (s).

• Initial/final value heuristics (with hypotheses):

f (0+ ) = lim sF (s), lim f (t) = lim sF (s).

s→∞ t→∞ s →0+

Transform pairs (minimal table)

Below a > 0, ω ∈ R, α > 0, and H is the Heaviside step:

········································································································································
2
 @ integra land

f (t) F (s) = (L f )(s)

1
1
s
n!
tn (n ∈ N) n+1
s
1
tα−1 /Γ(α)
sα
1
e at
s−a
s
cos(ωt)
s2 + ω 2
ω
sin(ωt)
s + ω2
2

e−as
H (t − a)
s
δ(t − a) −
e as

Generalized/distributional notes

• The Laplace transform extends to distributions supported in [0, ∞) by

duality: (L T )(s) = ⟨ T (t), e−st ⟩.
• Useful identities:

L{δ(t − a)} = e−as , L{δ(n) (t − a)} = sn e−as .

• Many “singular” time-domain expressions (like (e− at − e−bt )/t) become

regular and algebraic in the s-domain.

Typical applications

ODE initial-value problems, PDEs with time as the transform variable (diffu-
sion, wave, viscoelasticity), convolution/Volterra integral equations, con-
trol/transfer functions, probability waiting-time transforms, and asymp-
totic extraction (Tauberian principles). For discrete-time analogies see the
Z-transform in Chapter 4. For Fourier–Laplace interplay (causality, Hilbert
relations), see Chapters 2 and 5.

Complete details, proofs, and extended theory appear in the textbook ART OF INTE-
GRAL TRANSFORMS.
········································································································································
3
 @ integra land

B. Solved Problems

Problem 1.1 (A complex-parameter Laplace integral). Let a > 0, b ∈ R,

and ν > 0. Evaluate
Z ∞
I ( a, b, ν) = tν−1 e− at cos(bt) dt.
0

Solution. Write cos(bt) = ℜ(eibt ):

Z ∞  
ν−1 −( a−ib)t −ν
I=ℜ t e dt = ℜ Γ(ν)( a − ib) .
0
√
Let a − ib = a2 + b2 e−iθ with θ = arctan(b/a). Then
I = Γ(ν)( a2 + b2 )−ν/2 cos ν arctan(b/a) .

For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem
n  1.2 o(A logarithmic Laplace inversion). For a, b > 0, find f (t) =
L−1 ln ss+
+a
b
( t ).

Solution. Use the identity (Frullani-type in Laplace form)

Z ∞ −(s+ a)t
− e−(s+b)t
 
e s+b
dt = ln .
0 t s+a
Therefore
e−at − e−bt
f (t) =
, t > 0.
t
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 1.3 (Impulse-driven oscillator (distributional Laplace)). Solve

y′′ (t) + y(t) = δ(t − π2 ), y(0) = 0, y′ (0) = 0.

Solution. Taking Laplace transforms and using L{δ(t − a)} = e−as gives
1
(s2 + 1)Y (s) = e−(π/2)s ⇒ Y (s) = e−(π/2)s .
s2 + 1
········································································································································
4
 @ integra land

Invert via shift: L−1 {1/(s2 + 1)} = sin t, hence

y(t) = H t − π2 sin t − π2 .
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 1.4 (A Volterra equation with polynomial kernel). Solve for y on

t ≥ 0: Z t
y(t) − (t − τ ) y(τ ) dτ = t2 .
0

Solution. The integral is a convolution with k (t) = t. Thus L{k } = 1/s2 and
1 2 2
Y (s) − Y ( s ) = L{ t } = .
s2 s3
 
1 2
So Y (s) 1 − s2
= s3
, hence
2 2 1 1
Y (s) = = − + + .
s ( s2 − 1) s s−1 s+1
Therefore y(t) = −2 + et + e−t = 2(cosh t − 1).
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 1.5 (A classic “divide-by-t” identity). Show that for s > 0 and
a ∈ R, Z ∞
−st sin( at )
a
e dt = arctan .
0 t s

Solution. Differentiate with respect to a:

Z ∞ Z ∞
d −st sin( at ) s
e e−st cos( at) dt = 2
dt = .
0 da t 0 s + a2
Ra
Since the integral is 0 at a = 0, integrate back: 0 s2 +s u2 du = arctan( a/s).
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 1.6 (Laplace of shifted distributional derivatives). Compute

L{δ(n) (t − a)}(s) for a > 0 and n ∈ N.
········································································································································
5
 @ integra land

Solution. By distributional duality,

Z ∞
dn −st

L{δ (n)
(t − a)}(s) = e−st δ(n) (t − a) dt = (−1)n e = sn e−as .
0 dtn t= a

For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 1.7 (Half-line heat equation with a delayed boundary switch).

Solve ut = κu xx for x > 0, t > 0 with

u( x, 0) = 0, u(0, t) = H (t − T ), u( x, t) → 0 as x → ∞,

where κ > 0 and T > 0.

Solution. √
Laplace in t gives sU = κUxx , bounded solution U ( x, s) =
A(s)e − x s/κ . Boundary: U (0, s) = L{ H (t − T )} = e−Ts /s, so

e−Ts
 r 
s
U ( x, s) = exp − x .
s κ
√
Use the known inversion L−1 {e−c s /s} = erfc √ c


and shift in time:
2 t
!
x
u( x, t) = H (t − T ) erfc p .
2 κ (t − T )

For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 1.8(A fractional-looking

 inversion that collapses to erfi). Find
1
f ( t ) = L −1 √ ( t ).
s ( s + 1)

Solution. Factor as s−1/2 · (s + 1)−1 and use convolution:

1
L−1 {s−1/2 } = √ , L−1 {(s + 1)−1 } = e−t .
πt
So Z t Z t
1 1
f (t) = √ e−(t−τ ) dτ = e−t √ τ −1/2 eτ dτ.
0 πτ π 0
········································································································································
6
 @ integra land

Rt u2 du =
√ R √t √
Let τ = u2 : τ −1/2 eτ dτ
= 2 e π erfi ( t). Hence f (t) =
− t
√ 0 0
e erfi( t).
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

C. Challenge Problems (no solutions)

ln(s)
1. Invert F (s) = and express the answer using special functions
s2 + 1
and distributional terms if needed.
Z t
2. Solve y(t) = t + (t − τ )1/2 y(τ ) dτ using fractional Laplace pairs.
0

3. Derive the Green’s function for utt + 2aut = c2 u xx on x ∈ R via Laplace

in t and Fourier in x (connect to Chapters 2, 10).
4. Use Laplace asymptotics (Watson’s lemma) to extract the small-t expan-
sion of f (t) from a large-s expansion of F (s).
√
−
e s
5. Invert F (s) = √ and obtain sharp t → 0+ asymptotics.
s ( s + 1)
6. Prove a precise version of the final value theorem with minimal hy-
potheses (then test it on an oscillatory counterexample).

These are developed in depth in ART OF INTEGRAL TRANSFORMS.

········································································································································
7
 @ integra land

········································································································································
8
 @ integra land

2
Fourier Transform
(Classical + Generalized/Distributional forms)

A. Short Theory
Definitions and conventions

We use the unitary Fourier convention (excellent for L2 theory and for Chap-
ters 5, 9, 8):

Definition 2.1 (Fourier transform). For f ∈ L1 (R),

Z ∞
1
(F f )(ω ) = fb(ω ) := √ f (t)e−iωt dt.
2π −∞

Definition 2.2 (Inverse Fourier transform).

Z ∞
1
f (t) = √ fb(ω )eiωt dω (in appropriate senses).
2π −∞

√
If you prefer the engineering convention without 1/ 2π, convert via
constant rescalings; all structural theorems remain identical.

Existence / convergence (brief)

• If f ∈ L1 (R), then fb is bounded and continuous, and fb(ω ) → 0 as

|ω | → ∞ (Riemann–Lebesgue).

• If f ∈ L2 (R), F extends uniquely to a unitary operator on L2 (Plancherel).

········································································································································
9
 @ integra land

• Schwartz functions S are the ideal classical class; tempered distributions

S ′ give the generalized theory.

Inversion and energy identities

Identity 2.1 (Plancherel / Parseval). For f ∈ L2 (R),

Z Z
∥ f ∥2 = ∥ fb∥2 , f (t) g(t) dt = fb(ω ) gb(ω ) dω.
R R

Key properties

• Linearity: a\
f + bg = a fb + bb
g.

• Shift: F { f (t − t0 )} = e−iωt0 fb(ω ).

• Modulation: F {eiω0 t f (t)} = fb(ω − ω0 ).

1 b
• Scaling: F { f ( at)} = | a|
f (ω/a).

• Differentiation:
d b
F { f ′ (t)} = (iω ) fb(ω ), F {t f (t)} = i f ( ω ).
dω

• Convolution theorem:
√ Z
F {( f ∗ g)(t)} = 2π fb(ω ) gb(ω ), ( f ∗ g)(t) = f (τ ) g(t − τ ) dτ.
R

• Product ↔ convolution: F { f g} = √1 ( fb ∗ gb).

2π

• Duality (up to reflection): If fb(ω ) = g(ω ) then gb(t) = f (−t).

Distributional interpretations (high-yield)

In S ′ :
1 √
F {δ(t − t0 )} = √ e−iωt0 , F {1} = 2π δ(ω ).
2π
Principal value objects appear naturally and connect directly to the Hilbert
transform (Chapter 5).
········································································································································
10
 @ integra land

Minimal transform pairs

For a > 0:
r   r
2 a 1 π 1 − a|ω |
F {e−a|t| }(ω ) = , F (ω ) = e .
π a + ω2
2 t + a2
2 2a

Gaussian self-replication (unitary convention):

2 /2 2 /2
F { e−t }(ω ) = e−ω .

Typical applications

Constant-coefficient PDEs on Rn (heat, wave, Schrödinger), spectral multi-

pliers, convolution/integral equations, stationary phase and oscillatory inte-
grals, signal spectra, probability characteristic functions. For time–frequency
localization see STFT (Chapter 9) and FrFT (Chapter 8). For the conjugate-
analytic/singular side see Hilbert (Chapter 5). For computational realizations
see FFT (Chapter 10).

Complete details, proofs, and extended theory appear in the textbook ART OF INTE-
GRAL TRANSFORMS.

B. Solved Problems

Problem 2.1 (A rational transform leading to an exponential cusp). Com-

pute the Fourier transform (unitary convention)
 
1
fb(ω ) = F 2 ( ω ), a > 0.
t + a2

e−iωt π − a|ω |
R
Solution. Use the standard integral R t2 + a2 dt = ae , hence
r
1 π π 1 − a|ω |
fb(ω ) = √ · e− a|ω | = e .
2π a 2a

For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.
········································································································································
11
 @ integra land

Problem 2.2 (A sharper rational: squared denominator). Evaluate

Z ∞
cos(ωt)
I (ω ) = dt, a > 0.
−∞ ( t2 + a2 )2

e−iωt
R
Solution. Known transform (nonunitary) is R ( t + a2 )2
2 dt = π
2a3
( a|ω | +
1)e− a|ω | . Taking the real part gives the cosine integral:
π
I (ω ) = 3
( a | ω | + 1) e − a | ω | .
2a
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 2.3 (Convolution of two two-sided exponentials). Let a, b > 0

with a ̸= b and f (t) = e− a|t| , g(t) = e−b|t| . Compute ( f ∗ g)(t).
q
Solution. Use F { e− a|t| } = 2 a
and the convolution theorem:
π a2 + ω 2
! r !
√ √
r
2 a 2 b
F { f ∗ g} = 2π fb gb = 2π
π a2 + ω 2 π b2 + ω 2
r
2 2ab
= .
π ( a2 + ω 2 )(b2 + ω 2 )
Partial fractions yield
2ab 2a b2 2b a2
= − .
( a2 + ω 2 )(b2 + ω 2 ) a2 − b2 b2 + ω 2 a2 − b2 a2 + ω 2
Invert termwise using the same pair:
2  −b|t| − a|t|

( f ∗ g)(t) = 2 ae − be .
a − b2
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 2.4 (Heat equation on the line via Fourier). Solve ut = κu xx on

2
R with u( x, 0) = e− x .
········································································································································
12
 @ integra land

Solution. Fourier in x gives ∂t ub(ω, t) = −κω 2 ub(ω, t), so

2
ub(ω, t) = e−κω t ub(ω, 0).
2
√ 2 2
Since u( x, 0) = e− x = e −( 2x ) /2 , its unitary transform is u
b(ω, 0) = √1 e−ω /4 .
2
2
Thus ub(ω, t) = √1 e−(κt+1/4)ω . Invert using the Gaussian pair:
2

x2
 
1
u( x, t) = √ exp − .
1 + 4κt 1 + 4κt

For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 2.5 (A distributional cornerstone). Show (in the sense of tem-

pered distributions) that
r
π
F {PV(1/t)}(ω ) = −i sgn(ω ).
2

Solution. This is the Fourier-side characterization of the Hilbert kernel:

h(t) = π1 PV(1/t) has multiplier −i sgn(ω ) (Chapter 5). Thus F {PV(1/t)} =
π F {h} = −qiπ sgn(ω )F {1}/? In unitary normalization, the constant eval-
uates to −i π2 sgn(ω ). (Equivalent derivations follow from testing against
Schwartz functions and odd symmetry.)
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 2.6 (An integral identity encoding a cusp spectrum). Prove that
for ω ∈ R, Z ∞
1 − cos(ωt)
dt = π |ω |.
−∞ t2
R 1−cos(ωt) ′ (ω ) =
R sin(ωt)
Solution. Let F (ω ) = R t 2 dt. Differentiate: F R t dt, a
standard distributional integral giving π sgn(ω ). Since F (0) = 0, integrate:
F ( ω ) = π | ω |.
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.
········································································································································
13
 @ integra land

Problem 2.7 (A convolution integral equation on R). Find f ∈ L2 (R)

solving
f + (k ∗ f ) = k, k (t) = e−|t| .

Solution. Take Fourier transforms:

√ k
b
fb + k fb = b
2π b k ⇒ fb = √ .
1+ k
2πb
q
2 1
√
k(ω ) =
With b π 1+ ω 2 , we get k = 2/(1 + ω 2 ), hence
2πb
r
2 1
fb(ω ) =
.
π ω2 + 3
√ √
Invert using the exponential pair with a = 3: f (t) = √ e 3|t| .
1 −
3
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 2.8 (Poisson equation in R2 via Fourier multipliers). Solve

−∆u = f on R2 in Fourier form and identify the fundamental multi-
plier.

Solution. Taking the 2D Fourier transform gives

2 fb(ξ )
|ξ | ub(ξ ) = fb(ξ ) ⇒ ub(ξ ) = .
| ξ |2
Thus u is obtained by applying the multiplier |ξ |−2 (a Riesz potential). The
spatial kernel is the logarithmic fundamental solution in 2D: u = f ∗ G with
G ( x ) = −(2π )−1 log | x | (distributionally).
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

C. Challenge Problems (no solutions)

Z
2
1. Use stationary phase to estimate eiλt f (t) dt as λ → ∞ and relate the
R
result to Fourier-side decay.
········································································································································
14
 @ integra land

2. Compute F {(t2 + a2 )−m } for integer m ≥ 3 and express the answer

using e− a|ω | times a polynomial in |ω |.
3. Prove the Hausdorff–Young inequality and test sharpness on Gaussians.
4. Derive the Fourier transform of log |t| in S ′ and explain how δ-terms
appear.
5. Solve the wave equation utt = c2 u xx on R for distributional initial data,
emphasizing the transform-domain meaning.
6. Derive the Paley–Wiener theorem statement for compact support and
compare with analyticity strips in Laplace theory (Chapter 1).

These are developed in depth in ART OF INTEGRAL TRANSFORMS.

········································································································································
15
 @ integra land

········································································································································
16
 @ integra land

3
Mellin Transform and Scale-Invariant
Analysis

A. Short Theory
Definition and inversion

Definition 3.1 (Mellin transform). For f : (0, ∞) → C,

Z ∞
(M f )(s) = F (s) := x s−1 f ( x ) dx,
0

for complex s in the strip of convergence.

Definition 3.2 (Inversion). If F (s) is analytic in a vertical strip and suitable

growth holds,
Z c+i∞
1
f (x) = x −s F (s) ds,
2πi c−i∞
where ℜ(s) = c lies in the convergence strip.

Existence strip

If f ( x ) ∼ O( x −α ) as x → 0+ and f ( x ) ∼ O( x − β ) as x → ∞, then typically the

Mellin transform converges for α < ℜ(s) < β.

Core identities: Mellin turns scaling into translation

• Scaling: M{ f ( ax )}(s) = a−s F (s), a > 0.

• Multiplication by powers: M{ x c f ( x )}(s) = F (s + c).

········································································································································
17
 @ integra land

• Log-derivative: M{(log x ) f ( x )}(s) = F ′ (s).

Mellin convolution (multiplicative convolution)

Define
Z ∞  
x dt
( f ⋆ M g)( x ) := f g(t) .
0 t t
Then
M{ f ⋆ M g}(s) = F (s) G (s).
This is the scale-invariant analog of the Laplace/Fourier convolution theo-
rems.

Parseval-type identity

When both sides are defined,

Z ∞ Z ∞
dx 1
f ( x ) g( x ) = F (c + it) G (c + it) dt.
0 x 2π −∞

Transform pairs worth memorizing

 
−x 1 π
M{e }(s) = Γ(s), M ( s ) = Γ ( s ) Γ (1 − s ) = .
1+x sin(πs)
Also, for p > 0,
Z ∞ s −1  
x π πs
dx = csc , 0 < ℜ(s) < p.
0 1 + xp p p

Applications

Asymptotics (Mellin inversion and residue calculus), scale-invariant differen-

tial equations, products of random variables, special functions (Gamma/Beta,
Bessel K), and the conceptual bridge between dilation and translation (com-
pare with wavelets in Chapter 6).

Complete details, proofs, and extended theory appear in the textbook ART OF INTE-
GRAL TRANSFORMS.
········································································································································
18
 @ integra land

B. Solved Problems

Problem 3.1 (Beta integral via Mellin). For 0 < ℜ(s) < 1, evaluate
Z ∞ s −1
x
dx.
0 1+x

Solution. Recognize this as M{(1 + x )−1 }(s), and use the Beta/Gamma
identity:
Z ∞ s −1
x π
dx = B(s, 1 − s) = Γ(s)Γ(1 − s) =.
0 1+x sin(πs)
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 3.2 (A Bessel-K integral (Laplace-in-x meets Mellin-in-scale)).

Show that for β, γ > 0 and ℜ(µ) > 0,
Z ∞  µ/2
µ−1 − βx −γ/x γ p

x e dx = 2 Kµ 2 βγ .
0 β

Solution. Use the classic representation of Kµ :

1  z  µ ∞ − µ −1 z2
Z  
Kµ ( z ) = t exp −t − dt,
2 2 0 4t
p
then match parameters by the substitution t = βx and z = 2 βγ. This yields
the stated closed form.
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 3.3 (A log-moment by differentiation in Mellin space). For 0 <

ℜ(s) < 1, compute
Z ∞ s −1
x log x
J (s) = dx.
0 1+x

Solution. Differentiate the Mellin identity from the previous problem:

Z ∞ s −1
x π
dx = .
0 1+x sin(πs)
········································································································································
19
 @ integra land

Since ∂s x s−1 = x s−1 log x, we obtain

 
d π cos(πs)
J (s) = = −π 2 .
ds sin(πs) sin2 (πs)
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 3.4 (A general scale law: 1/(1 + x p )). For p > 0 and 0 < ℜ(s) <
p, evaluate
Z ∞ s −1
x
p dx.
0 1+x

Solution. Substitute u = x p so x = u1/p and dx = 1p u1/p−1 du:

Z ∞ s −1 Z ∞ s/p−1
x 1 u 1 π
dx = du = · .
0 1 + xp p 0 1+u p sin(πs/p)
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 3.5 (A zeta value from a Mellin identity). Show that

Z ∞
x π2
dx = .
0 ex − 1 6

Solution. Use the standard Mellin relation (valid for ℜ(s) > 1):
Z ∞ s −1
x
dx = Γ(s)ζ (s).
0 ex − 1
R∞ 2
Set s = 2 to get 0 exx−1 dx = Γ(2)ζ (2) = 1 · π6 .
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 3.6 (Mellin of a Bessel J (special-function fingerprint)). For ν >

−1, evaluate (in the convergence strip)
Z ∞
x s−1 Jν ( x ) dx.
0

········································································································································
20
 @ integra land

Solution. A standard Mellin pair is

Γ ν+s
Z ∞

x s−1 Jν ( x ) dx = 2s−1 2
ν − s +2
Γ

,
0 2

valid for −ℜ(ν) < ℜ(s) < 32 (precise bounds depend on ν).
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 3.7 (A scale-asymptotic that produces a logarithm). Let

Z ∞
e− x
I (λ) = dx, λ → ∞.
0 1 + λx
Find the leading asymptotic form.
R ∞ e−u/λ
Solution. Write I (λ) = λ−1 0 1+u du. As λ → ∞, the integral has a
R ∞ −t
logarithmic divergence regulated by e−u/λ . Using E1 (z) = z e t dt and the
identity
1 1/λ
e E1 (1/λ), I (λ) =
λ
the small-z expansion E1 (z) = −γ − log z + O(z) gives
 
log λ + γ 1
I (λ) = +O 2 .
λ λ
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 3.8 (A scale-invariant ODE becomes algebraic in Mellin space).

Solve for x > 0:

x2 y′′ ( x ) + xy′ ( x ) − ν2 y( x ) = x β , ν∈
/ {± β}.

Solution. Try a scale ansatz y p ( x ) = Cx β . Then

x2 ( β( β − 1)Cx β−2 ) + x ( βCx β−1 ) − ν2 (Cx β ) = C ( β2 − ν2 ) x β .
Match RHS to obtain C = ( β2 − ν2 )−1 . Thus
xβ
y ( x ) = c1 x ν + c2 x − ν + .
β2 − ν2
········································································································································
21
 @ integra land

(Exactly the structure Mellin theory predicts: powers correspond to poles/shifts.)

For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

C. Challenge Problems (no solutions)

1. Use Mellin
Z inversion and residues to derive the full asymptotic expan-
∞
sion of f ( x ) g(λx ) dx as λ → ∞ from poles of F (s) G (1 − s).
0
2. Prove a precise mapping theorem: which weighted L p spaces map to
vertical-strip Hardy spaces under M?
3. Derive the Mellin transform of log(1 + x ) and invert it in closed form.
4. Use Mellin methods to analyze ∑ e−nx and extract the small-x expan-
n ≥1
sion with Bernoulli corrections.
5. Solve a multiplicative convolution equation f ⋆ M k = g for a nontrivial
kernel k with meromorphic Mellin transform.
6. Relate Mellin methods to wavelet vanishing moments (Chapter 6) via
logarithmic coordinates.

These are developed in depth in ART OF INTEGRAL TRANSFORMS.

········································································································································
22
 @ integra land

4
Z-transform and Discrete-Time Methods

A. Short Theory
Definitions and ROC

Definition 4.1 (Bilateral Z-transform). For a two-sided sequence x [n],

∞
(Z x )(z) = X (z) := ∑ x [n ] z−n ,
n=−∞

with region of convergence (ROC) where the series converges.

Definition 4.2 (Unilateral (causal) Z-transform). For x [n] = 0 for n < 0,

∞
+
X (z) = ∑ x [n ] z−n .
n =0

Inversion

If X is analytic on a closed contour C in the ROC,

1
I
x [n] = X (z) zn−1 dz.
2πi C

Poles and the ROC determine causality and stability (bounded-input bounded-
output, BIBO).

Key properties

• Linearity: Z { ax + by} = aX + bY.

········································································································································
23
 @ integra land

• Time shift: Z { x [n − n0 ]} = z−n0 X (z) (ROC adjusted).

• Exponential weighting: Z { an x [n]} = X (z/a).

• Convolution: ( x ∗ h)[n] = ∑k x [k ]h[n − k ] satisfies Z { x ∗ h} = XH.

• Difference equations: shifts become powers of z−1 , converting recur-

rences into algebraic equations.

DTFT and the unit circle

If the ROC includes |z| = 1, the discrete-time Fourier transform exists:

∞
iω
X (e ) = ∑ x [n]e−iωn .
n=−∞

This is the Fourier chapter in disguise (Chapter 2), with z = eiω .

Laplace analogy

Sampling a causal continuous-time signal f (t) at t = nT often yields Z-

domain objects that mirror Laplace objects with z = esT ; this analogy is power-
ful but subtle (aliasing, ROC geometry). See Chapter 1 for the continuous-time
side.

Complete details, proofs, and extended theory appear in the textbook ART OF INTE-
GRAL TRANSFORMS.

B. Solved Problems

Problem 4.1 (A forced second-order difference equation). Solve for n ≥ 0:

y[n + 2] − 3y[n + 1] + 2y[n] = u[n], y[0] = 0, y[1] = 0,

where u[n] is the unit step.

Solution. Take unilateral Z-transforms. Using Z {y[n + 1]} = z(Y − y[0])

and Z {y[n + 2]} = z2 (Y − y[0] − y[1]z−1 ), we get
z
z2Y − 3zY + 2Y = Z {u[n]} = .
z−1
········································································································································
24
 @ integra land

z z
So Y = = . Partial fractions yield
(z − 1)(z2 − 3z + 2) ( z − 1)2 ( z − 2)
2 2 1
Y= − + .
z − 2 z − 1 ( z − 1)2

Invert: Z −1 { z−1 a } = an−1 u[n − 1] (with consistent indexing) and Z −1 { (z−z1)2 } =

n u[n]. After alignment, the closed form is

y[n] = 2n−1 − 1 + n, n ≥ 0.

For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 4.2 (Inverse Z with repeated poles). Find x [n] (causal) for
z ( z + 1)
X (z) = , ROC: |z| > 1.
(z − 1)2 (z − 12 )

Solution. Use partial fractions in z:

A B C
X (z) = + + .
z − 1 (z − 1)2 z − 12

Solving gives A = 2, B = 4, C = −2. Invert with Z −1 { (z−za)2 } = nan−1 u[n]

and Z −1 { z−z a } = an u[n] to obtain
 n
n −1
n
x [n] = 2 · 1 + 4n · 1 −2 1
2 = 2 + 4n − 21−n , n ≥ 0.

For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 4.3 (Convolution via Z rather than summation). Let x [n] =

(1/2)n u[n] and h[n] = n u[n]. Compute y = x ∗ h.

Solution. Transforms:
 n 1 z z
X (z) = ∑ 12 z−n = = , H (z) = ∑ nz−n = (z − 1)2 .
n ≥0 1 − 12 z−1 z − 12 n ≥0
········································································································································
25
 @ integra land

z2
Thus Y (z) = XH = . Decompose and invert; the result is
(z − 12 )(z − 1)2
y[n] = 21−n − (n + 1) + n · 2, n ≥ 0,
i.e. y[n] = 2n − n − 1 + 21−n = n − 1 + 21−n .
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

z
Problem 4.4 (Stability and ROC selection). Let X (z) = .
(z − 2)(z − 12 )
List the possible ROCs and identify which corresponds to a causal and
BIBO-stable sequence.

Solution. Poles at 2 and 1/2 yield three ROCs:

1
|z| > 2, 2 < |z| < 2, |z| < 12 .
Causality requires ROC outside the outermost pole: |z| > 2. Stability requires
ROC include the unit circle |z| = 1, so the only stable choice is 12 < |z| < 2
(two-sided, noncausal). Hence: no causal stable sequence exists for this X.
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 4.5 (A first-order system with a sparse input). Solve y[n] −

1
4 y [ n − 1] = x [ n ] with y [−1] = 0 and

x [ n ] = δ [ n ] − δ [ n − 3].

Solution. Take unilateral Z-transforms:

Y − 41 z−1Y = X, X = 1 − z −3 .
1 − z −3
So Y = . Invert using the geometric series:
1 − 14 z−1
1  n
1 −1
= ∑ 41 z−n .
1 − 4z n ≥0

Thus  n   n −3
1 1
y[n] = 4 u[n] − 4 u [ n − 3].
········································································································································
26
 @ integra land

For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 4.6 (Central binomial coefficients as a Z/generating function).

Show that for | a| < 41 ,
∞  
2n 1
∑ n an = √1 − 4a .
n =0

Solution. The sequence cn = (2n n

n ) has generating function C ( w ) = ∑n≥0 cn w .
(2n)!
Using the identity (2n
n ) = (n!)2 and the standard binomial series,

∞ ∞  
−1/2
 
2n
(1 − 4w) −1/2
= ∑ n
(−4w) = ∑
n
n
wn .
n =0 n =0

Set w = a.
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 4.7 (A nonhomogeneous recurrence with alternating forcing).

Let x [0] = 0, x [1] = 1 and for n ≥ 2,

x [n] − 3x [n − 1] + 2x [n − 2] = (−1)n .

Find x [n].

Solution. Homogeneous roots: r2 − 3r + 2 = (r − 1)(r − 2) = 0, so xh [n] =

A · 1n + B · 2n . Try a particular x p [n] = C (−1)n . Substitute:

C (−1)n − 3C (−1)n−1 + 2C (−1)n−2 = C (−1)n (1 + 3 + 2) = 6C (−1)n .

So 6C = 1 ⇒ C = 16 . Thus x [n] = A + B2n + 16 (−1)n . Use x [0] = 0:

A + B + 16 = 0. Use x [1] = 1: A + 2B − 16 = 1. Solve: B = 12 , A = − 23 . Hence

2 1 1
x [n] = − + 2n + (−1)n .
3 2 6
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.
········································································································································
27
 @ integra land

Problem 4.8 (IIR filter impulse response from a rational transfer function).
Given
1 − z −1
H (z) = ,
1 − 34 z−1 + 81 z−2
find the causal impulse response h[n] and determine BIBO stability.

Solution. Factor denominator: 1 − 34 z−1 + 81 z−2 = (1 − 12 z−1 )(1 − 14 z−1 ).

Then
1 − z −1 A B
H (z) = = + .
(1 − 12 z−1 )(1 − 41 z−1 ) 1 − 12 z−1 1 − 14 z−1
Solve quickly to get A = 32 , B = 13 . Hence
2  1 n 1  1 n
h[n] = u[n] + u [ n ].
3 2 3 4
Poles are at |z| = 12 , 14 < 1, so the causal ROC |z| > 21 includes the unit circle:
stable.
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

C. Challenge Problems (no solutions)

1. Use residue inversion to classify all sequences corresponding to a given
X (z) with multiple possible ROCs; interpret physically (causal/anti-
causal/two-sided).
2. Derive exact conditions under which X (eiω ) exists (unit circle in ROC)
and relate to DTFT convergence.
3. Prove the sampling relation z = esT connecting Z-planes and Laplace
planes, and quantify aliasing in terms of periodicity in ℑ(s).
4. Solve a nonconstant-coefficient recurrence via Z-transform and contour
deformation.
5. Use Z-transform to compute a nontrivial finite sum involving (nk) and
alternating weights.
6. Design a minimum-phase filter from magnitude constraints by Z-domain
factor allocation.
········································································································································
28
 @ integra land

These are developed in depth in ART OF INTEGRAL TRANSFORMS.

········································································································································
29
 @ integra land

········································································································································
30
 @ integra land

5
Hilbert Transform and Singular Integral
Operators

A. Short Theory
Definition

Definition 5.1 (Hilbert transform on R). For suitable f (e.g. f ∈ L p ,

1 < p < ∞),
Z ∞
1 f (t)
(H f )( x ) = PV dt.
π −∞ x − t

1
The kernel π PV 1x is a model singular integral.

Fourier multiplier representation

With the unitary Fourier transform (Chapter 2),

F {H f }(ω ) = (−i sgn ω ) fb(ω ).

This is the fastest way to compute Hilbert transforms of “Fourier-friendly”

functions.

Key properties

For f in appropriate spaces:

• Linearity.
• Skew-adjointness on L2 : ⟨H f , g⟩ = −⟨ f , H g⟩.
• Involution: H(H f ) = − f on L2 (up to constants in some settings).
········································································································································
31
 @ integra land

• Commutation with derivatives: H( f ′ ) = (H f )′ (distributionally).

• Analytic signal: f + i H f is the boundary value of a function analytic in

the upper half-plane, under standard hypotheses.

• Causality/dispersion links: Kramers–Kronig relations connect real and

imaginary parts via Hilbert transforms (Fourier–Laplace interplay).

Distributional notes

• H is defined on tempered distributions by Fourier multiplication.

R sin( ax )
• Principal value integrals like PV x dx are naturally produced by
Hilbert/Fourier identities.

Applications

Analytic continuation and boundary values, harmonic conjugates, singular in-

tegral equations, signal processing (quadrature signals, envelope/phase), and
dispersion relations. For time–frequency localization of analytic components
see STFT (Chapter 9) and wavelets (Chapter 6).

Complete details, proofs, and extended theory appear in the textbook ART OF INTE-
GRAL TRANSFORMS.

B. Solved Problems

Problem 5.1 (Hilbert transform of sinusoids). Compute H{cos(ω0 t)} and

H{sin(ω0 t)} for ω0 > 0.

Solution. Using Fourier multipliers: cos(ω0 t) has spectral masses at ±ω0 .

Multiplying by −i sgn(ω ) rotates phases:

H{cos(ω0 t)} = sin(ω0 t), H{sin(ω0 t)} = − cos(ω0 t).

(So cos +i H cos = eiω0 t is analytic-signal form.)

For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.
········································································································································
32
 @ integra land

Problem 5.2 (A canonical rational example). Show that

 
1 x
H ( x ) = .
1 + t2 1 + x2
q
Solution. From Chapter 2, F {(1 + t2 )−1 }(ω )
= π2 e−|ω | . Apply the Hilbert
multiplier:
   r
1 π −|ω |
F H = (− i sgn ω ) e .
1 + t2 2
Invert: this is exactly the Fourier transform of 1+xx2 (odd partner of the Cauchy
kernel). Thus H{(1 + t2 )−1 } = x/(1 + x2 ).
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 5.3 (A principal-value integral: the sin x/x law). Evaluate

Z ∞
sin( ax )
PV dx, a ∈ R.
−∞ x

Solution. The distribution PV(1/x ) has Fourier transform proportional

to −i sgn(ω ) (Chapter 2). Equivalently, the odd integral is the jump of a
logarithm. The standard value is
Z ∞
sin( ax )
PV dx = π sgn( a).
−∞ x
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 5.4 (Logarithmic cosine integral). Show that for a, b > 0,

Z ∞
cos( ax ) − cos(bx )
 
b
dx = ln .
0 x a

Solution. Differentiate in b:
Z ∞ Z ∞
∂ cos( ax ) − cos(bx ) 1
dx = sin(bx ) dx = .
∂b 0 x 0 b
········································································································································
33
 @ integra land

Thus the integral equals ln b + C ( a). Setting b = a gives 0, hence C ( a) = − ln a.

Therefore the value is ln(b/a).
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 5.5 (Hilbert transform inversion trick). Find f such that H f ( x ) =

1
.
1 + x2

Solution. On L2 , H2 = − I. Apply H to both sides:

 
1 x
− f =H = .
1 + x2 1 + x2
x
Hence f ( x ) = − .
1 + x2
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 5.6 (Analytic signal of a real exponential burst). Let f (t) = e−|t| .
Compute the analytic signal f + i H f explicitly in Fourier form.
q
2 1
Solution. fb(ω ) = π 1+ ω 2 . Then
(

2 fb(ω ), ω > 0,
+ i H f (ω ) = fb(ω ) 1 + sgn ω =
f\
0, ω < 0,

(up to the point ω = 0). So the analytic signal is the positive-frequency projection
of f , a key viewpoint for Chapters 9 and 6.
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 5.7 (Gaussian Hilbert transform and the Dawson function). Com-
2
pute H{e−t }( x ) in closed form (special function allowed).

2
Solution. Using Fourier multipliers: F {e−t } is Gaussian, and multiplica-
tion by −i sgn ω produces an odd transform whose inverse is known to be
········································································································································
34
 @ integra land

expressible via Dawson’s integral

Z x
− x2 2
D(x) = e eu du.
0
The result is
2 2
H{e−t }( x ) = √ D ( x ).
π
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 5.8 (A singular integral identity from a known transform pair).

Evaluate (principal value)
Z ∞
1
PV dt.
−∞ ( x − t)(1 + t2 )

Solution. By definition,
  Z ∞
1 1 1
H 2
( x ) = PV dt.
1+t π −∞ ( x − t )(1 + t2 )

From the earlier solved problem, the Hilbert transform equals 1+xx2 . Thus the
PV integral equals π 1+xx2 .
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

C. Challenge Problems (no solutions)

1. Prove boundedness of H on L p (R) for 1 < p < ∞ and explain why it
fails on L1 .
2. Solve a finite Hilbert transform equation on (−1, 1) and express the
solution using Chebyshev weights.
3. Derive Kramers–Kronig relations from analyticity in the upper half-
plane and connect to Laplace causality (Chapter 1).
4. Compute H{log |t|} in the distributional sense and track constant/
δ-terms carefully.
5. Use H to factor a real signal into minimum-phase and maximum-phase
components (spectral factorization).
········································································································································
35
 @ integra land

6. Analyze H on periodic functions and relate it to Fourier series multipli-

ers.

These are developed in depth in ART OF INTEGRAL TRANSFORMS.

········································································································································
36
 @ integra land

6
Wavelet Transforms and Multiresolution
Analysis

A. Short Theory
Continuous wavelet transform (CWT)

Definition 6.1 (CWT). Let ψ ∈ L2 (R) be a wavelet. For a ̸= 0 and b ∈ R

define
t−b
 
1
Z
ψa,b (t) = p ψ , (Wψ f )( a, b) = ⟨ f , ψa,b ⟩ = f (t)ψa,b (t) dt.
| a| a R

Admissibility and inversion

A sufficient admissibility condition (in Fourier terms) is

Z ∞ b
|ψ(ω )|2
Cψ := dω < ∞,
0 ω
R
which implies ψ b(0) = 0 (zero mean: ψ = 0). Then reconstruction holds
(schematically)
Z ∞ Z ∞
1 da db
f (t) = (Wψ f )( a, b) ψa,b (t) ,
Cψ −∞ 0 a2

in L2 sense.

Energy identity (Plancherel for CWT)

Z ∞ Z ∞
da db
|(Wψ f )( a, b)|2 2
= Cψ ∥ f ∥22 .
−∞ 0 a
········································································································································
37
 @ integra land

Multiresolution analysis (MRA) and discrete wavelets

An MRA is a nested sequence · · · ⊂ V−1 ⊂ V0 ⊂ V1 ⊂ · · · with scaling by

dilation: f (t) ∈ Vj ⇔ f (2t) ∈ Vj+1 , and j Vj = {0}, j Vj = L2 . A scaling
T S

function ϕ generates V0 by integer shifts. The wavelet space Wj satisfies

Vj+1 = Vj ⊕ Wj .

Two-scale relations (filters)

√ √
ϕ(t) = 2 ∑ hk ϕ(2t − k ), ψ(t) = 2 ∑ gk ϕ(2t − k ),
k k

with quadrature mirror relations gk = (−1)k h1−k in the orthonormal case.

Connections

Wavelets are localized in time and scale (contrast Fourier’s global frequency
view, Chapter 2). Log-scale behavior links to Mellin analysis (Chapter 3).
Time–frequency tradeoffs connect to STFT (Chapter 9) but with dilation (not
translation) as the primary symmetry.

Complete details, proofs, and extended theory appear in the textbook ART OF INTE-
GRAL TRANSFORMS.

B. Solved Problems

ProblemR6.1 (Vanishing moments annihilate polynomials). Assume ψ

satisfies R tk ψ(t) dt = 0 for k = 0, 1, . . . , m − 1. Show that for any poly-
nomial P of degree < m,

(Wψ P)( a, b) = 0 for all a ̸= 0, b ∈ R.

−1
Solution. Write P(t) = ∑m k
k =0 ck ( t − b ) after shifting the center b. Then

t−b
 
1
Z
⟨ P, ψa,b ⟩ = P(t) p ψ dt.
| a| a

Substitute u = (t − b)/a:
q Z  m −1  q m −1 Z
⟨ P, ψa,b ⟩ = | a| ∑ ck ( au) k
ψ(u) du = | a| ∑ ck a k
uk ψ(u) du = 0.
k =0 k =0
········································································································································
38
 @ integra land

For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 6.2 (Haar scaling function from the two-scale equation). Let
h0 = h1 = √1 and suppose ϕ satisfies
2
√

2 h0 ϕ(2t) + h1 ϕ(2t − 1) = ϕ(2t) + ϕ(2t − 1),
ϕ(t) =

with ϕ ∈ L2 supported in [0, 1] and ϕ = 1. Identify ϕ.

R

Solution. The refinement equation with support in [0, 1] forces ϕ to be

constantR on dyadic subintervals and equal to 1 almost everywhere on [0, 1) to
match ϕ = 1. Thus ϕ(t) = 1[0,1) (t).
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 6.3 (Haar DWT of a length-8 signal). Compute the Haar wavelet
decomposition (two levels) of

x = [3, 1, 0, 4, 8, 6, 5, 5].
√
Use the orthonormal Haar averaging/differencing with factor 1/ 2.

(1) x2k + x2k+1 (1) x2k − x2k+1

Solution. Level 1 averages ak = √ and details dk = √ :
2 2
h i h i
a (1)
= √4 , √4 , √
14 √
, 10 , d (1)
= −4 , √2 , √0
√2 , √ .
2 2 2 2 2 2 2 2

Level 2 on a(1) :
h i h i
a (2) = 8 24
2, 2 = [4, 12], d (2) = 0 4
2, 2 = [0, 2].

So the two-level orthonormal Haar coefficients are ( a(2) , d(2) , d(1) ):

[4, 12 | 0, 2 | −4 , √2 , 0].
√2 , √
2 2 2

For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.
········································································································································
39
 @ integra land

Problem 6.4 (Admissibility forces zero mean). Assume ψ ∈ L2 (R) and

Z ∞ b
|ψ(ω )|2
Cψ = dω < ∞.
0 ω
R
b(0) = 0 and thus
Show ψ ψ(t) dt = 0.

Solution. If ψ b(ω )|2 stays bounded below near ω = 0, making

b(0) ̸= 0, then |ψ
R ε |ψb(ω )|2 R ε dω
dω diverge like 0 ω . Hence ψ (0) = 0, and by Fourier inversion
b
0 ω
b(0) = √1
R
ψ ψ(t) dt = 0.
2π
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 6.5 (CWT of a sinusoid: scale-frequency localization). Let f (t) =

cos(ω0 t) and assume ψ is a bandpass wavelet with ψ b concentrated near
ω = ωψ > 0. Explain where |(Wψ f )( a, b)| is largest as a function of a.

| a|e−iωb ψb( aω ).
p
Solution. In frequency, dilation scales frequencies: ψ
d a,b ( ω ) =
The sinusoid has spectral masses at ±ω0 , so the overlap is maximal when
aω0 ≈ ωψ , i.e.
ωψ
a≈ .
ω0
Time translation b contributes only a phase (for pure tones), so magnitude
depends primarily on scale.
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 6.6 (Energy conservation in an orthonormal DWT). For an or-

thonormal wavelet basis, show that the sum of squares of wavelet coeffi-
cients equals ∥ f ∥22 .

Solution. In an orthonormal basis {ψj,k } and scaling part {ϕ J,k },

f = ∑ a J,k ϕ J,k + ∑ ∑ d j,k ψj,k ,

k j≥ J k
········································································································································
40
 @ integra land

and Parseval in Hilbert space gives

∥ f ∥22 = ∑ | a J,k |2 + ∑ ∑ |d j,k |2 .

k j≥ J k

For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 6.7 (A soft-threshold denoising step). Detail coefficients at a

scale are
d = [−0.12, 0.04, 1.35, −0.80, 0.09].
Apply soft-thresholding with threshold τ = 0.1.

Solution. Soft-threshold: Sτ ( x ) = sgn( x ) max{| x | − τ, 0}. Thus

dnew = [−0.02, 0, 1.25, −0.70, 0].

For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 6.8 (Wavelet view of scaling vs Mellin translation). Let f λ (t) =

f (λt), λ > 0. Show how CWT coefficients transform.

Solution. Z
(Wψ f λ )( a, b) = f (λt)ψa,b (t) dt.

Let u = λt:
1
Z Z
(Wψ f λ )( a, b) = f (u)ψa,b (u/λ) du = f (u)ψaλ, bλ (u) du = (Wψ f )( aλ, bλ).
λ
So dilation in time translates in log-scale, paralleling Mellin scaling laws
(Chapter 3).
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

C. Challenge Problems (no solutions)

1. Construct a compactly supported orthonormal wavelet with two van-
ishing moments and derive the associated filter bank constraints.
········································································································································
41
 @ integra land

2. Prove the Calderón reproducing formula and derive the exact recon-
struction constant Cψ for a given ψ.
3. Compare STFT and wavelet tilings of the time–frequency plane (Chap-
ter 9); quantify resolutions.
4. Show how wavelet shrinkage yields minimax rates over Besov classes
(outline the key inequality).
5. Derive the continuous wavelet transform of a chirp and explain scale-
dependent ridge behavior (compare with FrFT, Chapter 8).
6. Establish the connection between MRA and subdivision schemes for
refinement equations.

These are developed in depth in ART OF INTEGRAL TRANSFORMS.

········································································································································
42
 @ integra land

7
Hankel Transform and Radial Symmetry
Problems

A. Short Theory
Definition (order ν)

Definition 7.1 (Hankel transform). For ν > −1 and suitable f ,

Z ∞
(Hν f )(k) = f (r ) Jν (kr ) r dr,
0

where Jν is the Bessel function of the first kind.

A common inversion (with matching convention) is

Z ∞
f (r ) = (Hν f )(k) Jν (kr ) k dk.
0

Plancherel

Under appropriate hypotheses,

Z ∞ Z ∞
2
| f (r )| r dr = |(Hν f )(k)|2 k dk.
0 0

Radial Fourier connection

In Rd , the Fourier transform of a radial function reduces to a Hankel transform

of order ν = d2 − 1. Thus Hankel transforms are the natural diagonalization
tool for radial PDEs (compare Chapter 2).
········································································································································
43
 @ integra land

Operational rules

Differential operators in r tied to radial Laplacians become multipliers in

k under Hankel, mirroring Fourier: the radial Laplacian in 2D, urr + 1r ur ,
corresponds to multiplication by −k2 under the order-0 Hankel transform.

Applications

Radial heat and wave equations, Bessel operator eigenfunction expansions,

axisymmetric potential problems, optics (Fraunhofer with circular symmetry),
special-function integral evaluations.

Complete details, proofs, and extended theory appear in the textbook ART OF INTE-
GRAL TRANSFORMS.

B. Solved Problems

Problem 7.1 (A Gaussian–Bessel integral). For a > 0, evaluate

Z ∞
2
e− ar J0 (br ) r dr.
0

2
Solution. This is the order-0 Hankel transform of f (r ) = e−ar at frequency b.
The known closed form is
Z ∞  2
2 1 b
e− ar J0 (br ) r dr = exp − .
0 2a 4a

For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 7.2 (Radial heat equation in 2D). Solve for r > 0:

 
1 2
ut = κ urr + ur , u(r, 0) = e− ar , a > 0.
r

Solution. Take the order-0 Hankel transform in r:

Z ∞
2 1 −k2 /(4a)
∂t U (k, t) = −κk U (k, t), 2
U (k, 0) = e− ar J0 (kr ) r dr = e .
0 2a
········································································································································
44
 @ integra land

1 −(κt+1/(4a))k 2
Thus U (k, t) = 2a e . Invert using the same Gaussian–Bessel iden-
tity (self-reciprocal structure) to obtain
ar2
 
1
u(r, t) = exp − .
1 + 4aκt 1 + 4aκt
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 7.3 (A modified Bessel kernel from a Hankel inversion). Show

that for a > 0 and r > 0,
Z ∞
k
J0 (kr ) dk = K0 ( ar ),
0 k 2 + a2
where K0 is the modified Bessel function of the second kind.

Solution. Interpret the left side as the inverse Hankel transform of F (k ) =

1
k 2 + a2
(up to the k measure). The standard Hankel pair is exactly

1
Hν {K0 ( ar )}(k) = ,
k 2 + a2
hence the inversion yields the stated identity.
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 7.4 (Radial Fourier in R3 reduces to a sine transform). For a

radial function f ( x ) = F (| x |) on R3 , show that its Fourier transform is
radial and can be written as
2 ∞
r Z
sin(ρr ) 2
fb(ρ) = F (r ) r dr.
π 0 ρr
q
3 1 2
Solution. In d = 3, ν = − 1 = Using the identity J1/2 (z) = πz
2 2. sin z and
the general radial Fourier–Hankel reduction, the radial transform becomes an
order-1/2 Hankel transform, which simplifies to the displayed sine-kernel
form.
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.
········································································································································
45
 @ integra land

Problem 7.5 (Bessel orthogonality as a distributional identity). State the

orthogonality relation (in distribution sense)
Z ∞
1
Jν (kr ) Jν (kr ′ ) k dk = δ(r − r ′ ).
0 r

Solution. This is the kernel statement of Hankel self-inversion:

Z ∞ Z ∞ 
f (r ) = f (r ′ ) Jν (kr ′ ) r ′ dr ′ Jν (kr ) k dk.
0 0
R∞
Comparing with f (r ) = 0 f (r ′ )δ(r − r ′ ) dr ′ , one reads off
Z ∞
1
Jν (kr ) Jν (kr ′ ) k dk = δ(r − r ′ ).
0 r
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 7.6 (A radial Poisson problem (2D, tempered behavior)). Solve

2
−∆u = f on R2 for radial f (r ) = e−ar in transform form.

Solution. For radial data in 2D, apply order-0 Hankel:

FH (k )
k2 U (k) = FH (k ) ⇒ U (k) =
,
k2
Z ∞
2 1 2
FH (k ) = e− ar J0 (kr ) r dr = e−k /(4a) .
0 2a
2
1 e−k /(4a)
Thus U (k ) = 2a k2
, and
Z ∞ 2
1 ∞ e−k /(4a)
Z
u (r ) = U (k) J0 (kr ) k dk = J0 (kr ) dk,
0 2a 0 k
a logarithmic-type potential (consistent with 2D fundamental solutions).
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 7.7 (Radial convolution becomes Hankel multiplication (concep-

tual computation)). Let f , g be radial on R2 with order-0 Hankel trans-
forms FH , GH . State the transform-domain form of the radial convolution
h = f ∗ g.
········································································································································
46
 @ integra land

Solution. Radial convolution in R2 corresponds to multiplication of 2D

Fourier transforms. Under the radial Fourier–Hankel identification, this
becomes
HH (k ) = 2π FH (k ) GH (k )
(up to the chosen normalization constants). Thus Hankel is the natural algebra
for radial convolutions, mirroring Fourier convolution (Chapter 2) and FFT
computation (Chapter 10).
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 7.8 (A Bessel moment integral from Hankel calculus). Evaluate

Z ∞
J0 (r ) e−ar dr, a > 0.
0

Solution. Use the Laplace–Bessel identity:

Z ∞
1
e−ar J0 (r ) dr = √ .
0 a2 + 1
(One route: view J0 as the inverse Hankel of a compact spectral measure;
another route uses known Laplace pairs.)
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

C. Challenge Problems (no solutions)

1. Derive the full radial Fourier transform formula in Rd and track con-
stants precisely for the unitary normalization.
2. Solve the radial wave equation in R3 using the order-1/2 Hankel trans-
form and compare with Kirchhoff’s formula.
2
3. Compute Hν {r α e− ar } in terms of confluent hypergeometric functions.
4. Derive a Hankel convolution theorem with explicit kernel and apply it
to compute a nontrivial radial integral.
5. Use Hankel methods to solve an axisymmetric boundary-value problem
on a disk (Bessel series meets Hankel).
········································································································································
47
 @ integra land

6. Relate Hankel transforms to the FrFT viewpoint for radial Gaussian

optics (Chapter 8).

These are developed in depth in ART OF INTEGRAL TRANSFORMS.

········································································································································
48
 @ integra land

8
Fractional Fourier Transform (FrFT) and
Time–Frequency Analysis

A. Short Theory
Definition (one common unitary convention)

Definition 8.1 (FrFT of order/angle α). For α ̸≡ kπ,

1 − i cot α ∞
r  2
t + u2
Z 
(Fα f )(u) = exp i cot α − itu csc α f (t) dt.
2π −∞ 2
For α = 2kπ, Fα f = f ; for α = (2k + 1)π, Fα f (u) = f (−u).

At α = π/2, this reduces to the unitary Fourier transform of Chapter 2.

Core properties

• Additivity: Fα ◦ F β = Fα+ β (angles add modulo 2π).

• Unitarity: ∥Fα f ∥2 = ∥ f ∥2 .

• Eigenfunctions: Hermite functions hn satisfy Fα hn = e−inα hn (up to

convention-dependent global phases).

• Time–frequency rotation: Fα corresponds to a rotation by α in the (t, ω )

plane for quadratic-phase structures, making it ideal for chirps.

Connections

• α = π/2 is Fourier, α = 0 is identity.

········································································································································
49
 @ integra land

• STFT (Chapter 9) analyzes local spectra; FrFT rotates the global axes
before spectral reading.
• FFT (Chapter 10) computes discrete Fourier; discrete FrFT variants exist
for fast computation.

Complete details, proofs, and extended theory appear in the textbook ART OF INTE-
GRAL TRANSFORMS.

B. Solved Problems

Problem 8.1 (Composition law). Show formally that Fα (F β f ) = Fα+ β f

(mod 2π) for nice f .

Solution. The kernel is a quadratic-phase Gaussian integral. Composing two

FrFT kernels yields another quadratic kernel whose parameters add as angles
(a symplectic/rotation identity). Thus the operator forms a one-parameter
unitary group:
Fα F β = Fα+ β .
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 8.2 (FrFT at special angles). Verify that Fπ/2 = F (unitary

Fourier) and Fπ f (u) = f (−u).

Solution. At α = π/2, cot α = 0 and csc α = 1, so the kernel becomes

√1 e−itu , i.e. the unitary Fourier transform. At α = π, the operator corre-
2π
sponds to rotation by π in phase space, giving time reversal: f (−u).
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 8.3 (A fixed point: the ground Hermite function). Let h0 (t) =
2
π −1/4 e−t /2 . Compute Fα h0 .

Solution. Hermite functions are FrFT eigenfunctions. For n = 0,

F α h0 = h0
········································································································································
50
 @ integra land

under this convention (consistent with h0 being Fourier-invariant and even

under time reversal).
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 8.4 (FrFT of a shifted delta). Compute Fα {δ(t − t0 )}(u) for α ̸≡

kπ.

Solution. Insert δ(t − t0 ) into the integral:

r !
2 + u2
1 − i cot α t
Fα {δ(· − t0 )}(u) = exp i 0 cot α − it0 u csc α .
2π 2
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 8.5 (Chirp alignment: choosing the “right” fractional order). A

linear chirp has instantaneous frequency ω (t) = ω0 + at. Explain how to
choose α so that the FrFT concentrates energy near a single frequency in
the rotated domain.

Solution. In the (t, ω ) plane the chirp is a line of slope a. A rotation by α

aligns this line with the new frequency axis when
tan α ≈ a
(up to convention-dependent scaling). Then the chirp becomes approximately
a tone in the rotated coordinates, producing concentration.
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 8.6 (Unitarity check via Plancherel). State the L2 energy conser-
vation for FrFT and give the one-line justification.

Solution.
∥Fα f ∥2 = ∥ f ∥2 .
Justification: {Fα }α∈R is a unitary group generated by a self-adjoint oper-
ator (harmonic oscillator viewpoint), or directly from kernel unitarity and
composition F−α as inverse.
········································································································································
51
 @ integra land

For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 8.7 (A FrFT identity connecting to STFT). Explain why FrFT is

often more effective than Fourier for globally chirped signals, whereas
STFT is better for locally time-varying spectra.

Solution. FrFT performs a global rotation in time–frequency, turning chirps

(tilted lines) into concentrated features. STFT uses windowing, giving lo-
cal frequency content and tracking nonstationarity. Thus: FrFT excels for
quadratic-phase (global) structure; STFT excels when spectral content changes
with time.
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 8.8 (Hermite eigenvalues across angles). Assuming Fα hn =

e−inα hn , compute Fπ/2 hn and relate to classical Fourier eigenvalues.

Solution. At α = π/2,
Fπ/2 hn = e−inπ/2 hn = (−i )n hn ,
which matches the classical Fourier eigenvalue pattern for Hermite functions
under the unitary convention.
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

C. Challenge Problems (no solutions)

1. Derive the FrFT kernel from the harmonic oscillator propagator and
carefully track global phase conventions.
2. Compute the FrFT of a rectangular pulse and describe its caustics/oscillatory
behavior.
3. Build a fast discrete FrFT algorithm and compare complexity with FFT
(Chapter 10).
4. Prove a rotation theorem for the Wigner distribution under FrFT and
relate to chirp detection.
········································································································································
52
 @ integra land

5. Derive uncertainty principles in rotated coordinates and compare with

STFT (Chapter 9).
6. Solve a PDE with quadratic potential using FrFT diagonalization.

These are developed in depth in ART OF INTEGRAL TRANSFORMS.

········································································································································
53
 @ integra land

········································································································································
54
 @ integra land

9
Short-Time Fourier Transform (STFT)

A. Short Theory
Definition

Definition 9.1 (STFT). For f , g ∈ L2 (R) (window g),

Z ∞
(Vg f )( x, ω ) = f (t) g(t − x ) e−iωt dt.
−∞

The spectrogram is |Vg f ( x, ω )|2 .

Key identities

• Covariance: time shift and modulation move the STFT in ( x, ω ) up to

phases.
• Moyal identity:
dx dω
Z Z
|Vg f ( x, ω )|2 = ∥ f ∥22 ∥ g∥22 .
R R 2π
• Reconstruction (one standard form):
1
Z Z
f (t) = 2
(Vg f )( x, ω ) g(t − x ) eiωt dx dω,
2π ∥ g∥2 R R
in L2 sense.

Resolution tradeoff

Short windows localize in time but blur frequency; long windows sharpen
frequency but smear time. Gaussian windows saturate the uncertainty bound
and interact cleanly with Fourier/FrFT (Chapters 2, 8).
········································································································································
55
 @ integra land

Connections

• Hilbert transform (Chapter 5) provides analytic signals; applying STFT

to analytic signals avoids negative-frequency artifacts.
• Wavelets (Chapter 6) replace fixed window length with scale-dependent
windows.
• FFT (Chapter 10) is the computational engine for discrete STFT (sliding
FFT).

Complete details, proofs, and extended theory appear in the textbook ART OF INTE-
GRAL TRANSFORMS.

B. Solved Problems

Problem 9.1 (STFT of a pure tone). Let f (t) = eiω0 t and g ∈ L1 ∩ L2 .

Compute Vg f ( x, ω ).

Solution.
Z Z
−iωt
Vg f ( x, ω ) = e iω0 t
g(t − x )e dt = g(t − x )e−i(ω −ω0 )t dt.

Let u = t − x:
− i ( ω − ω0 ) x
Z √
Vg f ( x, ω ) = e g(u)e−i(ω −ω0 )u du = e−i(ω −ω0 ) x gb(ω − ω0 ) 2π

(with gb the unitary Fourier transform). Thus magnitude depends only on

ω − ω0 .
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 9.2 (Gaussian window gives Gaussian frequency localization).

2 2
Let g(t) = e−t /(2σ ) and f (t) = eiω0 t . Find |Vg f ( x, ω )|.

Solution. From the previous problem, |Vg f | is proportional to | gb(ω − ω0 )|.

The Fourier transform of a Gaussian is Gaussian:
σ 2 2
gb(ξ ) = √ e−σ ξ /2 (up to unitary constants),
1
········································································································································
56
 @ integra land

so
2 (ω −ω 2 /2
|Vg f ( x, ω )| = Cσ e−σ 0) ,
independent of x for a pure tone.
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 9.3 (Covariance under time shift and modulation). Let

( Tx0 f )(t) = f (t − x0 ) and ( Mω0 f )(t) = eiω0 t f (t). Show

Vg ( Mω0 Tx0 f )( x, ω ) = e−i(ω −ω0 ) x0 Vg f ( x − x0 , ω − ω0 ).

Solution. Insert Mω0 Tx0 f into the definition and substitute t′ = t − x0 . Phase
factors separate: time shift moves the window center; modulation shifts
frequency. The remaining integral is exactly Vg f at shifted ( x, ω ), with the
displayed phase.
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 9.4 (Moyal identity as a one-line orthogonality statement). State

Moyal’s identity and explain why it implies energy conservation of the
spectrogram.

Solution. Moyal:

dx dω
ZZ
|Vg f ( x, ω )|2 = ∥ f ∥22 ∥ g∥22 .
2π
Thus the total spectrogram energy (integrated over time–frequency) equals
the signal energy times the window energy: the spectrogram is an energy
redistribution, not an energy creation.
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 9.5 (Uncertainty: Gaussian window saturates the bound). For

a normalized Gaussian window, state the time–frequency uncertainty
product and its optimal value.
········································································································································
57
 @ integra land

2 / (2σ2 )
Solution. For Gaussian g(t) ∝ e−t , the standard deviations satisfy
1
∆t ∆ω =
2
(in unitary Fourier normalization). This is minimal among L2 windows,
explaining Gaussian popularity in STFT and FrFT contexts (Chapter 8).
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

a 2
Problem 9.6 (STFT of a linear chirp: ridge structure). Let f (t) = ei(ω0 t+ 2 t )
and take a localized window g. Describe where the STFT magnitude
concentrates.

Solution. The chirp has instantaneous frequency ω (t) = ω0 + at. A localized

window around x sees approximately a tone at ω0 + ax, so
|Vg f ( x, ω )| is largest near ω ≈ ω0 + ax,
producing a ridge line in the time–frequency plane. This is the local analog of
FrFT chirp alignment (Chapter 8).
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 9.7 (Reconstruction formula in practice). Assume g ̸= 0 and

f , g ∈ L2 . Write the synthesis formula that reconstructs f from Vg f .

Solution. One standard L2 reconstruction is

1
Z Z
f (t) = 2
(Vg f )( x, ω ) g(t − x ) eiωt dx dω.
2π ∥ g∥2 R R
It is a continuous superposition of time-shifted windows modulated by the
STFT coefficients.
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 9.8 (Analytic signal + STFT to suppress negative frequencies).

Let f be real and set f a = f + i H f (analytic signal). Explain why Vg f a has
primarily positive-frequency support.
········································································································································
58
 @ integra land

Solution. In Fourier terms, f a is the positive-frequency projection of f

(Chapter 5). STFT is a localized Fourier analysis; applying it to f a largely
removes negative-frequency interference, producing cleaner time–frequency
ridges for real oscillatory signals.
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

C. Challenge Problems (no solutions)

1. Prove the inversion formula for STFT under minimal assumptions on g
and interpret it as a resolution of the identity.
2. Derive frame conditions for Gabor systems {eimbt g(t − na)} and com-
pare with wavelet frame conditions (Chapter 6).
3. Quantify spectral leakage for a sinusoid whose frequency lies between
DFT bins; relate to the Dirichlet kernel (Chapter 10).
4. Compute the STFT of a sum of two chirps and derive a separation
condition for ridge resolvability.
5. Connect STFT to pseudo-differential operators: interpret window choice
as symbol smoothing.
6. Compare STFT and FrFT on the same chirp and show when global
rotation beats local windowing.

These are developed in depth in ART OF INTEGRAL TRANSFORMS.

········································································································································
59
 @ integra land

········································································································································
60
 @ integra land

10
FFT

A. Short Theory
DFT definition

For x [0], . . . , x [ N − 1], define the DFT

N −1
X [k] = ∑ x [n] e−i2πnk/N , k = 0, . . . , N − 1,
n =0

with inverse
N −1
1
x [n] =
N ∑ X [k] ei2πnk/N .
k =0
The DFT is Fourier analysis on the finite cyclic group Z N .

FFT idea

A direct DFT costs O( N 2 ) operations. FFT algorithms exploit factorization of

N and symmetry of roots of unity to compute the same output in O( N log N ).
The Cooley–Tukey radix-2 method (for N = 2m ) is the canonical example:
split even/odd indices, recurse, combine with “twiddle factors”.

Convolution and FFT

Circular convolution corresponds to pointwise multiplication in the DFT

domain:
DFT( x ⊛ h) = X · H.
Linear convolution is computed by zero-padding to length at least N ≥
Nx + Nh − 1 and then using circular convolution at that padded length.
········································································································································
61
 @ integra land

FFT in signal pipelines

FFT is the computational backbone for:

• spectral estimation and filtering (connect to Z-transform, Chapter 4),
• fast correlation and matched filtering,
• discrete STFT (Chapter 9),
• spectral PDE solvers on periodic grids.

Complete details, proofs, and extended theory appear in the textbook ART OF INTE-
GRAL TRANSFORMS.

B. Solved Problems

Problem 10.1 (DFT of a sparse periodic pattern). Let N = 8 and x =

[1, 0, 1, 0, 1, 0, 1, 0]. Compute X [k].

Solution. This is x [n] = 1 for even n, 0 for odd n. So

3 3
X [k] = ∑ e −i2πk(2m)/8
= ∑ e−i2πkm/4 .
m =0 m =0

This sum equals 4 if k ≡ 0 (mod 4), and 0 otherwise. Hence

X = [4, 0, 0, 0, 4, 0, 0, 0].

For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 10.2 (Circular convolution by FFT logic). Let x = [1, 2, 0, 1] and

h = [1, −1, 2, 0] (length 4). Compute the circular convolution y = x ⊛ h.

Solution. By definition, y[n] = ∑3m=0 x [m]h[n − m(mod 4)]. Compute

quickly:
y[0] = 1 · 1 + 2 · 0 + 0 · 2 + 1 · (−1) = 0,
y[1] = 1 · (−1) + 2 · 1 + 0 · 0 + 1 · 2 = 3,
y[2] = 1 · 2 + 2 · (−1) + 0 · 1 + 1 · 0 = 0,
········································································································································
62
 @ integra land

y[3] = 1 · 0 + 2 · 2 + 0 · (−1) + 1 · 1 = 5.
So y = [0, 3, 0, 5]. (FFT would compute the same via Y = XH in k-space.)
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 10.3 (Linear convolution via zero-padding). Let x = [1, 2, 3] and

h = [1, 0, −1]. Compute the linear convolution.

Solution. Linear convolution length is 3 + 3 − 1 = 5:

y[0] = 1, y[1] = 2, y[2] = 3 − 1 = 2, y[3] = −2, y[4] = −3.

So y = [1, 2, 2, −2, −3]. (FFT method: zero-pad both to length N ≥ 5, DFT

multiply, inverse DFT.)
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 10.4 (Operation count comparison). Estimate the ratio of multi-

plications for direct DFT vs FFT when N = 1024.

Solution. Direct DFT: O( N 2 ) ≈ 10242 ≈ 1.05 × 106 complex mults. Radix-2

FFT: ≈ N2 log2 N = 512 · 10 = 5120 complex mults (order-of-magnitude).
Ratio ≈ 1.05 × 106 /5120 ≈ 205.
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 10.5 (Polynomial multiplication as convolution). Let p(z) =

1 + 2z + z2 and q(z) = 1 − z + 2z2 . Compute p(z)q(z) using convolution
logic.

Solution. Coefficient vectors: p = [1, 2, 1], q = [1, −1, 2]. Convolution gives
coefficients:

r0 = 1, r1 = 2 − 1 = 1, r2 = 1 − 2 + 2 = 1, r3 = 4 − 1 = 3, r4 = 2.

So p(z)q(z) = 1 + z + z2 + 3z3 + 2z4 . (FFT computes this quickly for high

degree.)
········································································································································
63
 @ integra land

For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 10.6 (Periodic Poisson solver by FFT diagonalization). On Z N

with periodic boundary, solve

u[n + 1] − 2u[n] + u[n − 1] = f [n], n = 0, . . . , N − 1,

assuming ∑n f [n] = 0 (compatibility).

Solution. DFT diagonalizes shifts: if U [k] is the DFT of u, then

(ei2πk/N − 2 + e−i2πk/N )U [k] = F [k],
i.e. (−4 sin2 (πk/N ))U [k ] = F [k ]. For k ̸= 0,
F [k]
U [k] = − .
4 sin2 (πk/N )
Set U [0] = 0 (mean zero) and invert by inverse DFT.
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 10.7 (Fast correlation identity). Let ( x ⋆ y)[n] = ∑m x [m]y[m + n]

(circular correlation). Show how FFT computes x ⋆ y from DFTs.

Solution. DFT of correlation satisfies

DFT( x ⋆ y) = X [k] Y [k].
Thus: FFT x 7→ X, FFT y 7→ Y, multiply X · Y, inverse FFT to obtain correla-
tion.
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

Problem 10.8 (Spectral leakage formula for an off-bin sinusoid). Let x [n] =
/ Z. Express X [k ] in closed form.
ei2π (m+δ)n/N with integer m and δ ∈

Solution. Compute the geometric sum:

N −1 N −1
X [k] = ∑ e i2π (m+δ)n/N −i2πkn/N
e = ∑ ei2π (m+δ−k)n/N .
n =0 n =0
········································································································································
64
 @ integra land

Thus
1 − ei2π (m+δ−k)


sin π ( m + δ − k )
X [k] = i2π ( m + − k ) /N
= eiπ ( N −1)(m+δ−k)/N ·
.
1−e δ sin π (m + δ − k )/N
Magnitude is a Dirichlet-kernel profile, explaining leakage.
For full derivations, alternative methods, and deeper extensions, see ART OF INTE-
GRAL TRANSFORMS.

C. Challenge Problems (no solutions)

1. Derive radix-2 Cooley–Tukey FFT from the DFT definition and show
the exact butterfly structure.
2. Prove that zero-padding increases interpolation density in the frequency
domain but does not increase true resolution.
3. Implement overlap-add / overlap-save convolution and analyze com-
plexity vs direct time-domain filtering.
4. Construct a numerically stable FFT-based method for computing the
Hilbert transform of discrete data (connect to Chapter 5).
5. Build an FFT-based STFT with a chosen window and hop size; derive
the perfect reconstruction condition (connect to Chapter 9).
6. Analyze roundoff error propagation in FFT recursion and propose miti-
gation strategies.

These are developed in depth in ART OF INTEGRAL TRANSFORMS.

········································································································································
65
 @ integra land

········································································································································
66
 @ integra land

ACompact Transform-Pairs Appendix (high-

value)

How to use this appendix

These are core pairs and identities that unlock most computations in the solved
problems. For larger tables, derivations, and parameter variants, consult ART
OF INTEGRAL TRANSFORMS.

Laplace pairs (unilateral)

f (t) F (s) = L{ f }(s)

1 1/s
tn (n ∈ N) n!/sn+1
tα−1 /Γ(α) (α > 0) 1/sα
e at 1/(s − a)
cos(ωt) s/(s2 + ω 2 )
sin(ωt) ω/(s2 + ω 2 )
H (t − a) e−as /s
δ(t − a) e−as
δ(n) ( t − a ) sn e− as
(e−at − e√−bt )/t ln ss+ b


√+a
e−t erfi( t) 1/( s(s + 1))

Fourier pairs (unitary convention)

········································································································································
67
 @ integra land

f (t) fb(ω ) = F { f }(ω )

2 −ω 2 /2
e−t /2 eq
e− a|t| (a > 0) 2 a
π a2 + ω 2
1 q
π 1 − a|ω |
(a > 0) 2 ae
t2 + a2
1
δ ( t − t0 ) √ e−iωt0
√ 2π
1 q δ(ω )
2π
PV(1/t) −i π2 sgn(ω )

Mellin pairs

f (x) F (s) = M{ f }(s)

e− x Γ(s)
(1 + x ) −1 Γ(s)Γ(1 − s) = π/ sin(πs)
(1 + x p ) −1 π
p csc( πs/p )
(log x ) f ( x ) F ′ (s)
f ( ax ) a−s F (s )

Z-transform essentials

x [n] (typically n ≥ 0) X ( z ) = ∑ n ≥0 x [ n ] z − n
z
u[n] z −1
z
an u[n] z− a
az
nan u[n] ( z − a )2
δ [ n − n0 ] z − n0

Hilbert transform Fourier multiplier

F {H f }(ω ) = (−i sgn ω ) fb(ω ).

Hankel transform (order ν) identity

Z ∞ Z ∞
(Hν f )(k) = f (r ) Jν (kr ) r dr, f (r ) = (Hν f )(k) Jν (kr ) k dk.
0 0

········································································································································
68
 @ integra land

BFurther Reading
If this mini-book felt like a useful “methods upgrade,” the natural continua-
tion is the full text:

• ART OF INTEGRAL TRANSFORMS (primary reference). Complete

proofs, careful distributional frameworks, multiple convention maps,
extended transform tables, and large graded problem sets (including
many of the challenge problems solved in several ways).

Additional standard references (for complementary perspectives):

• Classic transform methods for ODE/PDE and physics.
• Harmonic analysis texts for Fourier/Hilbert and L p theory.
• Wavelet and time–frequency analysis monographs for MRA, frames, and
applications.
• Numerical analysis and signal processing references for FFT-based pipelines.

Complete details, proofs, and extended theory appear in the textbook ART OF INTE-
GRAL TRANSFORMS.

········································································································································
69
 @ integra land

Integral Transforms Problem Bank

Conventions (used throughout).

Z ∞
• Laplace (unilateral): L{ f (t)}(s) = e−st f (t) dt.
0
Z ∞
• Fourier (non-unitary): F { f }(ω ) = f (t)e−iωt dt, F −1 { F }(t) =
Z ∞ −Z∞
1 1
Z
iωt 2
F (ω )e dω. Hence Plancherel: | f | dt = |F f |2 dω.
2π −∞ 2π
∞
• Z-transform (bilateral): X (z) = Z { x [n]}(z) = ∑ x [n]z−n with ROC
n=−∞
logic; causal/unilateral: X + (z) = ∑∞
n =0 x [ n ] z−n .
1 ∞ f (t) Z
• Hilbert transform: (H f )( x ) = PV dt. With the above
π −∞ x − t
Fourier convention: F {H f }(ω ) = (−i sgn ω ) F { f }(ω ).
Z ∞
• Mellin: M{ f }(s) = x s−1 f ( x ) dx (strip of convergence).
0
Z ∞
• Hankel (order ν > −1): Hν { f }(k ) = f (r ) Jν (kr ) r dr, with inversion
Z ∞ 0
f (r ) = (Hν f )(k) Jν (kr ) k dk.
0
Z ∞
• STFT: Vg f ( x, ω ) = f (t) g(t − x )e−iωt dt.
−∞

t−b
 
1
Z
• CWT wavelet: Wψ f ( a, b) = f (t) p ψ dt.
| a| a
• FrFT (matching the Fourier kernel at α = π/2): for α ̸≡ kπ,
Z ∞  
i 2 2
(Fα f )(u) = exp 2 (t + u ) cot α − iut csc α f (t) dt,
−∞
and Fα |α=π/2 = F , Fα |α=0 = Id (by definition).

Problem 1: Log-ratio Laplace inversion [Laplace]

n  o
Find f (t) = L −1 s+b
ln s+ a (t) for a, b > 0.
········································································································································
1
 @ integra land

Solution. Use the standard Laplace identity (Frullani form)

Z ∞ −(s+ a)t
− e−(s+b)t
 
e s+b
dt = ln .
0 t s+a
Hence
e−at − e−bt
f (t) = , t > 0.
t

Problem 2: Cauchy kernel spectrum [Fourier]



1
Compute F 2 (ω ) for a > 0.
t + a2
Solution. A standard contour evaluation gives
Z ∞
e−iωt π − a|ω |
dt = e .
−∞ t2 + a2 a
π − a|ω |
Thus F {(t2 + a2 )−1 }(ω ) = e .
a

Problem 3: Two-sided Z-inversion from a strip ROC [Z]

Let
z 1
X (z) = , ROC: 2 < |z| < 2.
(z − 2)(z − 12 )
Find x [n].
Solution. Partial fractions:
z 4/3 1/3
1
= − .
(z − 2)(z − 2 ) z − 2 z − 12

Note 1/(z − a) = z−1 · z/(z − a). In the given ROC, the a = 12 term is
right-sided, while the a = 2 term is left-sided:
( )  
−1 1 1 n −1 −1 1
Z 1
= ( 2 ) u [ n − 1], Z = −2 n−1 u[−n].
z− 2 z−2

Therefore
4 n −1 1  1  n −1
x [n] = − 2 u[−n] − u [ n − 1].
3 3 2
········································································································································
2
 @ integra land

Problem 4: Scale-law integral [Mellin]

For p > 0 and 0 < ℜ(s) < p, evaluate

Z ∞ s −1
x
I (s) = dx.
0 1 + xp

Solution. Let u = x p , so dx = 1p u1/p−1 du:

Z ∞ s/p−1
1 u 1 s  s π  πs 
I (s) = du = Γ Γ 1− = csc .
p 0 1+u p p p p p

Problem 5: A principal-value identity (Hilbert kernel test) [Hilbert]

Evaluate, for x ∈ R, Z ∞
1
PV dt.
−∞ ( x − t)(1 + t2 )
Solution. By definition,
1 f (t) 1
Z
(H f )( x ) = PV dt, f (t) = .
π x−t 1 + t2
x
A standard Hilbert pair is H{(1 + t2 )−1 }( x ) = . Hence the PV integral
1 + x2
equals Z ∞
1 x
PV dt = π .
−∞ ( x − t)(1 + t2 ) 1 + x2

Problem 6: Gaussian–Bessel integral [Hankel]

For a > 0, evaluate Z ∞

2
e− ar J0 (br ) r dr.
0

Solution. This is the order-0 Hankel transform of a radial Gaussian:

Z ∞  2
2 1 b
e− ar J0 (br ) r dr = exp − .
0 2a 4a

········································································································································
3
 @ integra land

Problem 7: STFT of a monochromatic tone [STFT]

Let f (t) = eiω0 t and let g ∈ L1 ∩ L2 . Compute Vg f ( x, ω ) in terms of F { g}.

Solution.
Z Z
−iωt
Vg f ( x, ω ) = e iω0 t
g(t − x )e dt = g(t − x )e−i(ω −ω0 )t dt.

With u = t − x,
Z
− i ( ω − ω0 ) x
Vg f ( x, ω ) = e g(u)e−i(ω −ω0 )u du = e−i(ω −ω0 ) x F { g}(ω − ω0 ).

For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.

Problem 8: Admissibility forces zero mean [Wavelet]

Assume ψ ∈ L2 (R) satisfies the wavelet admissibility condition

Z ∞
|F {ψ}(ω )|2
Cψ = dω < ∞.
0 ω
R∞
Show that −∞ ψ ( t ) dt = 0.
Solution. If F {ψ}(0) ̸= 0, then |F {ψ}(ω )|2 /ω behaves Rlike c/ω near 0,
forcing
R divergence. Hence F {ψ}(0) = 0. Since F {ψ}(0) = ψ(t) dt, we get
ψ = 0.

Problem 9: FrFT of a shifted delta [FrFT]

For α ̸≡ kπ, compute Fα {δ(t − t0 )}(u).

Solution. Insert δ(t − t0 ) into the defining integral:
 
i 2 2
Fα {δ(· − t0 )}(u) = exp 2 (t0 + u ) cot α − iut0 csc α .

Problem 10: Half-line heat with delayed boundary activation [Laplace]

Solve ut = κu xx for x > 0, t > 0 with

u( x, 0) = 0, u(0, t) = H (t − T ), lim u( x, t) = 0,
x →∞
········································································································································
4
 @ integra land

where κ > 0, T > 0, and H is the Heaviside step.

√
Solution. Laplace in t gives sU = κUxx , bounded solution U ( x, s) = A(s)e − x s/κ .

Boundary: U (0, s) = L{ H (t − T )} = e−Ts /s. Hence

e−Ts
 r 
s
U ( x, s) = exp − x .
s κ
√
Use L−1 {e−c s /s} = erfc √ c


and the time shift T:
2 t
!
x
u( x, t) = H (t − T ) erfc p .
2 κ (t − T )

Problem 11: Quadrature pair via Hilbert multiplier [Hilbert↔Fourier]

Show that for ω0 > 0,

H{cos(ω0 t)} = sin(ω0 t), H{sin(ω0 t)} = − cos(ω0 t).

Solution. F {cos(ω0 t)} = π (δ(ω − ω0 ) + δ(ω + ω0 )). Multiply by −i sgn(ω ):

−i sgn(ω )F {cos(ω0 t)} = −iπ (δ(ω − ω0 ) − δ(ω + ω0 )) = F {sin(ω0 t)}.
Similarly, F {sin(ω0 t)} = π
i (δ(ω − ω0 ) − δ(ω + ω0 )) gives F {H sin} =
−F {cos}.

Problem 12: Sampling bridge [Z↔Laplace]

Let f (t) = e at H (t) with a ∈ C and sampling period T > 0. Define x [n] =
f (nT ) = e anT u[n]. Find X + (z) and identify the pole in terms of a and T.
Solution.
∞
1 z
+
X (z) = ∑ eanT z−n = 1 − eaT z−1 = z − eaT , ROC: |z| > |e aT |.
n =0

Pole at z = e aT (compare with the Laplace pole at s = a under z = esT ).

Problem 13: A clean Mellin scaling evaluation [Mellin]

Let 0 < β < 1 and λ > 0. Evaluate

Z ∞ β −1
x
I (λ) = dx.
0 1 + λx
········································································································································
5
 @ integra land

Solution. Scale out λ: with u = λx,

Z ∞ β −1
−β u π
I (λ) = λ du = λ− β Γ( β)Γ(1 − β) = λ− β .
0 1+u sin(πβ)

Problem 14: Radial heat flow (Hankel diagonalization) [Hankel]

Solve for r > 0, t > 0:

 
1 2
ut = κ urr + ur , u(r, 0) = e−ar ( a > 0).
r

Solution. Apply order-0 Hankel transform: Ut = −κk2 U and

Z ∞
2 1 −k2 /(4a)
U (k, 0) = e− ar J0 (kr ) r dr = e .
0 2a
1 −(κt+1/(4a))k2
Thus U (k, t) = 2a e . Invert using the same Gaussian–Bessel iden-
tity to obtain
ar2
 
1
u(r, t) = exp − .
1 + 4aκt 1 + 4aκt
For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.

Problem 15: A nontrivial Hilbert pair by involution [Hilbert]

 
t
Compute H ( x ).
1 + t2
x
Solution. A standard pair is H{(1 + t2 )−1 }( x ) = . On suitable classes,
1 + x2
t x
H2 = − I. Since is exactly with variable renamed,
1 + t2 1 + x2
 
t 1
H ( x ) = − .
1 + t2 1 + x2

Problem 16: Cusp identity from Fourier differentiation [Fourier]

Show that for ω ∈ R,

Z ∞
1 − cos(ωt)
dt = π |ω |.
−∞ t2
········································································································································
6
 @ integra land

R 1−cos(ωt)
Solution. Let F (ω ) = R t2
dt. Differentiate:

sin(ωt)
Z
′
F (ω ) = dt = π sgn(ω ),
R t
a standard PV/distributional identity. Since F (0) = 0, integrate back to get
F ( ω ) = π | ω |.

Problem 17: Volterra equation with polynomial kernel [Laplace]

Solve, for t ≥ 0, Z t
y(t) − (t − τ ) y(τ ) dτ = t2 .
0

Solution. The integral is a convolution with k (t) = t, so L{k } = 1/s2 . Thus

 
1 2 2 2 1 1
Y (s) 1 − 2 = L{t2 } = 3 ⇒ Y (s) = = − + + .
s s s ( s2 − 1) s s−1 s+1
Therefore y(t) = −2 + et + e−t = 2(cosh t − 1).

Problem 18: STFT ridge for a chirp [STFT]

a 2
Let f (t) = ei(ω0 t+ 2 t ) with real a ̸= 0. For a localized window g, where does
|Vg f ( x, ω )| concentrate?
Solution. The instantaneous frequency is ω (t) = ω0 + at. A window centered
at x sees f approximately as a tone with frequency ω0 + ax. Hence the
dominant ridge satisfies
ω ≈ ω0 + ax.

Problem 19: Vanishing moments annihilate low-degree polynomials [Wavelet]

Assume ψ satisfies tk ψ(t) dt = 0 for k = 0, 1, . . . , m − 1. Show that for any

R

polynomial P with deg P < m, Wψ P( a, b) = 0 for all a ̸= 0, b ∈ R.

−1
Solution. Write P(t) = ∑m k
k =0 ck ( t − b ) . Then

t−b
 
1
Z
Wψ P( a, b) = P(t) p ψ dt.
| a| a

········································································································································
7
 @ integra land

With u = (t − b)/a,
q m −1 Z
Wψ P( a, b) = | a| ∑ ck a k
uk ψ(u) du = 0.
k =0

Problem 20: ROC logic: stable vs causal [Z]

Let
z
X (z) = .
(z − 2)(z − 12 )
List the three possible ROCs and determine whether any choice yields both
causality and BIBO stability.
Solution. Possible ROCs: |z| > 2, 12 < |z| < 2, |z| < 12 . Causality requires
ROC outside the outermost pole: |z| > 2. BIBO stability requires the unit
circle |z| = 1 lie in the ROC, so only 12 < |z| < 2 is stable. No ROC satisfies
both ⇒ no causal stable realization.

Problem 21: A Mellin-type Bessel K evaluation [Mellin]

For ℜ(s) > 0 and γ > 0, evaluate

Z ∞
I (s, γ) = x s−1 e− x−γ/x dx.
0

Solution. Use the standard representation

Z ∞  s/2
γ
x s−1 e− βx−γ/x dx = 2
p

Ks 2 βγ , ℜ(s) > 0, β, γ > 0,
0 β
with β = 1. Hence
√

I (s, γ) = 2 γs/2 Ks 2 γ .

For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.

Problem 22: Inverse Hankel producing a modified Bessel kernel [Hankel]

Show that for a > 0 and r > 0,

Z ∞
k
J0 (kr ) dk = K0 ( ar ).
0 k 2 + a2
········································································································································
8
 @ integra land

Solution. A standard Hankel pair (order 0) is

Z ∞
1
K0 ( ar ) J0 (kr ) r dr = .
0 k2 + a2
R∞
Apply the inversion formula f (r ) = 0 (Hν f )(k ) J0 (kr ) k dk with (Hν f )(k ) =
1
k 2 + a2
to get f (r ) = K0 ( ar ), i.e. the stated identity.

Problem 23: Fourier multiplier solution of Poisson [Fourier]

On R2 , solve −∆u = f in transform form and identify the singular multiplier.

Solution. Let ξ ∈ R2 . Fourier transform gives
F { f }(ξ )
|ξ |2 F {u}(ξ ) = F { f }(ξ ) ⇒ F {u}(ξ ) = .
| ξ |2
Thus u is obtained by applying the singular multiplier |ξ |−2 (distributionally;
the ξ = 0 behavior encodes the harmonic/mean component).

Problem 24: Shifted distributional derivatives under Laplace [Laplace]

Compute L{δ(n) (t − a)}(s) for a > 0 and n ∈ N.

Solution. By distributional integration by parts,
Z ∞
−st (n) dn −st

L{δ (n)
(t − a)}(s) = e δ (t − a) dt = (−1) n e n
= sn e−as .
0 dt t= a

Problem 25: Alternating forcing in a recurrence [Z]

Solve for n ≥ 0:

x [n] − 3x [n − 1] + 2x [n − 2] = (−1)n , x [0] = 0, x [1] = 1,

interpreting x [n] = 0 for n < 0.

Solution. Homogeneous: (r − 1)(r − 2) = 0, so xh [n] = A + B2n . Try
x p [n] = C (−1)n . Substitute:
1
C (−1)n − 3C (−1)n−1 + 2C (−1)n−2 = 6C (−1)n = (−1)n ⇒ C = .
6

········································································································································
9
 @ integra land

1
Use x [0] = 0: A + B + 6 = 0. Use x [1] = 1: A + 2B − 16 = 1. Solve: B = 12 ,
A = − 23 . Hence
2 1 1
x [n] = − + 2n + (−1)n , n ≥ 0.
3 2 6

Problem 26: FrFT composition law (formal) [FrFT]

Assuming the FrFT kernel above, state the group property relating Fα at
angles α and β.
Solution. Quadratic-phase kernels compose by angle addition (rotation in
the time–frequency plane):


Fα F β f = Fα+ β f (mod 2π ),
with Fπ/2 = F and F0 = Id by convention.

Problem 27: Haar two-level DWT of a length-8 signal [Wavelet]

√
Using orthonormal Haar averaging/differencing with factor 1/ 2, compute
two levels of Haar coefficients for
x = [2, 6, 1, 5, 0, 4, 7, 3].

Solution. Level 1:
1 1
a(1) = √ [8, 6, 4, 10], d(1) = √ [−4, −4, −4, 4].
2 2
Level 2 on a(1) :
1 8 − 6 4 − 10
   
(2) 1 8 + 6 4 + 10 (2)
a =√ √ , √ = [7, 7], d =√ √ , √ = [1, −3].
2 2 2 2 2 2
Thus coefficients (coarse-to-fine) are ( a(2) , d(2) , d(1) ):
√ √ √ √
[7, 7 | 1, −3 | −4/ 2, −4/ 2, −4/ 2, 4/ 2].

Problem 28: Sliding-window FFT identity for discrete STFT [STFT↔Fourier]

Let x [n] be a discrete sequence and define a discrete STFT-like map with
window w[n]:
S[m, k ] = ∑ x [n]w[n − m]e−i2πkn/N .
n
········································································································································
10
 @ integra land

Show that for fixed m, S[m, k ] is the length-N DFT of the windowed segment
x [n]w[n − m] (over n = 0, . . . , N − 1 after periodic or zero extension).
Solution. Fix m and define ym [n] = x [n]w[n − m]. Then S[m, k ] = ∑n ym [n]e−i2πkn/N ,
which is exactly the DFT of ym (with the assumed length-N summation rule).
Hence each time-frame STFT column is a DFT, enabling fast computation by
FFT.
For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.

Problem 29: Mellin log-moment by differentiation [Mellin]

For 0 < ℜ(s) < 1, evaluate

Z ∞ s −1
x log x
J (s) = dx.
0 1+x

Solution. Start from

Z ∞ s −1
x π
dx = Γ(s)Γ(1 − s) = .
0 1+x sin(πs)

Differentiate in s (since ∂s x s−1 = x s−1 log x):

 
d π cos(πs)
J (s) = = −π 2 2 = −π 2 cot(πs) csc(πs).
ds sin(πs) sin (πs)

Problem 30: Logarithmic cosine integral [Hilbert/PV]

Show that for a, b > 0,

Z ∞
cos( ax ) − cos(bx )
 
b
dx = ln .
0 x a

Solution. Differentiate with respect to b:

Z ∞ Z ∞
∂ cos( ax ) − cos(bx ) 1
dx = sin(bx ) dx = .
∂b 0 x 0 b
Thus the integral equals ln b + C ( a). Setting b = a gives 0, hence C ( a) =
− ln a.
········································································································································
11
 @ integra land

Problem 31: Radial Fourier reduction in R3 [Hankel↔Fourier]

Let f ( x ) = F (| x |) be radial on R3 . Show that F { f }(ξ ) is radial and can be

written (with ρ = |ξ |)
Z ∞
sin(ρr ) 2
F { f }(ρ) = 4π F (r ) r dr.
0 ρr

Solution. Write ξ along the polar axis and integrate over the sphere:
Z Z ∞Z 1
−iξ · x
F { f }(ξ ) = F (| x |)e dx = 2π F (r )e−iρrµ r2 dµ dr.
R3 0 −1
R1
Since −1 e
−iρrµ dµ = 2 sinρr(ρr) , the stated formula follows.

Problem 32: A Laplace integral producing arctan [Laplace]

Show that for s > 0 and a ∈ R,

Z ∞
−st sin( at )
a
e dt = arctan .
0 t s
Solution. Differentiate in a:
d ∞ −st sin( at)
Z ∞
s
Z
e dt = e−st cos( at) dt = 2 .
da 0 t 0 s + a2
Ra
Since the integral is 0 at a = 0, integrate back: 0 s2 +s u2 du = arctan( a/s).

Problem 33: IIR impulse response and stability [Z]

Given
1 − z −1
H (z) = ,
1 − 34 z−1 + 81 z−2
find the causal impulse response h[n] and decide BIBO stability.
Solution. Factor:
3 1  1  1 
1 − z −1 + z −2 = 1 − z −1 1 − z −1 .
4 8 2 4
Partial fraction:
A B 2 1
H (z) = + , A= , B= .
1 − 12 z−1 1 − 14 z−1 3 3
········································································································································
12
 @ integra land

Thus  n
1 1 n
 
2 1
h[n] = u[n] + u [ n ].
3 2 3 4
Poles at z = 12 , 14 lie inside |z| = 1, so causal ROC |z| > 1
2 contains the unit
circle ⇒ BIBO stable.

Problem 34: Choosing the FrFT angle to focus a chirp [FrFT]

A real chirp has instantaneous frequency ω (t) = ω0 + at. Explain how to

choose α so that Fα concentrates its energy (qualitatively) in the rotated
frequency coordinate.
Solution. In the (t, ω ) plane the chirp traces a line of slope a. FrFT acts as
a rotation by α in time–frequency. Choose α so that the line becomes nearly
vertical (frequency-localized) after rotation; equivalently,

tan α matches the slope scale ≈ a

(up to convention-dependent units). Then the chirp behaves like a narrow-

band tone in the rotated domain.

Problem 35: Soft-threshold wavelet shrinkage step [Wavelet]

Given detail coefficients d = [−0.18, 0.07, 1.10, −0.52, 0.09], apply soft-
thresholding with τ = 0.1: Sτ ( x ) = sgn( x ) max{| x | − τ, 0}.
Solution.
S0.1 (d) = [−0.08, 0, 1.00, −0.42, 0].

For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.

Problem 36: Fourier of a constant (distributional) [Fourier]

Compute F {1}(ω ) in the sense of distributions.

Solution. Since F −1 {δ}(t) = 1
2π under our convention, the identity F {1} =
2π δ(ω ) follows.

········································································································································
13
 @ integra land

Problem 37: Hankel orthogonality kernel (distributional) [Hankel]

State (and justify briefly) the distributional identity

Z ∞
1
Jν (kr ) Jν (kr ′ ) k dk = δ(r − r ′ ), ν > −1.
0 r
Solution. Hankel inversion says
Z ∞Z ∞ 
′ ′ ′ ′
f (r ) = f (r ) Jν (kr ) r dr Jν (kr ) k dk.
0 0
R∞
Comparing with f (r ) = 0 f (r ′ )δ(r − r ′ ) dr ′ yields the kernel identity:
Z ∞
1
Jν (kr ) Jν (kr ′ ) k dk = δ(r − r ′ )
0 r

Problem 38: Multiplicative convolution equation [Mellin]

R∞
Define the multiplicative convolution ( f ⋆ M g)( x ) = 0 f ( x/t) g(t) dtt . Sup-
pose f ⋆ M g = h and M{ g}(s) has no zeros on the line ℜ(s) = c. Express f
in terms of Mellin transforms.
Solution. Mellin convolution diagonalizes:
M{ f ⋆ M g}(s) = M{ f }(s) M{ g}(s).
Hence on ℜ(s) = c,
M{h}(s)
M{ f }(s) = ,
M{ g}(s)
and f is recovered by Mellin inversion:
Z c+i∞
1 M{h}(s)
f (x) = x −s ds.
2πi c−i∞ M{ g}(s)

Problem 39: Moyal identity constant check [STFT]

State the Moyal identity for Vg under the Fourier convention used here, and
deduce the total spectrogram energy.
Solution. With Vg f ( x, ω ) = f (t) g(t − x )e−iωt dt, one has
R
Z ∞ Z ∞
|Vg f ( x, ω )|2 dx dω = 2π ∥ f ∥22 ∥ g∥22 .
−∞ −∞
········································································································································
14
 @ integra land

Thus the spectrogram redistributes energy: its total integral equals 2π times
the product of signal and window energies.

Problem 40: Heat kernel via Fourier in space and Laplace in time [Fourier↔Laplace]

Solve ut = κu xx on R with u( x, 0) = δ( x ) and write u( x, t) explicitly.

Solution. Fourier in x gives ∂t ub(ω, t) = −κω 2 ub(ω, t) with ub(ω, 0) = F {δ}(ω ) =
2
1. So ub(ω, t) = e−κω t . Invert Fourier:
Z ∞
x2
 
1 −κω 2 t iωx 1
u( x, t) = e e dω = √ exp − .
2π −∞ 4πκt 4κt

Problem 41: Window-length tradeoff quantified for a Gaussian [STFT]

2 2
Let gσ (t) = e−t /(2σ ) . Using the Fourier convention above, show that F { gσ }
is Gaussian and identify the frequency width scaling in σ.
Solution. A standard Gaussian transform gives
Z
2 / (2σ2 ) √ σ2 ω 2
F { gσ }(ω ) = e−t e−iωt dt = 2π σ e− 2 .

Thus increasing σ (longer window) narrows the frequency Gaussian with

width ∼ 1/σ, quantifying the STFT time–frequency tradeoff.

Problem 42: A PV transform identity behind Hilbert multipliers [Hilbert↔Fourier]

Show that, in the sense of tempered distributions,

 
1
F PV (ω ) = −iπ sgn(ω ).
t

Solution. The Hilbert kernel is h(t) = π1 PV 1t and satisfies F {H f } =

(−i sgn ω )F { f }. Setting f = δ gives H δ = h, and F {δ} = 1, hence
 
1
F {h}(ω ) = −i sgn(ω ) ⇒ F PV (ω ) = π F {h}(ω ) = −iπ sgn(ω ).
t

For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.
········································································································································
15
 @ integra land

Problem 43: Hankel image of a disk indicator [Hankel]

Compute Hν { f }(k ) (order 0) for f (r ) = 1[0,R] (r ).

Solution. Z R
R
(Hν f )(k) = J0 (kr ) r dr = J (kR).
0 k 1

Problem 44: Impulse-driven oscillator with delay [Laplace]

Solve y′′ (t) + ω 2 y(t) = δ(t − T ) for t ≥ 0 with y(0) = y′ (0) = 0, where
ω > 0, T > 0.
Solution. Laplace: (s2 + ω 2 )Y (s) = e−Ts . Hence

e−Ts sin(ω (t − T ))
Y (s) = 2 ⇒ y(t) = H (t − T ) .
s + ω2 ω

Problem 45: Two-sided Z-inversion from a band ROC [Z]

Let
z z 1
X (z) = − , ROC: < |z| < 2.
z − 21 z − 2 2
Find x [n].
1 z
Solution. In < | z |, corresponds to the right-sided sequence ( 21 )n u[n]. In
2 z− 21
z
|z| < 2, z −2 corresponds to the left-sided sequence −2n u[−n − 1]. Therefore
 n
1
x [n] = u[n] + 2n u[−n − 1].
2

Problem 46: Fourier transform of the Heaviside step [Fourier]

Compute F { H (t)}(ω ) in the sense of tempered distributions.

Solution. Using H ′ = δ and F {δ} = 1, we have iω F { H } = 1 distribution-
ally, hence  
1
F { H }(ω ) = PV + c δ ( ω ).
iω

········································································································································
16
 @ integra land

1
Since H (t) = 2 + 12 sgn(t) and F {1} = 2πδ, one gets c = π. Thus
 
1
F { H (t)}(ω ) = PV + π δ ( ω ).
iω

Problem 47: CWT of a shifted delta [Wavelet]

Let f (t) = δ(t − t0 ) and ψ ∈ L2 (R). Compute Wψ f ( a, b).

Solution. Insert the delta:
t−b t0 − b
   
1 1
Z
Wψ f ( a, b) = δ ( t − t0 ) p ψ dt = p ψ .
| a| a | a| a

Problem 48: Mellin as Fourier on the log-axis [Mellin↔Fourier]

Let f : (0, ∞) → C and fix c ∈ R in the Mellin strip. Define g(y) = ecy f (ey )
on R. Show that
M{ f }(c + iω ) = F { g}(−ω ).

Solution. Set x = ey , dx = ey dy:

Z ∞ Z ∞
c+iω −1
M{ f }(c + iω ) = x f ( x ) dx = e(c+iω )y f (ey ) dy
0 −∞
Z
= g(y)eiωy dy = F { g}(−ω ).

Problem 49: A PV identity from Hilbert quadrature [Hilbert]

For ω0 > 0, evaluate Z ∞

sin(ω0 t)
PV dt.
−∞ x−t
R f (t)
Solution. By definition, PV x −t dt = π (H f )( x ). For ω0 > 0, H{sin(ω0 t)}( x ) =
− cos(ω0 x ), hence
Z ∞
sin(ω0 t)
PV dt = −π cos(ω0 x ).
−∞ x−t

For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.
········································································································································
17
 @ integra land

Problem 50: STFT modulation rule [STFT]

Show that for any real ω0 ,

Vg {eiω0 t f (t)}( x, ω ) = Vg f ( x, ω − ω0 ).

Solution. Z
iω0 ·
Vg (e f )( x, ω ) = f (t)eiω0 t g(t − x )e−iωt dt
Z
= f (t) g(t − x )e−i(ω −ω0 )t dt = Vg f ( x, ω − ω0 ).

Problem 51: Fourier transform of a Gaussian (parameterized) [Fourier]

2
For a > 0, compute F {e− at }(ω ).
Solution. A standard completion-of-squares gives
Z ∞
ω2
r  
− at2 − at2 −iωt π
F {e }(ω ) = e e dt = exp − .
−∞ a 4a

Problem 52: Inverse Laplace of a logarithmic symbol [Laplace]

 
ln s
Find f (t) = L −1 (t) for t > 0 (Euler constant γ allowed).
s2
d
Solution. Use L{ln t}(s) = −(γ + ln s)/s and the rule L{t h(t)} = − ds L{h(t)}.
One computes
1 − γ − ln s
L{t ln t}(s) = .
s2
Hence
ln s 1−γ

= − L{ t ln t } ⇒ f ( t ) = ( 1 − γ ) t − t ln t = t 1 − γ − ln t .
s2 s2

Problem 53: A Z-transform with a double pole and a difference operator [Z]

Let a ∈ (0, 1) and

1 − z −1
X (z) = , ROC: |z| > | a|.
(1 − az−1 )2
Find x [n] for n ∈ Z.
········································································································································
18
 @ integra land

Solution. Let Y (z) = (1 − az−1 )−2 , so y[n] = (n + 1) an u[n]. Multiplication

by (1 − z−1 ) is the first difference: x [n] = y[n] − y[n − 1]. Thus x [0] = 1, and
for n ≥ 1,

x [n] = (n + 1) an − nan−1 = an−1 (n + 1) a − n ,

and x [n] = 0 for n < 0.

Problem 54: 2D modified Helmholtz Green’s function [Hankel/Fourier]

On R2 , find the radial fundamental solution G (r ) satisfying

(−∆ + µ2 ) G = δ (µ > 0),

in the sense of distributions.
1
Solution. Taking the 2D Fourier transform gives F2 { G }(ξ ) = | ξ |2 + µ2
. The
inverse transform is radial and is known to be
1
G (r ) = K0 (µr ), r > 0,
2π
where K0 is the modified Bessel function.

Problem 55: A sharp Mellin–Beta evaluation [Mellin]

For 0 < ℜ(s) < 2, evaluate

Z ∞
x s −1
I (s) = dx.
0 (1 + x )2

Solution. This is B(s, 2 − s):

π
I ( s ) = Γ ( s ) Γ (2 − s ) = (1 − s ) Γ ( s ) Γ (1 − s ) = (1 − s ) .
sin(πs)

Problem 56: FrFT of a delta derivative [FrFT]

For α ̸≡ kπ and t0 ∈ R, compute Fα {δ′ (t − t0 )}(u).

i 2
+ u2 ) cot α − iut csc α . Distributionally,


Solution. Let Kα (t, u) = exp 2 (t
Z
′
Fα {δ (· − t0 )}(u) = δ′ (t − t0 )Kα (t, u) dt = −∂t Kα (t, u) t = t0
.
········································································································································
19
 @ integra land

Since ∂t Kα = (it cot α − iu csc α)Kα , we get

 
′ i 2 2


Fα {δ (t − t0 )}(u) = i u csc α − t0 cot α exp 2 ( t0 + u ) cot α − iut0 csc α .

For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.

Problem 57: Hilbert transform commutes with differentiation [Hilbert]

Assume f is sufficiently nice (e.g. Schwartz). Show that (H f ′ ) = (H f )′ .

Solution. Fourier multiplier form gives

F {H f ′ } = (−i sgn ω ) F { f ′ } = (−i sgn ω ) (iω )F { f }

= (iω ) (−i sgn ω )F { f } = F {(H f )′ }.
Invert F to conclude H f ′ = (H f )′ .

Problem 58: Derivative-of-Gaussian wavelet spectrum [Wavelet]

d −t2 /2
2 R∞
Let ψ(t) = e = −te−t /2 . (i) Show −∞ ψ(t) dt = 0. (ii) Compute
dt
F {ψ}(ω ).
2
Solution. (i) ψ = (e−t /2 )′ dt = 0. (ii) Use F { f ′ } = iω F { f } and
R R
2 √ 2
F {e−t /2 } = 2π e−ω /2 :
√ 2
F {ψ}(ω ) = iω 2π e−ω /2 .

Problem 59: STFT of a shifted delta [STFT]

Let f (t) = δ(t − t0 ). Compute Vg f ( x, ω ).

Solution.
Z
Vg f ( x, ω ) = δ(t − t0 ) g(t − x )e−iωt dt = g(t0 − x ) e−iωt0 .

Problem 60: Fourier transform of sgn(t) [Fourier]

Compute F {sgn(t)}(ω ) in the sense of distributions.

········································································································································
20
 @ integra land

Solution. Since (sgn)′ = 2δ and F {δ} = 1, we have iω F {sgn} = 2. Oddness

excludes a δ(ω ) term, hence
 
1
F {sgn(t)}(ω ) = 2 PV .
iω

Problem 61: Z-domain “finite difference” shaping [Z]

Let y[n] = (n + 1) an u[n] with | a| < 1, and define x [n] = y[n] − y[n − 1] (with
y[−1] = 0). Find X (z) = Z { x [n]} for the causal ROC.
Solution. Y (z) = ∑n≥0 (n + 1) an z−n = 1
(1− az−1 )2
for |z| > | a|. Then x [n] =
y[n] − y[n − 1] corresponds to multiplication by (1 − z−1 ):

1 − z −1
X ( z ) = ( 1 − z − 1 )Y ( z ) = , ROC: |z| > | a|.
(1 − az−1 )2

Problem 62: A log-oscillatory Mellin integral [Mellin]

Let 0 < c < 1 and τ ∈ R. Evaluate

Z ∞ c −1
x cos(τ ln x )
I (c, τ ) = dx.
0 1+x

Solution. Write cos(τ ln x ) = ℜ( xiτ ) so

Z ∞ (c+iτ )−1  
x π
I (c, τ ) = ℜ dx = ℜ .
0 1+x sin(π (c + iτ ))
Using sin(π (c + iτ )) = sin(πc) cosh(πτ ) + i cos(πc) sinh(πτ ),
sin(πc) cosh(πτ )
I (c, τ ) = π .
sin2 (πc) + sinh2 (πτ )

Problem 63: 2D radial Fourier reduces to Hankel0 [Hankel↔Fourier]

Let f ( x ) = F (| x |) be radial on R2 . Show that the 2D Fourier transform is

radial and satisfies
Z ∞
F2 { f }(ρ) = 2π F (r ) J0 (ρr ) r dr, ρ = | ξ |.
0
········································································································································
21
 @ integra land

Solution. Align ξ with the polar axis and integrate over angle:
Z ∞ Z 2π Z ∞
−iρr cos θ
F2 { f }(ξ ) = F (r ) e r dθ dr = 2π F (r ) J0 (ρr ) r dr,
0 0 0
R 2π
using 0 e−iρr cos θ dθ = 2π J0 (ρr ).
For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.

Problem 64: Analytic signal of a cosine [Hilbert]

Let f (t) = cos(ω0 t) with ω0 > 0. Compute the analytic signal f a (t) =
f ( t ) + i H f ( t ).
Solution. For ω0 > 0, H{cos(ω0 t)} = sin(ω0 t), hence

f a (t) = cos(ω0 t) + i sin(ω0 t) = eiω0 t .

Problem 65: Final value theorem check [Laplace]

s+2
Let F (s) = and f (t) = L−1 { F (s)}. Compute limt→∞ f (t).
s(s + 1)(s + 3)
Solution. All poles of sF (s) are at −1, −3 (left half-plane), so the final value
theorem applies:
s+2 2
lim f (t) = lim sF (s) = lim = .
t→∞ s →0 s→0 ( s + 1)( s + 3) 3

Problem 66: A half-line cosine integral [Fourier]

Evaluate, for a ∈ R, Z ∞
cos( ax )
dx.
0 1 + x2
R∞ e−iax
Solution. Since − ∞ 1+ x 2 dx = πe−|a| , taking the real part and halving gives
Z ∞
cos( ax ) π −|a|
dx = e .
0 1 + x2 2

········································································································································
22
 @ integra land

Problem 67: Inverse system and stability [Z]

Consider the discrete-time transfer function

1 − 0.8z−1
H (z) = .
1 − 1.25z−1
(i) Find an inverse transfer function G (z) = 1/H (z). (ii) Decide whether G
can be causal and BIBO stable.
1 − 1.25z−1
Solution. (i) G (z) = . (ii) G has a pole at z = 0.8 (inside the unit
1 − 0.8z−1
circle), so the causal ROC |z| > 0.8 contains |z| = 1. Hence G admits a causal,
BIBO-stable realization.

Problem 68: STFT of a two-tone signal [STFT]

Let f (t) = eiω1 t + eiω2 t with ω1 ̸= ω2 . Express Vg f ( x, ω ) in terms of F { g}

and describe the interference term in |Vg f |2 .
Solution. By linearity and the tone formula,

Vg f ( x, ω ) = e−i(ω −ω1 ) x F g(ω − ω1 ) + e−i(ω −ω2 ) x F g(ω − ω2 ).

Thus
|Vg f |2 = |F g(ω − ω1 )|2 + |F g(ω − ω2 )|2
 
+2ℜ e−i(ω2 −ω1 )x F g(ω − ω1 )F g(ω − ω2 ) ,
where the last term is the oscillatory cross-term (time-dependent interference).

Problem 69: FrFT of the origin delta [FrFT]

For α ̸≡ kπ, compute Fα {δ(t)}(u) and its modulus.

Solution. Insert δ(t) into the kernel:
 
i 2
Fα {δ}(u) = exp u cot α .
2
Therefore |Fα {δ}(u)| = 1 for all u.

········································································································································
23
 @ integra land

Problem 70: Exponential–Bessel moment [Hankel]

For a > 0 and b ≥ 0, evaluate

Z ∞
e− ar J0 (br ) r dr.
0

R∞ 1
Solution. Use the standard identity 0 e−ar J0 (br ) dr = √ and differ-
a2 + b2
entiate in a:
Z ∞  
d 1 a
− re− ar J0 (br ) dr = √ =− .
0 da a2 + b2 ( a2 + b2 )3/2
Hence Z ∞
a
e− ar J0 (br ) r dr = .
0 ( a2 + b2 )3/2
For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.

Problem 71: Mellin transform of a log-Gaussian [Mellin]

Let σ > 0 and define f ( x ) = exp − (ln x )2 /(2σ2 ) on (0, ∞). Compute

M{ f }(s) for s ∈ C.
Solution. Set x = ey , dx = ey dy:
Z ∞
s−1 −(ln x )2 /(2σ2 )
Z ∞
2 / (2σ2 ) √ σ 2 s2
M{ f }(s) = x e dx = esy e−y dy = 2π σ e 2 .
0 −∞

Problem 72: Bedrosian-type quadrature rule [Hilbert]

Assume f ∈ L2 (R) has Fourier support in |ω | < Ω, and let ω0 > Ω. Show
that
H{ f (t) cos(ω0 t)} = f (t) sin(ω0 t), H{ f (t) sin(ω0 t)} = − f (t) cos(ω0 t).

Solution. In frequency, multiplication by cos(ω0 t) shifts F { f } to bands near

±ω0 with no overlap around 0. On these separated bands, the multiplier
−i sgn(ω ) acts as a constant phase ∓i on the ±ω0 bands. This exactly converts
cosine carriers to sine carriers (and vice versa with a sign), yielding the stated
identities.
········································································································································
24
 @ integra land

Problem 73: CWT energy identity (Plancherel form) [Wavelet]

R∞
Assume ψ is admissible with Cψ = 0 ω < ∞. State the energy
|F ψ(ω )|2 dω
identity relating ∥ f ∥22 to Wψ f .
Solution. The CWT satisfies
Z ∞ Z ∞
|Wψ f ( a, b)|2
2
da db = Cψ ∥ f ∥22 .
−∞ 0 a

Problem 74: Laplace–Fourier bridge on a vertical line [Laplace↔Fourier]

Let f be causal ( f (t) = 0 for t < 0) and assume e−σt f (t) ∈ L1 (0, ∞) for some
σ > 0. Show that the Fourier transform of e−σt f (t) H (t) equals L{ f }(σ + iω ).
Solution. By definitions,
Z ∞ Z ∞
−σt −σt −iωt
F {e f (t) H (t)}(ω ) = f (t)e e dt = f (t)e−(σ+iω )t dt = L{ f }(σ + iω ).
0 0

Problem 75: Z-identity for n an [Z]

For | a| < 1, compute Z {nan u[n]}(z) and state a general differentiation iden-
tity connecting Z {nx [n]} to X (z).
Solution. For X (z) = Z { x [n]}, one has the identity
d
Z {nx [n]}(z) = − z
X ( z ),
dz
(valid when the differentiated series converges in the ROC). For x [n] = an u[n],
1
X (z) = 1− az −1 (ROC | z | > | a |), hence

az−1
 
n d 1
Z {na u[n]}(z) = −z = .
dz 1 − az−1 (1 − az−1 )2

Problem 76: Hilbert as a singular convolution operator [Hilbert↔Fourier]

1
Show that H f = f ∗ h with h(t) = π PV 1t and explain (via Fourier multipliers)
why H is a ±90◦ phase shifter.
Solution. By definition,
1 f (t) 1 1
Z Z
(H f )( x ) = PV dt = f (t) h( x − t) dt = ( f ∗ h)( x ), h(t) = PV .
π x−t π t
········································································································································
25
 @ integra land

Taking Fourier transforms gives F {h}(ω ) = −i sgn(ω ), hence

F {H f }(ω ) = F { f }(ω ) F {h}(ω ) = (−i sgn ω )F { f }(ω ),
which multiplies positive frequencies by −i and negative frequencies by +i,
i.e. a ∓90◦ phase rotation.

Problem 77: Laplace solution with step forcing [Laplace]

Solve y′′ (t) + y(t) = H (t − a) for t ≥ 0 with y(0) = y′ (0) = 0 and a > 0.
Solution. Laplace: (s2 + 1)Y (s) = L{ H (t − a)} = e− as /s. Thus
e−as
 
1 s
Y (s) = = e−as − .
s ( s2 + 1) s s2 + 1
Invert and apply the delay:


y(t) = H (t − a) 1 − cos(t − a) .

For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.

Problem 78: Exponential cusp in frequency space [Fourier]

Let a > 0 and f (t) = e− a|t| . Compute F { f }(ω ) and deduce

Z ∞
e− at cos(ωt) dt.
0

Solution.
Z ∞ Z ∞
− a|t| − a|t| −iωt 2a
F {e }(ω ) = e e dt = 2 e− at cos(ωt) dt = .
−∞ 0 a2 + ω 2
Z ∞
a
Hence e− at cos(ωt) dt = .
0 a2 + ω 2

Problem 79: A square-root Laplace symbol [Laplace]

For c > 0, compute ( √ )

e−c s
L −1 √ ( t ), t > 0.
s
········································································································································
26
 @ integra land

Solution. Use the classical identity

Z ∞ 2
√
e−c /(4t) e − c s
e−st √ dt = √ , ℜs > 0,
0 πt s
so ( √ )
e − c s 1 2
L −1 √ (t) = √ e−c /(4t) .
s πt

Problem 80: Causal but unstable from the ROC [Z]

Let
1 − 21 z−1
X (z) = , ROC: |z| > 1.2.
(1 − 0.8z−1 )(1 − 1.2z−1 )
Find x [n] and decide whether x ∈ ℓ1 .
Solution. With q = z−1 , partial fractions give
− 34 7
4
X (z) = + .
1 − 0.8z−1 1 − 1.2z−1
Since ROC is exterior, both terms are right-sided:
3 7
x [n] = − (0.8)n u[n] + (1.2)n u[n].
4 4
Because (1.2)n grows, x ∈
/ ℓ1 (not BIBO-stable as an impulse response).

Problem 81: Order-1 Hankel image of a decaying radial [Hankel]

For a > 0, compute the order-1 Hankel transform

Z ∞
(H1 f )(k) = f (r ) J1 (kr ) r dr for f (r ) = e− ar .
0
R∞
Solution. Differentiate 0 e− ar J0 (kr ) dr = √ 21 2 in k:
a +k
  Z ∞ Z ∞
d 1 − ar d
√ = e J0 (kr ) dr = − e− ar rJ1 (kr ) dr.
dk a2 + k 2 0 dk 0

Thus Z ∞
− ar k
(H1 e )(k) = e− ar rJ1 (kr ) dr = .
0 ( a2 + k2 )3/2
········································································································································
27
 @ integra land

Problem 82: A Mellin–Beta evaluation with higher power [Mellin]

For 0 < ℜ(s) < 3, evaluate

Z ∞
x s −1
I (s) = dx.
0 (1 + x )3

Solution. This is a Beta integral:

Γ ( s ) Γ (3 − s ) 1 π (s − 1)(s − 2)
I (s) = B(s, 3 − s) = = Γ ( s ) Γ (3 − s ) = .
Γ (3) 2 2 sin(πs)

Problem 83: A cosine principal value integral [Hilbert]

Let ω0 > 0. Evaluate Z ∞

cos(ω0 t)
PV dt.
−∞ x−t
R f (t)
Solution. By definition, PV x −t dt = π (H f )( x ). Since H{cos(ω0 t)}( x ) =
sin(ω0 x ) for ω0 > 0, we get
Z ∞
cos(ω0 t)
PV dt = π sin(ω0 x ).
−∞ x−t

Problem 84: Haar multiresolution reveals a hidden jump [Wavelet]

Let x ∈ R8 be
x = [0, 1, 2, 3, 10, 11, 12, 13].
Using
√ the orthonormal Haar DWT (pairwise averages/details with factor
1/ 2), compute level-1 details d , level-2 details d(2) , and level-3 detail d(3) .
( 1 )

Identify which scale captures the jump.

Solution. Level 1:
√
1 2
d(1) = √ [0 − 1, 2 − 3, 10 − 11, 12 − 13] = − [1, 1, 1, 1],
2 2
√
1 2
a(1) = √ [0 + 1, 2 + 3, 10 + 11, 12 + 13] = [1, 5, 21, 25].
2 2

········································································································································
28
 @ integra land

Level 2 (apply Haar to a(1) ):

1 √ √ √ √
d(2) = √ [1/ 2 − 5/ 2, 21/ 2 − 25/ 2] = [−2, −2],
2
1 √ √ √ √
a(2) = √ [1/ 2 + 5/ 2, 21/ 2 + 25/ 2] = [3, 23].
2
Level 3 (apply Haar to a(2) ):

(3) 3 − 23 √ (3) 3 + 23 √
d = √ = −10 2, a = √ = 13 2.
2 2
The large coefficient d(3) captures the coarse-scale jump between the first four
and last four samples.
For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.

Problem 85: FrFT becomes a scaled Fourier transform after dechirping [FrFT↔Fourier]

i 2 cot α
Fix α ̸≡ kπ and define f (t) = e− 2 t g(t). Show that
i 2 cot α
(Fα f )(u) = e 2 u F { g}(u csc α).

Solution. Insert the definition of Fα :

Z Z
i 2 2 csc α − 2i t2 cot α i 2
(Fα f )(u) = e 2 ( t + u ) cot α −iut e g(t) dt = e 2 u cot α g(t)e−i(u csc α)t dt,
i 2 cot α
which is e 2 u F { g}(u csc α).

Problem 86: A Mellin oscillatory moment [Mellin]

For 0 < ℜ(s) < 1, evaluate

Z ∞
x s−1 sin x dx.
0
R∞
Solution. Use 0 x s−1 eix dx = eiπs/2 Γ(s) in the principal branch and take
imaginary parts: Z ∞  πs 
s −1
x sin x dx = Γ(s) sin .
0 2
········································································································································
29
 @ integra land

Problem 87: Deconvolving a step through a stable IIR [Z]

 n
1
Let h[n] = 2 u[n] and y[n] = u[n]. If y = x ∗ h, find x [n] (assume a causal
ROC).
Solution.
1 1
H (z) = , Y (z) = , ROC: |z| > 1.
1 − 12 z−1 1 − z −1
Thus
Y (z) 1 − 12 z−1 1 −1
2z
X (z) = = = 1 + .
H (z) 1 − z −1 1 − z −1
Invert:
1
x [n] = δ[n] + u [ n − 1].
2

Problem 88: Integro-differential equation by Laplace [Laplace]

Solve for t ≥ 0:
Z t
′
y (t) + y(τ ) dτ = e−t , y(0) = 0.
0

Solution. Taking Laplace transforms:

 
1 1 1 1
(sY (s) − y(0)) + Y (s) = ⇒ Y (s) s + = .
s s+1 s s+1
Hence
s s+1 1
Y (s) = = − .
(s + 1)(s2 + 1) 2( s2 + 1) 2( s + 1)
Invert:
1
1
y(t) = cos t + sin t − e−t .
2 2

Problem 89: Hilbert transform of a pure tone [Hilbert]

Let ω0 ̸= 0 and f (t) = eiω0 t . Compute H f .

Solution. In frequency, F {eiω0 t } = 2π δ(ω − ω0 ), so
F {H f }(ω ) = (−i sgn ω ) 2πδ(ω − ω0 ) = −i sgn(ω0 ) 2πδ(ω − ω0 ).
Invert:
H{eiω0 t } = −i sgn(ω0 ) eiω0 t .
········································································································································
30
 @ integra land

Problem 90: A convolution equation with an explicit closed form [Fourier]

Solve for f ∈ L2 (R):

Z ∞
1
f (t) + e−|t−τ | f (τ ) dτ = e−|t| .
2 −∞

Solution. Let k (t) = 12 e−|t| . Then F {e−|t| }(ω ) = 1+2ω2 and F {k }(ω ) = 1
1+ ω 2
.
Fourier transforming:
 1  2 2
F (ω ) 1 + = ⇒ F ( ω ) = .
1 + ω2 1 + ω2 ω2 + 2
√
Match with F {e−a|t| } = a2 + 2a
ω2
using a = 2:
√ √
1 2 2 1
F (ω ) = √ · 2 ⇒ f ( t ) = √ e − 2| t | .
2 ω +2 2

Problem 91: STFT covariance under time shifts [STFT]

Let f t0 (t) = f (t − t0 ). Show that

Vg f t0 ( x, ω ) = e−iωt0 Vg f ( x − t0 , ω ).

Solution. Z
Vg f t0 ( x, ω ) = f (t − t0 ) g(t − x )e−iωt dt.

With u = t − t0 :
Z
−iωt0
Vg f t0 ( x, ω ) = e f (u) g(u − ( x − t0 ))e−iωu du = e−iωt0 Vg f ( x − t0 , ω ).

For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.

Problem 92: A damped Dirichlet integral [Hilbert/Fourier]

For a ∈ R and b > 0, evaluate

Z ∞
sin( at)
PV e−b|t| dt.
−∞ t

········································································································································
31
 @ integra land

Solution. The integrand is even, so PV is not needed after symmetrization:

Z ∞ Z ∞
−b|t| sin( at ) sin( at)
PV e dt = 2 e−bt dt.
−∞ t 0 t
Differentiate in a:
d ∞ −bt sin( at)
Z ∞
b
Z
e dt = e−bt cos( at) dt = 2 .
da 0 t 0 a + b2
Integrate back with value 0 at a = 0:
Z ∞
−bt sin( at )
a
e dt = arctan ,
0 t b
hence the full integral is 2 arctan( a/b).

Problem 93: A logarithmic integral via Laplace differentiation [Laplace]

For s > 0 and b ∈ R, evaluate

Z ∞
1 − cos(bt)
I= e−st dt.
0 t
Solution. Differentiate in b:
Z ∞
dI b
= e−st sin(bt) dt = 2 .
db 0 s + b2
Integrate b from 0 to b and use I (0) = 0:
b2
Z b  
β 1
I= dβ = ln 1 + 2 .
0 s2 + β2 2 s

Problem 94: Left-sided sequence from an interior ROC [Z]

Let | a| > 0 and

1
X (z) = , ROC: |z| < | a|.
1 − az−1
Find x [n].
Solution. For |z| < | a|,
∞  m −1
1 z/a z
1 − az − 1
= −
1 − (z/a)
= − ∑ a
= − ∑ an z−n .
m =1 n=−∞
Thus
x [n] = − an u[−n − 1].
········································································································································
32
 @ integra land

Problem 95: Laplace-in-k produces a radial kernel [Hankel]

For c > 0 and r ≥ 0, evaluate

Z ∞
e−ck J0 (rk ) k dk.
0
R∞
Solution. Use 0 e−ck J0 (rk ) dk = √ 1 and differentiate in c:
c2 +r 2
Z ∞  
−ck d 1 c
− ke J0 (rk ) dk = √ =− .
0 dc c2 + r 2 (c2 + r2 )3/2
Hence Z ∞
c
e−ck J0 (rk ) k dk = .
0 (c2 + r2 )3/2

Problem 96: Mellin log-moment of a rational profile [Mellin]

For p > 0 and 0 < ℜ(s) < p, evaluate

Z ∞ s −1
x ln x
dx.
1 + xp
0

R∞ x s −1
 
Solution. First, 0 1+ x p
π
dx = p csc πs
p . Differentiate in s under the integral
sign:
Z ∞ s −1
π2
      
x ln x d π πs πs πs
dx = csc = − 2 csc cot .
0 1 + xp ds p p p p p

Problem 97: CWT of a step function [Wavelet]

Let f (t) = H (t) and ψ ∈ L2 (R). Show that for a ̸= 0,

q Z ∞
Wψ f ( a, b) = | a| ψ(u) du.
−b/a

Deduce the small-scale behavior at b = 0.

Solution. Z ∞
t−b
 
1
Wψ f ( a, b) = p ψ dt.
0 | a| a

········································································································································
33
 @ integra land

Set u = (t − b)/a so dt = a du:

q Z ∞
Wψ f ( a, b) = | a| ψ(u) du.
−b/a
p R∞ p
At b = 0, WRψ f ( a, 0) = | a| 0 ψ(u) du, hence |Wψ f ( a, 0)| ∼ C | a| as a → 0
∞
(with C = 0 ψ ).

Problem 98: A chirp whose FrFT collapses to a delta [FrFT]

Fix α ̸≡ kπ and let f (t) = exp − 2i t2 cot α . Compute Fα f in the sense of

distributions.
Solution. Insert into the kernel:
Z Z
i 2 2 csc α − 2i t2 cot α i 2
(Fα f )(u) = e 2 ( t + u ) cot α −iut e dt = e 2 u cot α e−i(u csc α)t dt.

e−iβt dt = 2πδ( β), so

R
Distributionally
i 2 cot α
(Fα f )(u) = e 2 u 2π δ(u csc α) = 2π | sin α| δ(u).

For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.

Problem 99: Sinc-squared integral by Parseval [Fourier]

Evaluate
sin ω 2
Z ∞  
dω.
−∞ ω

Solution. Let f (t) = 1[−1,1] (t). Then

Z 1
2 sin ω
F (ω ) = F { f }(ω ) = e−iωt dt = .
−1 ω
1
| f |2 dt = 2π | F |2 dω, i.e.
R R
Parseval gives

sin ω 2 sin ω 2
Z ∞   Z ∞  
1
2= 4 dω ⇒ dω = π.
2π −∞ ω −∞ ω

········································································································································
34
 @ integra land

Problem 100: Laplace transform of Bessel oscillations [Laplace]

For a > 0 and ℜ(s) > 0, compute L{ J0 ( at)}(s).

Solution. A standard Laplace–Bessel identity gives
Z ∞
1
e−st J0 ( at) dt = √ .
0 s2 + a2
Hence L{ J0 ( at)}(s) = √ 1 .
s2 + a2

Problem 101: Step response of a first-order system [Z]

Let a causal LTI system have transfer function

1 + z −1
H (z) = , ROC: |z| > 12 .
1 − 21 z−1
Find the output y[n] to the input x [n] = u[n].
Solution.
1 1 + z −1
X (z) = , Y (z) = H (z) X (z) = .
1 − z −1 (1 − 12 z−1 )(1 − z−1 )
Partial fraction (in z−1 ) yields
4 3
Y (z) = − .
1 − z−1 1 − 12 z−1
Invert (right-sided by ROC):
 n
1
y[n] = 4u[n] − 3 2 u [ n ].

Problem 102: A scaled Hilbert pair [Hilbert]

n o
1
For a > 0, compute H t2 + a2
( x ).
x
Solution. Using the known pair H{ 1+1t2 }( x ) = 1+ x 2
and scaling 1
t2 + a2
=
1 1
a2 1+(t/a)2
, we get
 
1 1 x/a x
H (x) = = .
t2 + a2 a2 1 + ( x/a)2 a ( x 2 + a2 )
········································································································································
35
 @ integra land

Problem 103: Gamma values from log-frequency modulation [Mellin]

Let c > 0, µ > 0, and τ ∈ R. Evaluate

Z ∞
I= x c−1 e−µx cos(τ ln x ) dx.
0

Solution. Write cos(τ ln x ) = ℜ( xiτ ). Then

Z ∞
x c+iτ −1 e−µx dx = ℜ µ−(c+iτ ) Γ(c + iτ ) .


I=ℜ
0

Problem 104: Differentiating the K0 Hankel kernel [Hankel]

For a > 0 and r > 0, show that

Z ∞
k2
J1 (kr ) dk = aK1 ( ar ),
0 k 2 + a2
where K1 is the modified Bessel function.
Solution. Start from the standard identity
Z ∞
k
J0 (kr ) dk = K0 ( ar ), r > 0.
0 k 2 + a2
d d
Differentiate in r and use dr J0 (kr ) = −kJ1 (kr ) and dr K0 ( ar ) = − aK1 ( ar ):
Z ∞ Z ∞
k2

d k
J0 (kr ) dk = − J1 (kr ) dk = − aK1 ( ar ).
dr 0 k2 + a2 0 k 2 + a2
Multiply by −1 to obtain the claim.

Problem 105: A spectrogram marginal identity [STFT]

Show that for any f , g ∈ L2 (R) and any fixed x ∈ R,

Z ∞ Z ∞
2
Vg f ( x, ω ) dω = 2π | f (t)|2 | g(t − x )|2 dt.
−∞ −∞

Solution. For fixed x, Vg f ( x, ω ) = F {h x }(ω ) where h x (t) = f (t) g(t − x ). By

Plancherel (with our Fourier convention),
Z Z Z Z
2 2 2
|Vg f ( x, ω )| dω = |F h x (ω )| dω = 2π |h x (t)| dt = 2π | f (t)|2 | g(t − x )|2 dt.

For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.
········································································································································
36
 @ integra land

Problem 106: Two-sided stable but noncausal sequence [Z]

Let
x [n] = 0.9n u[n] + 1.1n u[−n − 1].
Compute X (z), its ROC, and decide whether x can be causal and whether
x ∈ ℓ1 .
Solution. Right-sided part:
1
∑ 0.9n z−n = 1 − 0.9z−1 , |z| > 0.9.
n ≥0

Left-sided part:
 z m z 1
∑ 1.1 zn −n
= ∑
1.1
=
1.1 − z
=−
1 − 1.1z −1
, |z| < 1.1.
n≤−1 m ≥1

Thus
1 1
X (z) = −
− , ROC: 0.9 < |z| < 1.1.
1 − 0.9z 1 1 − 1.1z−1
The ROC includes the unit circle, so x ∈ ℓ1 (absolutely summable), but the
sequence is two-sided and cannot be causal.

Problem 107: A Mellin–Laplace hybrid with log-oscillation [Mellin]

For c > 0, τ ∈ R, and µ > 0, evaluate

Z ∞
x c−1 e−µx sin(τ ln x ) dx.
0

Solution. Use sin(τ ln x ) = ℑ( xiτ ):

Z ∞ Z ∞
c−1 −µx
x c+iτ −1 e−µx dx = ℑ µ−(c+iτ ) Γ(c + iτ ) .


x e sin(τ ln x ) dx = ℑ
0 0

Problem 108: The analytic projection operator [Hilbert↔Fourier]

1


Define P+ f = 2 f − i H f on Schwartz functions. Show that
1
F { P+ f }(ω ) = 1{ω >0} F { f }(ω ) + F { f }(0) δ(ω )
2
········································································································································
37
 @ integra land

in the distributional sense (interpret the ω = 0 contribution appropriately).

Solution. Using F {H f } = (−i sgn ω )F { f },
1 
F { P+ f } = 1 − sgn ω F { f }.
2
Since 12 (1 − sgn ω ) = 1{ω >0} away from ω = 0, this is exactly the positive-
frequency projector, with the usual distributional convention assigning half-
weight at ω = 0.

Problem 109: Wavelet scaling at a jump discontinuity [Wavelet]

Let f (t) = H (t − t0 ) and suppose ψ is such that R ψ(t) dt = 0 and ψ ∈ L1 (R).

R

Show that q Z ∞
Wψ f ( a, t0 ) = | a| ψ(u) du,
0
p
and conclude |Wψ f ( a, t0 )| = O( | a|) as a → 0.
Solution. From the step formula (change of variables),
q Z ∞
Wψ f ( a, b) = | a| ψ(u) du.
−(b−t0 )/a

At b = t0 this becomes
q Z ∞
Wψ f ( a, t0 ) = | a| ψ(u) du,
0
p
which is a constant times | a |.

Problem 110: Inverse Laplace with a square-root pole [Laplace]

For a > 0, compute

 
−1 1
L √ ( t ), t > 0.
s+a
R∞ √
Solution. Use 0 t−1/2 e−(s+ a)t dt = Γ( 12 )(s + a)−1/2 = π (s + a)−1/2 . Thus
 
−1 √ 1 1
L (t) = √ e−at .
s+a πt

········································································································································
38
 @ integra land

Problem 111: A Plancherel evaluation of a rational integral [Fourier]

For a > 0 and b > 0, evaluate

Z ∞
1
dt.
−∞ (t2 + a2 )(t2 + b2 )

Solution. Let f a (t) = t2 +1 a2 and f b (t) = 1

t2 + b2
. Known Fourier transforms
(with our convention) are
π − a|ω | π −b|ω |
F { f a }(ω ) = e , F { f b }(ω ) = e .
a b
By Parseval,
Z ∞
1 1 π2
Z Z
f a (t) f b (t) dt = F { f a }(ω ) F { f b }(ω ) dω = e−(a+b)|ω | dω.
2π 2π ab −∞
R∞
Since −∞ e−(a+b)|ω | dω = a+2 b , we get
Z ∞
1 π
dt = .
−∞ (t2 + a2 )(t2 + b2 ) ab( a + b)

Problem 112: A cylindrical Green kernel integral [Hankel]

For a > 0 and r, R > 0, evaluate

Z ∞
k
J0 (kr ) J0 (kR) dk
0 k 2 + a2
in terms of modified Bessel functions. (Use r< := min{r, R} and r> :=
max{r, R}.)
Solution. This is the standard axisymmetric Green kernel for (−∆ + a2 ) in
cylindrical coordinates:
Z ∞
k
J0 (kr ) J0 (kR) dk = I0 ( ar< ) K0 ( ar> ),
0 k 2 + a2
where I0 and K0 are modified Bessel functions.
For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.

········································································································································
39
 @ integra land

Problem 113: Damped sinusoid in the z-plane [Z]

Let x [n] = r n sin(Ωn) u[n] with 0 < r < 1 and Ω ∈ (0, π ). Find X (z) and its
ROC.
Solution. With q = z−1 ,
1 1
∑ rn eiΩn qn = 1 − reiΩ q , ∑ rn e−iΩn qn = 1 − re−iΩ q .
n ≥0 n ≥0

Thus
ne
iΩn − e−iΩn n r sin Ω q r sin Ω z−1
X (z) = ∑r 2i
q =
1 − 2r cos Ω q + r2 q2
=
1 − 2r cos Ω z−1 + r2 z−2
,
n ≥0

with ROC |z| > r (right-sided).

Problem 114: A half-order Laplace–oscillatory integral [Laplace]

For s > 0 and b ∈ R, evaluate

Z ∞
I (s, b) = t−1/2 e−st sin(bt) dt.
0
R∞
Solution. Use 0 tν−1 e−(s−ib)t dt = Γ(ν)(s − ib)−ν with ν = 12 and take
imaginary parts: √
I (s, b) = π ℑ (s − ib)−1/2 .
√
Writing s − ib = ρe−iθ with ρ = s2 + b2 and θ = arctan(b/s),
r √ r p√
√  θ  π ρ−s π s2 + b2 − s
I (s, b) = π ρ−1/2 sin = = √ .
2 2 ρ 2 s2 + b2

Problem 115: Mellin of a quadratic denominator [Mellin]

For 0 < ℜ(s) < 2, evaluate

Z ∞ s −1
x
dx.
0 1 + x2

Solution. Let u = x2 so dx = 12 u−1/2 du:

Z ∞ s −1 Z ∞ s/2−1
x 1 u 1 s  s π  πs 
dx = du = Γ Γ 1− = csc .
0 1 + x2 2 0 1+u 2 2 2 2 2
········································································································································
40
 @ integra land

Problem 116: A differentiated principal value kernel [Hilbert]

Show that, for x ∈ R,

Z ∞
1 1 − x2
PV dt = π .
−∞ ( x − t )2 (1 + t2 ) (1 + x 2 )2

Solution. From the known identity

Z ∞
1 x
PV dt = π ,
−∞ ( x − t)(1 + t2 ) 1 + x2
differentiate in x (derivative passes under PV):
Z ∞
1 d x  1 − x2
PV dt = π = π .
−∞ ( x − t )2 (1 + t2 ) dx 1 + x2 (1 + x 2 )2

Problem 117: Hankel image of a weighted Gaussian [Hankel]

Let ν > −1 and a > 0. Evaluate

Z ∞
2
r ν+1 e−ar Jν (kr ) dr.
0

Solution. A standard Gaussian–Bessel identity gives

Z ∞  2
ν+1 − ar2 kν k
r e Jν (kr ) dr = exp − .
0 (2a)ν+1 4a

Problem 118: Vanishing moments =⇒ spectral zero at ω = 0 [Wavelet]

Assume ψ ∈ L1 (R) and tm ψ(t) dt = 0 for m = 0, 1, . . . , M − 1. Show that

R

F {ψ}(ω ) = O(|ω | M ) as ω → 0.

M −1 (−iωt)m
Solution. Expand e−iωt = ∑m =0 m! + R M (ωt) with R M (ξ ) = O(|ξ | M ).
Then
M −1
(−iω )m
Z Z Z
F {ψ}(ω ) = ψ(t)e −iωt
dt = ∑ m!
m
t ψ(t) dt + ψ(t) R M (ωt) dt = O(|ω | M ).
m =0

········································································································································
41
 @ integra land

Problem 119: STFT under time-dilation [STFT]

Let a > 0 and f a (t) = f ( at). Show that

1 ( ga )
Vg f a ( x, ω ) = Vg f ( ax, ω/a), where ga (t) = g(t/a),
a
( ga )
and Vg denotes the STFT computed with window ga .
Solution. Z
Vg f a ( x, ω ) = f ( at) g(t − x )e−iωt dt.

Set u = at so dt = du/a:
1
Z u 
Vg f a ( x, ω ) = f (u) g − x e−i(ω/a)u du
a a
1 1 ( ga )
Z
= f (u) ga (u − ax )e−i(ω/a)u du = Vg f ( ax, ω/a).
a a
For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.

Problem 120: The triangle function and sinc2 [Fourier]

Let Λ(t) = (1 − |t|)1{|t|<1} . Compute F {Λ}(ω ).

Solution. Let f (t) = 1[−1/2, 1/2] (t). Then Λ = f ∗ f and
Z 1/2
2 sin(ω/2)
F { f }(ω ) = e−iωt dt = .
−1/2 ω
Hence
2
4 sin2 (ω/2)

2 2 sin(ω/2)
F {Λ}(ω ) = F { f ∗ f }(ω ) = F { f }(ω ) = = .
ω ω2

Problem 121: Discrete Green’s function for a second difference [Z]

Solve for n ≥ 0:

x [n + 1] − 2x [n] + x [n − 1] = u[n], x [−1] = 0, x [0] = 0,

where u[n] is the unit step.

········································································································································
42
 @ integra land

Solution. Using the unilateral Z-transform, Z { x [n + 1]} = z( X (z) − x [0]) =

zX (z) and Z { x [n − 1]} = z−1 X (z) (since x [−1] = 0). Thus
1 z
( z − 2 + z −1 ) X ( z ) = U ( z ) = = .
1 − z −1 z−1
So
z z z2
X (z) = − 1
· = 3
= z −1 (1 − z −1 ) −3 .
z−2+z z−1 ( z − 1)
Since (1 − z−1 )−3 ←→ (n+ 2
2 ) u [ n ], the factor z
−1 delays by 1:

 
n+1 n ( n + 1)
x [n] = u [ n − 1] = u [ n − 1].
2 2

Problem 122: Fourier inversion as a special FrFT identity [FrFT↔Fourier]

Using the convention Fα π/2 = F , show that for suitable f ,

1

Fα −π/2 Fα π/2 f = f .
2π
1
Solution.
R∞ Since Fα π/2 = F , the left side is 2π Fα −π/2 (F f ). But Fα −π/2 F (t) =
iωt − 1 1 −1
−∞ F (ω ) e dω = 2π F { F }(t ). Therefore 2π Fα −π/2 (F f ) = F (F f ) =
f.

Problem 123: Mellin transform of a Lorentz profile [Mellin]

For 0 < ℜ(s) < 2, compute

 
1
M ( s ).
1 + x2

Solution. By Problem 115,

  Z ∞ s −1
1 x π  πs 
M (s) = dx = csc .
1 + x2 0 1 + x2 2 2

Problem 124: A rational Hilbert pair via differentiation [Hilbert]

 
t
Compute H ( x ).
(1 + t2 )2
········································································································································
43
 @ integra land

Solution. Use H{ f ′ } = (H f )′ and the known pair H{(1 + t2 )−1 }( x ) = x

1+ x 2
.
Since
d 1  2t
= − ,
dt 1 + t2 (1 + t2 )2
we get
1 − x2
 
2t d x 
H − ( x ) = = .
(1 + t2 )2 dx 1 + x2 (1 + x 2 )2
Hence
1 1 − x2
 
t
H ( x ) = − .
(1 + t2 )2 2 (1 + x 2 )2

Problem 125: Gaussian in k with Bessel kernel [Hankel]

For a > 0, evaluate Z ∞

2
e− ak J0 (rk ) k dk.
0

Solution. This is the order-0 Hankel transform of a Gaussian:

Z ∞  2
− ak2 1 r
e J0 (rk ) k dk = exp − .
0 2a 4a

Problem 126: Abel equation solved by Laplace [Laplace]

Solve for t ≥ 0: Z t
f (τ )
√ dτ = t.
0 t−τ

Solution. This is a convolution with k (t) = t−1/2 . Since L{t−1/2 } =

√ −1/2
πs and L{t} = 1/s2 ,
√ 1 1
F (s) π s−1/2 =
F (s) = √ s−3/2 . ⇒
π s2
√
Using L−1 {s−3/2 } = t1/2 /Γ(3/2) = 2t1/2 / π,
1 2t1/2 2√
f (t) = √ · √ = t.
π π π

For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.
········································································································································
44
 @ integra land

Problem 127: A forbidden real eigenvalue for Hilbert [Hilbert]

Assume f is a Schwartz function and H f = λ f for some real λ. Show that

f ≡ 0.
Solution. On Schwartz functions, H2 = − I. Apply H again:
− f = H2 f = H(λ f ) = λ H f = λ2 f .
Thus (λ2 + 1) f = 0. If λ ∈ R then λ2 + 1 > 0, hence f ≡ 0.

Problem 128: Z-transform of a causal cosine [Z]

For Ω ∈ (0, π ), find X (z) = Z {cos(Ωn)u[n]}(z) and its ROC.

Solution. With q = z−1 ,
1 − q cos Ω
 
1
∑ cos(Ωn)q n
=ℜ
1 − eiΩ q
=
1 − 2q cos Ω + q2
.
n ≥0
Hence
1 − cos Ω z−1
X (z) = , ROC: |z| > 1.
1 − 2 cos Ω z−1 + z−2

Problem 129: Mellin scaling and an exponential example [Mellin]

State the scaling rule for the Mellin transform and use it to compute M{e− ax }(s)
for a > 0.
Solution. Scaling: if f a ( x ) = f ( ax ) with a > 0, then M{ f a }(s) = a−s M{ f }(s).
Since M{e− x }(s) = Γ(s) for ℜ(s) > 0, we obtain
M{e−ax }(s) = a−s Γ(s).

Problem 130: A short Hilbert computation via a known pair [Hilbert]

 
1
Evaluate H ( x ).
(1 + t2 )2
n o
t
Solution. Differentiate H 1+t2 ( x ) = − 1+1x2 (a standard pair) and use
2
H{ f ′ } = (H f )′ . Since dtd 1+t t2 = (11+−tt2 )2 , one finds

1 − t2
   
d 1 2x
H ( x ) = − = .
(1 + t2 )2 dx 1 + x2 (1 + x 2 )2
········································································································································
45
 @ integra land

Now write
1 − t2 t2 1 − t2
 
1 1 1
= + = + − ,
(1 + t2 )2 (1 + t2 )2 (1 + t2 )2 (1 + t2 )2 1 + t2 (1 + t2 )2
so
1 − t2
     
1 1
2H =H +H .
(1 + t2 )2 (1 + t2 )2 1 + t2
Using H{(1 + t2 )−1 }( x ) = x
1+ x 2
and the first identity above,

x ( x 2 + 3)
   
1 1 2x x
H (x) = + = .
(1 + t2 )2 2 2
(1 + x ) 2 1 + x2 2(1 + x 2 )2

Problem 131: Gaussian windowed tone (explicit STFT) [STFT]

2 / (2σ2 )
Let f (t) = eiω0 t and gσ (t) = e−t with σ > 0. Compute Vg f ( x, ω ).
Solution. From the tone identity, Vg f ( x, ω ) = e−i(ω −ω0 ) x F { gσ }(ω − ω0 ).
√ 2 2
Also F { gσ }(ξ ) = 2π σ e−σ ξ /2 . Since F { gσ } is real,
√ − i ( ω − ω0 ) x

σ 2 ( ω − ω0 ) 2

Vg f ( x, ω ) = 2π σ e exp − .
2

Problem 132: Haar energy splitting in one step [Wavelet]

Let ( x0 , x1 ) ∈ R2 and define Haar coefficients

x0 + x1 x0 − x1
a= √ , d= √ .
2 2
Show a2 + d2 = x02 + x12 , and state the N-sample version.
Solution. Compute directly:

2 ( x0 + x1 )2 + ( x0 − x1 )2
2
a +d = = x02 + x12 .
2
For an orthonormal DWT (including Haar) on R N , the transform is an orthog-
onal matrix, hence
∥ x ∥22 = ∥ a∥22 + ∥d∥22
at each multiresolution split.
········································································································································
46
 @ integra land

Problem 133: A classic K0 Hankel integral [Hankel]

For a > 0 and k > 0, evaluate

Z ∞
r
J0 (kr ) dr.
0 r 2 + a2
Solution. A standard Hankel pair states
Z ∞
1
K0 ( ak ) J0 (kr ) k dk = .
0 r 2 + a2
By Hankel inversion (order 0), the transform of (r2 + a2 )−1 is K0 ( ak ), i.e.
Z ∞
r
J0 (kr ) dr = K0 ( ak ).
0 r 2 + a2
For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.

Problem 134: A resolvent-type Volterra equation [Laplace]

Solve Z t
y(t) = sin t + e−(t−τ ) y(τ ) dτ, t ≥ 0.
0

Solution. Laplace transforms give

 
1 1 1 1
Y (s) = 2 + Y (s) ⇒ Y (s) 1 − = .
s +1 s+1 s+1 s2 + 1
s+1 1 s 1
Hence Y (s) = 2
= − 2 + 2 , so
s ( s + 1) s s +1 s +1
y(t) = 1 − cos t + sin t.

Problem 135: Fourier transform of a squared Lorentzian [Fourier]

For a > 0, compute  

1
F ( ω ).
( t2 + a2 )2
Solution. Use F {(t2 + a2 )−1 }(ω ) = πa e−a|ω | and
∂ 1  2a 1 1 ∂ 1 
=− 2 ⇒ =− .
∂a t2 + a2 ( t + a2 )2 ( t2 + a2 )2 2a ∂a t2 + a2
········································································································································
47
 @ integra land

Therefore
 
1 1 ∂  π − a|ω |  π
= 3 1 + a |ω | e− a|ω | .


F =− e
( t + a2 )2
2 2a ∂a a 2a

Problem 136: A logarithmic generating function in Z [Z]

Let a ∈ C, a ̸= 0, and define

X (z) = ln(1 − az−1 ), ROC: |z| > | a|.
Find x [n].
Solution. For | az−1 | < 1,
∞
an −n
ln(1 − az −1
)=− ∑ nz .
n =1

Hence
an
x [n] = − u [ n − 1].
n

Problem 137: Mellin of a stretched exponential [Mellin]

For p > 0 and ℜ(s) > 0, evaluate

Z ∞
p
x s−1 e− x dx.
0

Solution. Let u = x p , dx = 1p u1/p−1 du:

Z ∞ Z ∞
s −1 − x p 1 s/p−1 −u 1 s
x e dx = u e du = Γ .
0 p 0 p p

Problem 138: Translation and scaling covariance of Hilbert [Hilbert]

Show that for suitable f and b ∈ R, a > 0,

H{ f (· − b)}( x ) = (H f )( x − b), H{ f ( a·)}( x ) = (H f )( ax ).

Solution. Translation: substitute t = u + b in the PV integral:

1 f (t − b) 1 f (u)
Z Z
H{ f (· − b)}( x ) = PV dt = PV du = (H f )( x − b).
π x−t π ( x − b) − u
········································································································································
48
 @ integra land

Scaling: substitute u = at so dt = du/a:

1 f ( at) 1 f (u) du
Z Z
H{ f ( a·)}( x ) = PV dt = PV
π x−t π x − u/a a
1 f (u)
Z
= PV du = (H f )( ax ).
π ax − u

Problem 139: Gaussian uncertainty product (computed) [STFT/Fourier]

Let
1 −t2 /(2σ2 )
ϕσ (t) = σ > 0, e ,
(πσ2 )1/4
so that ∥ϕσ ∥2 = 1. Using the convention F { f }(ω ) = f (t)e−iωt dt, compute
R

1
Z Z
2 2 2 2
(∆t) = t |ϕσ (t)| dt, (∆ω ) = ω 2 |F {ϕσ }(ω )|2 dω,
2π
and show ∆t ∆ω = 21 .
2 2
Solution. Since |ϕσ (t)|2 = (πσ2 )−1/2 e−t /σ is a centered Gaussian with
variance σ2 /2,
σ
∆t = √ .
2
2 2 √ 2 2 2 2
Also F {e−t /(2σ ) } = 2π σ e−σ ω /2 , hence |F {ϕσ }(ω )|2 = C e−σ ω for a
constant C. A direct Gaussian moment computation in the frequency formula
yields
1
∆ω = √ .
2σ
Therefore ∆t ∆ω = √σ · √1 = 12 .
2 2σ

Problem 140: 2D radial Gaussian Fourier transform [Hankel↔Fourier]

2
Let f ( x ) = e−a| x| on R2 , with a > 0. Compute F2 { f }(ξ ) in radial form.
Solution. For ρ = |ξ |, radial reduction gives
Z ∞  2
| ξ |2
 
− ar2 1 ρ π
F2 { f }(ρ) = 2π e J0 (ρr ) r dr = 2π · exp − = exp − .
0 2a 4a a 4a

For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.
········································································································································
49
 @ integra land

Problem 141: Fourier squared equals reflection (constant tracked) [Fourier↔FrFT]

Show that for suitable f ,

F (F f )(t) = 2π f (−t).

Solution. By the inversion formula,

Z ∞
1
f (−t) = F f (ω )e−iωt dω.
2π −∞

But F (F f )(t) = F f (ω )e−iωt dω, hence F (F f )(t) = 2π f (−t). (Inter-

R

pretable as a π-rotation in time–frequency, up to the 2π normalization.)

Problem 142: Causal accumulation of a polynomial-exponential [Z]

Let x [n] = (n + 1) an u[n] with | a| < 1, and define the cumulative sum y[n] =
∑nm=0 x [m]. Find a closed form for y[n].
Solution. For |z| > | a|,
1
X (z) = ∑ (n + 1)an z−n = (1 − az−1 )2 .
n ≥0

Summation corresponds to division by (1 − z−1 ):

X (z) 1
Y (z) = = .
1 − z −1 (1 − z−1 )(1 − az−1 )2
Partial fractions give
1 1 a 1 a 1
Y (z) = · − · − · .
(1 − a)2 1 − z−1 (1 − a)2 1 − az−1 1 − a (1 − az−1 )2
Invert:
1 a a
y[n] = 2
u[n] − 2
an u[n] − ( n + 1) a n u [ n ]
(1 − a ) (1 − a ) 1−a

1 − ( n + 2 ) a n +1 + ( n + 1 ) a n +2
= u [ n ].
(1 − a )2

········································································································································
50
 @ integra land

Problem 143: A scaled Mellin rational integral [Mellin]

For 0 < ℜ(s) < 2 and λ > 0, evaluate

Z ∞
x s −1
I (λ) = dx.
0 (1 + λx )2

Solution. Scale u = λx:

Z ∞
−s u s −1
I (λ) = λ 2
du = λ−s B(s, 2 − s)
0 (1 + u )
π
= λ − s Γ ( s ) Γ (2 − s ) = λ − s (1 − s ) .
sin(πs)

Problem 144: Hilbert transform and convolution [Hilbert↔Fourier]

For Schwartz f , g, show H( f ∗ g) = (H f ) ∗ g = f ∗ (H g).

Solution. Fourier transforms convert convolution to multiplication:

F {H( f ∗ g)} = (−i sgn ω )F { f ∗ g} = (−i sgn ω )F f F g.

Also

F {(H f ) ∗ g} = F {H f }F g = (−i sgn ω )F f F g,
and similarly for f ∗ (H g). Invert F to conclude the equality.

Problem 145: A clean Laplace evaluation giving log [Laplace]

For s > 0 and b ∈ R, evaluate

Z ∞
sin(bt)
e−st dt.
0 t

Solution. Differentiate in b:
Z ∞ Z ∞
d −st sin( bt ) s
e dt = e−st cos(bt) dt = .
db 0 t 0 s2 + b2
The integral vanishes at b = 0, hence
Z ∞ Z b  
sin(bt) s b
e−st dt = dβ = arctan .
0 t 0 s2 + β2 s
········································································································································
51
 @ integra land

Problem 146: Fourier transform of a causal decaying exponential [Fourier]

For a > 0, compute F {e− at H (t)}(ω ).

Solution. Directly,
Z ∞ Z ∞
− at − at −iωt 1
F {e H (t)}(ω ) = e e dt = e−(a+iω )t dt = .
0 0 a + iω

Problem 147: CWT of a pure tone [Wavelet↔Fourier]

Let f (t) = eiω0 t and assume ψ ∈ L1 ∩ L2 . Show that

q
Wψ f ( a, b) = | a| eiω0 b F {ψ}( aω0 ).
Solution.
t−b
 
1
Z
iω0 t
Wψ f ( a, b) = e p ψ dt.
| a| a
Set u = (t − b)/a so t = au + b, dt = a du:
q Z
iω0 b
Wψ f ( a, b) = | a| e ψ(u)eiω0 au du
q Z q
iω0 b
= | a| e ψ(u)e − i ( aω 0 ) u du = | a| eiω0 b F {ψ}( aω0 ).

For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.

Problem 148: A Bromwich-safe inverse Laplace with a square root [Laplace]

For a > 0, compute

 
1
L −1 √ √ ( t ), t > 0.
s ( s + a)
Solution. Use the algebraic split
 
1 1 1 1
√ √ = √ −√ .
s( s + a) a s s+a
2 √
Also L−1 {s−1/2 } = √1 . A standard identity is L{e a t erfc( a t)} = √1
s+ a
for
πt
ℜs > 0, a > 0. Hence

1

1

1 2 √ 
L −1 √ √ (t) = √ − e a t erfc( a t) .
s( s + a) a πt
········································································································································
52
 @ integra land

Problem 149: A Fourier–Hilbert bridge via the signum multiplier [Hilbert↔Fourier]

Let f ∈ S(R) and define m(ω ) = 1{ω >0} (with half-weight at ω = 0). Show
that the positive-frequency projection satisfies
1
P+ f = F −1 {m(ω )F f (ω )} =


f − iH f .
2

Solution. Since m(ω ) = 12 (1 + sgn ω ) distributionally (with the 0-frequency

convention),
1 1
F −1 {m F f } = F −1 {F f } + F −1 {sgn(ω )F f }.
2 2
But F −1 {F f } = f and F {H f } = (−i sgn ω )F f gives F −1 {sgn(ω )F f } =
i H f . Hence P+ f = 12 ( f − i H f ).

Problem 150: Exact energy of a discrete-time windowed tone [STFT]

Let x [n] = eiΩ0 n and consider the discrete STFT-like map

N −1
S[m, k ] = ∑ x [n]w[n − m]e−i2πkn/N ,
n =0

where w[·] is supported in {0, . . . , N − 1} (zero outside). Express S[m, k ] in

terms of the N-point DFT of w[·].
Solution.
N −1
S[m, k ] = ∑ w[n − m]e−i(2πk/N −Ω0 )n .
n =0
Let u = n − m; then
N −1
−i (2πk/N −Ω0 )m
S[m, k ] = e ∑ w[u]e−i(2πk/N −Ω0 )u .
u =0

Thus
S[m, k ] = e−i(2πk/N −Ω0 )m W (2πk/N − Ω0 ),
where W (θ ) = ∑uN=−01 w[u]e−iθu is the (DTFT/DFT-sampled) spectrum of the
window.

········································································································································
53
 @ integra land

Problem 151: Mellin transform of a power-log profile [Mellin]

For ℜ(α) > −1 and ℜ(s) > 0, compute

M{ x α e− x }(s) and M{ x α (ln x )e− x }(s).

Solution. Z ∞
α −x
M{ x e }(s) = x s+α−1 e− x dx = Γ(s + α).
0
Differentiate in s:
d
M{ x α (ln x )e− x }(s) = Γ ( s + α ) = Γ ( s + α ) ψ ( s + α ),
ds
where ψ is the digamma function.

Problem 152: A nontrivial inverse Z-transform with a repeated pole [Z]

Let | a| < 1 and

1
X (z) = , ROC: |z| > | a|.
(1 − az−1 )3
Find x [n].
Solution. Use the binomial series (1 − q)−3 = ∑∞ n +2 n −1
n=0 ( 2 ) q with q = az :
 
n+2 n (n + 1)(n + 2) n
x [n] = a u[n] = a u [ n ].
2 2

Problem 153: A Hankel kernel that reproduces the heat semigroup [Hankel]

Let ν > −1, κ > 0, and define

Z ∞
2
u(r, t) = e−κk t (Hν f )(k ) Jν (kr ) k dk,
0
R∞
where (Hν f )(k ) = f (ρ) Jν (kρ) ρ dρ. Show that u solves the radial PDE
0

ν2
 
1
ut = κ urr + ur − 2 u , u(r, 0) = f (r ).
r r

Solution. Using the Bessel eigen-identity

 2
ν2

d 1d
2
+ − 2 Jν (kr ) = −k2 Jν (kr ),
dr r dr r
········································································································································
54
 @ integra land

differentiate under the integral:

Z ∞
2
ut = (−κk2 )e−κk t (Hν f )(k) Jν (kr ) k dk,
0

and the radial operator applied to u produces the same factor (−k2 ) inside
the integral. Inversion at t = 0 gives u(r, 0) = f (r ).

Problem 154: A stable spectral factorization check [Z]

Let
(1 − 0.3z−1 )(1 − 2z−1 )
H (z) = .
(1 − 0.5z−1 )(1 − 0.8z−1 )
Assuming a causal realization, determine whether the system is BIBO stable.
Solution. Causality forces ROC outside the outermost pole. Poles at z = 0.5
and z = 0.8; thus causal ROC is |z| > 0.8. Since the unit circle |z| = 1 lies in
|z| > 0.8, the causal system is BIBO stable. (Zeros do not affect BIBO stability,
only poles and ROC.)

Problem 155: A principal-value identity with a rational numerator [Hilbert]

Evaluate, for x ∈ R, Z ∞
t
PV dt.
−∞ ( x − t)(1 + t2 )
Solution. Rewrite
t x t−x
= + ,
1 + t2 1 + t2 1 + t2
so
t 1 1
Z Z Z
PV dt = x PV dt − dt.
( x − t)(1 + t2 ) ( x − t)(1 + t2 ) 1 + t2
R∞
1
= π 1+xx2 and dt
R
Using PV ( x −t)(1+t2 )
dt − ∞ 1+ t2 = π,
Z ∞
x2
 
t π
PV dt = π −1 =− .
−∞ ( x − t)(1 + t2 ) 1 + x2 1 + x2

········································································································································
55
 @ integra land

Problem 156: A Fourier inversion integral yielding a Gaussian [Fourier]

Evaluate, for a > 0, Z ∞

2
e− aω eiωt dω.
−∞
2
Solution. This is the inverse Fourier integral of e− aω up to normalization.
Using the standard Gaussian integral,
Z ∞ r  2
2 π t
e− aω +iωt dω = exp − .
−∞ a 4a

For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.

Problem 157: Laplace transform of an error-function tail [Laplace]

√
For a > 0, compute L{erfc( a t)}(s) for ℜ(s) > 0.
2 √
Solution. Use the known identity L{e a t erfc( a t)} = √1 . By the shift
s+ a
2 2 √
property L{e− a t f (t)}(s) = F (s + a2 ), taking f (t) = e a t erfc( a t) gives
√ 1
L{erfc( a t)}(s) = √ .
s + a2 + a
Rationalize: √
√ s + a2 − a
L{erfc( a t)}(s) = .
s

Problem 158: Mellin inversion residue sketch (asymptotics) [Mellin]

Let f ( x ) = 1+1 x and F (s) = M{ f }(s) = π/ sin(πs) for 0 < ℜ(s) < 1. Use
Mellin inversion to state the leading asymptotics of f ( x ) as x → 0+ and as
x → ∞.
1
R c+i∞ −s
Solution. Mellin inversion uses f ( x ) = 2πi c−i∞ x F ( s ) ds. Poles of F are at
s ∈ Z with residues (−1) . As x → 0 , shift contour right: dominant pole at
n +

s = 0 gives f ( x ) ∼ 1. As x → ∞, shift contour left: dominant pole at s = 1

yields f ( x ) ∼ x −1 . (Indeed 1/(1 + x ) ∼ 1 near 0 and ∼ x −1 at ∞.)

········································································································································
56
 @ integra land

Problem 159: Wavelet detects polynomial trends by vanishing moments [Wavelet]

Let ψ have M ≥ 1 vanishing moments and let f (t) = P(t) + g(t) where P is a
polynomial with deg P < M. Show that Wψ f = Wψ g.
Solution. Linearity gives Wψ f = Wψ P + Wψ g. But Wψ P( a, b) = 0 for all
a ̸= 0 (vanishing moments annihilate polynomials of degree < M). Hence
Wψ f = Wψ g.

Problem 160: FrFT shifts become chirp-modulations [FrFT]

Let f t0 (t) = f (t − t0 ) and α ̸≡ kπ. Show that (Fα f t0 )(u) equals (Fα f )(u)
times a phase factor linear in u, up to a quadratic chirp.
Solution. Insert f (t − t0 ) and substitute v = t − t0 in the kernel:
Z
i 2 + u2 ) cot α −iu (v + t
(Fα f t0 )(u) = e 2 ((v+t0 ) 0 ) csc α f (v) dv.

Factor the v-independent terms and group the linear-in-v term:

Z
i 2 i 2 i 2 cot α −iuv csc α +ivt
(Fα f t0 )(u) = e 2 t0 cot α −iut0 csc α e 2 u cot α f (v) e 2 v 0 cot α dv.

This equals a chirp factor times (Fα f ) evaluated at a frequency-shifted argu-

ment in the rotated domain; in particular the dependence on u includes the
i 2
linear phase e−iut0 csc α and quadratic chirp e 2 u cot α , as claimed.

Problem 161: A Hankel–Gaussian semigroup check [Hankel]

For ν > −1, show that

Z ∞  2
r + ρ2
  
− ak2 1 rρ
e Jν (kr ) Jν (kρ) k dk = exp − Iν , a > 0,
0 2a 4a 2a
where Iν is the modified Bessel function.
Solution. This is the standard Weber–Sonine integral (heat kernel in ra-
dial/Bessel form). It is precisely the kernel of e a∆ν diagonalized by the Hankel
transform, yielding the stated closed form.

········································································································································
57
 @ integra land

Problem 162: A Fourier evaluation using contour residue [Fourier]

For a > 0, compute

Z ∞ Z ∞
e−iωt cos(ωt)
dω and deduce dω.
−∞ ω 2 + a2 −∞ ω 2 + a2

Solution. Standard contour integration gives

Z ∞
e−iωt π
2 2
dω = e− a|t| .
−∞ ω +a a
Taking real parts yields
Z ∞
cos(ωt) π − a|t|
dω = e .
−∞ ω 2 + a2 a

Problem 163: A subtle Z ROC choice for a two-sided exponential [Z]

Let | a| < |b| and define x [n] = an u[n] + bn u[−n − 1]. Find X (z) and its ROC.
Solution. Right-sided part: ∑n≥0 an z−n = 1− az
1
−1 with ROC | z | > | a |. Left-

sided part: ∑n≤−1 bn z−n = − 1−bz

1
−1 with ROC | z | < | b |. Thus

1 1
X (z) = − , ROC: | a| < |z| < |b|.
1 − az−1 1 − bz−1

Problem 164: A Laplace-convolution identity with power kernel [Laplace]

Solve for t ≥ 0: Z t
1 y(τ )
y(t) + √ √ dτ = 1.
π 0 t−τ

Solution. Let k (t) = √1 t−1/2 so L{k }(s) = s−1/2 . Taking Laplace transforms:
π

1 1 1 1
Y (s) 1 + s−1/2 = ⇒ Y (s) = · = √ √ .
s s 1 + s−1/2 s ( s + 1)
By Problem 148 with a = 1,
1 √
y(t) = √ − et erfc( t).
πt
········································································································································
58
 @ integra land

Problem 165: A clean Hilbert identity on cosh [Hilbert]

Show that
 
1 π  πω 
F (ω ) = sech , a > 0,
cosh( at) a 2a
and deduce H{(cosh( at))−1 } in frequency form.
Solution. A standard Fourier pair gives
Z ∞
−iωt π  πω 
sech( at) e dt = sech .
−∞ a 2a
Then F {H f }(ω ) = (−i sgn ω )F { f }(ω ) implies
π  πω 
F {H (sech( at))} (ω ) = −i sgn(ω ) sech .
a 2a
For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.

Problem 166: Mellin transform of a Gaussian in log-scale [Mellin]

1
exp − (ln x )2 /(2σ2 ) on (0, ∞). Compute M{ f }(s).


Let σ > 0 and f ( x ) = x
Solution. With x = ey , dx = ey dy, and f (ey ) = e−y exp(−y2 /(2σ2 )),
Z ∞
s−1 1 −(ln x )2 /(2σ2 )
Z ∞
2 2 √ σ 2 ( s −1)2
M{ f }(s) = x e dx = e(s−1)y e−y /(2σ ) dy = 2π σ e 2 .
0 x −∞

Problem 167: Fourier image of a weighted cusp [Fourier]

Let a > 0 and f (t) = |t|e−a|t| . Compute F { f }(ω ).

2a ∂ − a|t|
Solution. Use F {e− a|t| }(ω ) = 2 2
and e = −|t|e−a|t| :
a +ω ∂a
2( a2 − ω 2 )
 
− a|t| ∂ 2a
F {|t|e }(ω ) = − = 2 .
∂a a2 + ω 2 ( a + ω 2 )2

Problem 168: Sampling bridge (Laplace ↔ Z) [Z↔Laplace]

Let f (t) = e− at H (t) with a > 0, and sample it with period T > 0:
x [n] = f (nT ) = e− anT u[n].
········································································································································
59
 @ integra land

Find X (z) and express X (esT ) in a closed form.

Solution.
∞
1
X (z) = ∑ e−anT z−n = 1 − e−aT z−1 , ROC: |z| > e− aT .
n =0

With z = esT ,
∞
1
sT
X (e ) = ∑ e−(a+s)nT = 1 − e−(a+s)T .
n =0

Problem 169: A rational PV integral with a quadratic numerator [Hilbert]

Evaluate, for x ∈ R,
Z ∞
t2
PV dt.
−∞ ( x − t)(1 + t2 )

t2 1
Solution. Write = 1 − :
1 + t2 1 + t2
t2 1 1
Z Z Z
PV dt = PV dt − PV dt.
( x − t)(1 + t2 ) x−t ( x − t)(1 + t2 )
x
The first PV integral is 0, and the second equals π . Hence
1 + x2
Z ∞
t2 x
PV dt = − π .
−∞ ( x − t)(1 + t2 ) 1 + x2

Problem 170: Multiplicative convolution on (0, 1) [Mellin]

Define the multiplicative convolution

Z ∞
dt
( f ∗× g)( x ) = f ( x/t) g(t) .
0 t
Let f ( x ) = x β−1 1(0,1) ( x ) and g( x ) = x α−1 1(0,1) ( x ) with α ̸= β. Compute
( f ∗× g)( x ) for x > 0.
Solution. For x > 0,
Z 1   β −1 Z 1
x x
α−1 dt β −1
( f ∗× g)( x ) = 1(0,1) t =x tα− β−1 dt.
0 t t t max( x,0)
········································································································································
60
 @ integra land

Thus ( f ∗× g)( x ) = 0 for x ≥ 1, and for 0 < x < 1,

β −1 1 − x α− β
( f ∗× g)( x ) = x .
α−β

Problem 171: CWT of a constant and admissibility [Wavelet]

Let f (t) ≡ 1 and ψ ∈ L1 (R). Show that

q Z ∞
Wψ f ( a, b) = | a| ψ(u) du.
−∞
R
Deduce that if ψ = 0 then Wψ f ≡ 0.
Solution.
t−b
 
1
Z q Z
Wψ f ( a, b) = p ψ dt = | a| ψ(u) du.
| a| a
R
If ψ = 0 (zero mean), then Wψ f ( a, b) = 0 for all ( a, b).

Problem 172: Moyal identity for the STFT [STFT]

Show that for f , g ∈ L2 (R),

Z ∞ Z ∞
2
Vg f ( x, ω ) dω dx = 2π ∥ f ∥22 ∥ g∥22 .
−∞ −∞

Solution. For fixed x, Vg f ( x, ω ) = F { f (t) g(t − x )}(ω ), so by Plancherel,

Z Z
2
|Vg f ( x, ω )| dω = 2π | f (t)|2 | g(t − x )|2 dt.

Integrate over x and swap integrals:

ZZ Z Z 
2 2
|Vg f | dω dx = 2π | f (t)| | g(t − x )| dx dt = 2π ∥ f ∥22 ∥ g∥22 .
2

For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.

········································································································································
61
 @ integra land

Problem 173: Convolution of two two-sided exponentials [Fourier]

For a, b > 0, a ̸= b, compute (e− a|·| ∗ e−b|·| )(t).

Solution.
2a 2b
F {e−a|t| }(ω ) = , F {e−b|t| }(ω ) = .
a2 + ω 2 b2 + ω 2
Thus
 
4ab 4ab 1 1
F { f ∗ g} = 2 = − .
( a + ω 2 )(b2 + ω 2 ) b2 − a2 a2 + ω 2 b2 + ω 2
Using F −1 {( a2 + ω 2 )−1 }(t) = 1 − a|t|
2a e
gives
− a|·| −b|·| 2  − a|t| −b|t|

(e ∗e )(t) = 2 be −ae .
b − a2

Problem 174: Radial heat kernel from a ring delta [Hankel]

 
Consider ut = κ urr + 1r ur for r > 0, t > 0 with κ > 0. Take initial data
u(r, 0) = δ(r − R)/R (R > 0). Find u(r, t).
2
Solution. Hankel0 transform gives (Hν u)(k, t) = e−κk t (Hν u0 )(k), and
Z ∞
δ (r − R )
(Hν u0 )(k) = J0 (kr ) r dr = J0 (kR).
0 R
Thus
Z ∞  2
r + R2
  
−κk2 t 1 rR
u(r, t) = e J0 (kR) J0 (kr ) k dk = exp − I0 .
0 2κt 4κt 2κt

Problem 175: A forced first-order difference equation [Z]

Solve for n ≥ 0:
y[n] − 21 y[n − 1] = (−1)n , y[−1] = 0.

Solution. Let x [n] = (−1)n u[n], so X (z) = ∑n≥0 (−1)n z−n = 1+1z−1 (ROC
|z| > 1). Z-transform of the recursion gives (1 − 21 z−1 )Y (z) = X (z), hence
1 1/3 2/3
Y (z) = = + .
(1 − 12 z−1 )(1 + z−1 ) 1 − 12 z−1 1 + z−1
········································································································································
62
 @ integra land

Therefore
1  1 n 2
y[n] = 2 u [ n ] + (−1)n u[n].
3 3

Problem 176: FrFT of a dechirped Gaussian [FrFT↔Fourier]

Let a > 0 and α ̸≡ kπ. Define

2 i 2 cot α
f (t) = e−at e− 2 t .

Compute (Fα f )(u).

i 2 cot α 2
dechirp identity (Fα f )(u) = e 2 u
Solution. Using the q F {e−at }(u csc α),
2 π −ω 2 /(4a)
and F {e−at }(ω ) = a e , we obtain

u2 csc2 α
r  
i 2 π
(Fα f )(u) = e 2 u cot α exp − .
a 4a

Problem 177: An inverse Laplace with a logarithmic ratio [Laplace]

For a, b > 0, find

  
−1 1 s+b
L ln ( t ), t > 0.
s s+a

Solution. Use the standard identity (for ℜs > 0)

Z ∞ − at − e−bt  
−st e s+b
e dt = ln .
0 t s+a
− at −bt
Thus if g(t) = e −t e , then L{ g} = ln ss+ b


+ a and
Z t − aτ
− e−bτ
  Z t
−1 1 e
L L{ g}(s) (t) = g(τ ) dτ = dτ.
s 0 0 τ

Problem 178: A scaled Mellin rational integral [Mellin]

For 0 < ℜ(s) < 1 and λ > 0, evaluate

Z ∞
x s −1
dx.
0 1 + λx
········································································································································
63
 @ integra land

Solution. Scale u = λx:

Z ∞ Z ∞ s −1
x s −1 −s u π
dx = λ du = λ−s Γ(s)Γ(1 − s) = λ−s .
0 1 + λx 0 1+u sin(πs)

Problem 179: 3D radial Fourier reduction (spherical Bessel) [Fourier↔Hankel]

Let f ( x ) = F (| x |) be radial on R3 . Show that F3 { f } is radial and

Z ∞
sin(ρr ) 2
F3 { f }(ρ) = 4π F (r ) r dr, ρ = | ξ |.
0 ρr

Solution. Align ξ with the polar axis and integrate over the sphere:
Z ∞Z Z ∞ Z 1 
F3 { f }(ξ ) = F (r )e−iρr cos θ r2 dΩ dr = 2π F (r )r 2 e−iρrµ dµ dr,
0 S2 0 −1
R1 −iρrµ dµ 2 sin(ρr )
and −1 e = ρr , giving the formula.
For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.

Problem 180: Fourier transform of the principal value kernel [Fourier/Hilbert]

Show (in S ′ ) that  

1
F PV (ω ) = −iπ sgn(ω ).
t
 
1 1
Solution. Using F {H f } = (−i sgn ω )F { f } and H f = f ∗ PV , take π t
Fourier transforms:
   
1 1 1
F PV = −i sgn(ω ) ⇒ F PV = −iπ sgn(ω ).
π t t

Problem 181: Two-sided inverse Z with annular ROC [Z]

Let
1
X (z) = , ROC: 0.5 < |z| < 2.
(1 − 0.5z−1 )(1 − 2z−1 )
Find x [n].

········································································································································
64
 @ integra land

Solution. Partial fractions in q = z−1 :

−1/3 4/3
X (z) = + .
1 − 0.5z−1 1 − 2z−1
With ROC |z| > 0.5, the first term is right-sided: (0.5)n u[n]. With ROC |z| < 2,
the second term is left-sided: −2n u[−n − 1]. Thus
1 4
x [n] = − (0.5)n u[n] − 2n u[−n − 1].
3 3

Problem 182: Wave equation on a half-line with step boundary [Laplace]

Solve utt = c2 u xx for x > 0, t > 0 with u( x, 0) = ut ( x, 0) = 0, boundary

u(0, t) = H (t), and boundedness as x → ∞.
1
Solution. Laplace in t gives s2 U = c2 Uxx with U (0, s) = s and U (·, s)
bounded as x → ∞. Hence U ( x, s) = 1s e−sx/c . Invert:
 x
u( x, t) = H t − .
c

Problem 183: A squared-denominator Mellin integral [Mellin]

For p > 0 and 0 < ℜ(s) < 2p, evaluate

Z ∞
x s −1
dx.
0 (1 + x p )2

Solution. Let u = x p so dx = 1p u1/p−1 du:

Z ∞ Z ∞ s/p−1
x s −1
     
1 u 1 s s 1 s s
dx = du = B ,2− = Γ Γ 2− .
0 (1 + x p )2 p 0 (1 + u )2 p p p p p p

Problem 184: Radial Laplacian diagonalized by Hankel0 [Hankel]

Assume f is smooth and decays sufficiently fast so boundary terms vanish.

Show that for ν = 0,
 
1
Hν f rr + f r (k) = −k2 (Hν f )(k).
r

········································································································································
65
 @ integra land

R∞
Solution. Use (Hν f )(k ) = f (r ) J0 (kr ) r drand integrate
0  by parts twice,
d 2 1 d
moving the radial operator onto J0 (kr ). Since dr 2 + r dr J0 (kr ) = −k2 J0 (kr )
and boundary terms vanish, the identity follows.

Problem 185: A half-line cosine integral of order two [Fourier]

For a ∈ R and b > 0, evaluate

Z ∞
cos( ax )
dx.
0 ( x 2 + b2 )2

Solution. From F {(t2 + b2 )−2 }(ω ) = π

2b3
(1 + b|ω |)e−b|ω | ,
Z ∞
cos( ax ) π
2 2 2
dx = 3
(1 + b| a|)e−b|a| .
−∞ (x + b ) 2b
Halving gives
Z ∞
cos( ax ) π
dx = (1 + b| a|)e−b|a| .
0 ( x 2 + b2 )2 4b 3

Problem 186: CWT of a linear trend [Wavelet]

Let f (t) = t and ψ ∈ L1 (R) with

R
ψ(t) dt = 0. Show that for a ̸= 0,
Z ∞
3/2
Wψ f ( a, b) = a u ψ(u) du,
−∞

independent of b.
Solution. Change variables u = (t − b)/a, so t = au + b, dt = a du:
q  Z 
1
Z Z
Wψ f ( a, b) = ( au + b) p ψ(u) a du = | a| a uψ(u) du + b ψ(u) du .
| a|
p
If ψ = 0, the b-term vanishes and | a| a = a3/2 (for a > 0; for a < 0
R

interpret with | a| similarly), yielding the result.

For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.

········································································································································
66
 @ integra land

Problem 187: Inverse Z of a resonator [Z]

Let 0 < r < 1 and Ω ∈ (0, π ). Find x [n] such that

1
X (z) = , ROC: |z| > r.
1 − 2r cos Ω z−1 + r2 z−2
Solution. Factor the denominator:
1 − 2r cos Ω z−1 + r2 z−2 = (1 − reiΩ z−1 )(1 − re−iΩ z−1 ).
A standard causal inversion gives
r n sin((n + 1)Ω)
x [n] = u [ n ].
sin Ω

Problem 188: Inverse Laplace of a double harmonic pole [Laplace]

Compute L−1 {(s2 + a2 )−2 }(t) for a > 0.

a s2 − a2
Solution. Use L{sin( at)} = s2 + a2
and L{t cos( at)} = ( s2 + a2 )2
. Then

s2 − a2
   
1
1 1 1
L sin ( at ) − at cos ( at ) = 2 − = .
2a3 2a s2 + a2 ( s2 + a2 )2 ( s2 + a2 )2
Hence  
−1 1 1

L (t) = sin ( at ) − at cos ( at ) .
( s2 + a2 )2 2a3

Problem 189: A rational integral via Parseval [Fourier]

Evaluate, for a > 0, Z ∞

dω
.
−∞ ( a2 + ω 2 )2

Solution. From F {(t2 + a2 )−2 }(ω ) = π

2a3
(1 + a|ω |)e− a|ω | , set ω = 0 and use
Z ∞
e−iωt π
2 2 2
dt = 3
(1 + a|ω |)e−a|ω | .
−∞ (t + a ) 2a
Equivalently, compute the known integral:
Z ∞
dω π
= .
−∞ ( a2 + ω 2 )2 2a3
········································································································································
67
 @ integra land

Problem 190: STFT as a modulated convolution in time [STFT]

Let g̃(t) = g(−t). Show that

Vg f ( x, ω ) = e−iωx ( f (·)e−iω (·) ∗ g̃)( x ).

Solution.
Z Z
−iω (·) −iωt
( f (·)e ∗ g̃)( x ) = f (t)e g̃( x − t) dt = f (t)e−iωt g(t − x ) dt.

Multiplying by e−iωx yields

Z Z
−iωx −iωt
e f (t) g(t − x )e dt = f (t) g(t − x )e−iωt dt = Vg f ( x, ω ).

Problem 191: A PV identity from H{t/(t2 + a2 )} [Hilbert]

 
t
For a > 0, compute H 2 ( x ).
t + a2
 
u 1
Solution. Using the known pair H ( x ) = − and scaling
1 + u2 1 + x2
t 1 (t/a) 1
= · · , one gets
t2 + a2 a 1 + (t/a)2 a
 
t 1
H 2 2
(x) = − 2 .
t +a x + a2

Problem 192: A Hankel distributional pair [Hankel]

Show (distributionally) that for a > 0,

Z ∞
1
J0 ( ar ) J0 (kr ) r dr = δ(k − a),
0 a
i.e. Hν { J0 ( ar )}(k ) = δ(k − a)/a.
R∞
Solution. Hankel0 is self-inverse: if F (k ) = Hν f (k ), then f (r ) = 0 F (k ) J0 (kr ) k dk.
Choose F (k ) = δ(k − a)/a. Then
Z ∞
δ(k − a)
J0 (kr ) k dk = J0 ( ar ),
0 a
so indeed Hν { J0 ( ar )}(k ) = δ(k − a)/a, which is equivalent to the stated
orthogonality relation.
········································································································································
68
 @ integra land

Problem 193: PV integral for a squared Lorentzian [Hilbert]

For a > 0 and x ∈ R, evaluate

Z ∞
1
PV dt.
−∞ ( x − t)(t2 + a2 )2

Solution. This equals π H{(t2 + a2 )−2 }( x ). Using scaling from H{(1 +

2 − 2 x ( x 2 + 3)
t ) }( x ) = yields
2(1 + x 2 )2

x ( x2 + 3a2 )
 
1
H (x) = 3 2 .
( t2 + a2 )2 2a ( x + a2 )2
Therefore Z ∞
1 x ( x2 + 3a2 )
PV dt = π .
−∞ ( x − t)(t2 + a2 )2 2a3 ( x2 + a2 )2
For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.

Problem 194: Mellin image of the Euler operator [Mellin]

Assume f is such that boundary terms vanish. Show that

M{ x f ′ ( x )}(s) = −s M{ f }(s).

Solution.
Z ∞ Z ∞
′ s −1 ′
M{ x f ( x )}(s) = x x f ( x ) dx = x s f ′ ( x ) dx
0 0
∞ Z ∞
= xs f (x) 0 − s x s−1 f ( x ) dx = −s M{ f }(s).

0

Problem 195: Wave equation with impulsive boundary data [Laplace]

Solve utt = c2 u xx for x > 0, t > 0 with u( x, 0) = ut ( x, 0) = 0, boundary

u(0, t) = δ(t), and boundedness as x → ∞.
Solution. Laplace in t gives s2 U = c2 Uxx with U (0,
s) = 1. The bounded
solution is U ( x, s) = e − sx/c x
. Invert: u( x, t) = δ t − c .
········································································································································
69
 @ integra land

Problem 196: A two-sided realization from ROC choices [Z]

Let
1 − 1.2z−1
H (z) = .
1 − 0.6z−1
Find the impulse response h[n] for (i) ROC |z| > 0.6 and (ii) ROC |z| < 0.6,
and state which case is causal and BIBO stable.
−1 (1.2−0.6)z−1 −1 −1
Solution. Write H (z) = 1 − 1−0.6z
0.6z−1
− 1−0.6z−1
= 1 − 1−0.6z
0.6z−1
− 1−0.6z
0.6z−1
=
−1
1 − 1−1.2z
0.6z−1
.

1
(i) If |z| > 0.6, then 1−0.6z−1
↔ (0.6)n u[n], so

h[n] = δ[n] − 1.2(0.6)n−1 u[n − 1].

This is causal and stable (pole inside unit circle, ROC includes |z| = 1).
1
(ii) If |z| < 0.6, then 1−0.6z−1
↔ −(0.6)n u[−n − 1], so

h[n] = δ[n] + 1.2(0.6)n−1 u[−n].

This is anti-causal (left-sided tail); not causal, though it is absolutely

summable (stable) since |0.6| < 1.

Problem 197: A clean Hilbert PV formula (repackaged) [Hilbert]

Show that for a > 0,

Z ∞
t π
PV dt = − .
− ∞ ( t2 + a2 )( x − t) x2 + a2
n o
t
Solution. By definition, the PV integral equals π H t2 + a2
( x ). Using
n o
t 1
H t2 + a2 ( x ) = − x 2 + a2
(Problem 191), the claim follows.

Problem 198: A compact inverse Fourier integral for causality [Fourier↔Laplace]

For a > 0, evaluate

Z ∞
eiωt
dω
−∞ a + iω
········································································································································
70
 @ integra land

in the distributional sense, and interpret the result.

Solution. Since F {e− at H (t)}(ω ) = 1
a+iω , inversion gives
Z ∞
1 eiωt
dω = e− at H (t).
2π −∞ a + iω
Hence Z ∞
eiωt
dω = 2π e− at H (t),
−∞ a + iω
exhibiting causality via the half-line support.

Problem 199: Mellin of a log-derivative (safe identity) [Mellin]

Assume f decays so boundary terms vanish and set D f ( x ) = d(lnd x) f ( x ) =

x f ′ ( x ). Show that Mellin diagonalizes D as multiplication by −s.
Solution. This is exactly Problem 194:
M{ D f }(s) = M{ x f ′ ( x )}(s) = −s M{ f }(s).

Problem 200: A FrFT variant via differentiation of a Gaussian [FrFT↔Fourier]

Let a > 0 and α ̸≡ kπ. Define

2 i 2 cot α
f (t) = t e−at e− 2 t .
Compute (Fα f )(u).
Solution. By the dechirp identity,
i 2 cot α 2
F {te−at }(u csc α).
(Fα f )(u) = e 2 u
2
q 2
Using F {tϕ(t)} = i ∂ω F {ϕ} and F {e − at }(ω ) = πa e−ω /(4a) ,
r  r
− at2 d π −ω2 /(4a) π ω −ω2 /(4a)
F {te }(ω ) = i e = −i e .
dω a a 2a
Therefore
u2 csc2 α
r  
i 2 π u csc α
(Fα f )(u) = −i e 2 u cot α exp − .
a 2a 4a

For complete derivations, broader theory, and extended problem sets, see ART OF
INTEGRAL TRANSFORMS.
········································································································································
71
 @ integra land

Problem 201: A log-oscillatory Mellin integral with 1 + x2 [Mellin]

For 0 < c < 2 and τ ∈ R, evaluate

Z ∞ c −1
x cos(τ ln x )
dx.
0 1 + x2

Solution. Write cos(τ ln x ) = ℜ( xiτ ):

Z ∞ c −1 Z ∞ (c+iτ )−1   
x cos(τ ln x ) x π π (c + iτ )
dx = ℜ dx = ℜ csc .
0 1 + x2 0 1 + x2 2 2
sin a cosh b
With a = πc
2 and b = πτ
2 , ℜ(csc( a + ib)) = , so
sin2 a + sinh2 b
Z ∞ c −1
x cos(τ ln x ) π sin( πc πτ
2 ) cosh( 2 )
dx = .
0 1 + x2 2 sin2 ( πc ) + sinh2 ( πτ )
2 2

Problem 202: DTFT as Z on the unit circle [Z↔Fourier]

Let x ∈ ℓ1 (Z) and define the DTFT

∞
XDTFT (ω ) = ∑ x [n]e−iωn .
n=−∞

Show that XDTFT (ω ) = X (z) z=eiω

, where X (z) = Z { x [n]}(z).
Solution. By definition,
∞
X (z) = ∑ x [n ] z−n .
n=−∞

For z = eiω (which lies in the ROC when x ∈ ℓ1 ),

∞
iω
X (e ) = ∑ x [n]e−iωn = XDTFT (ω ).
n=−∞

Problem 203: Modified Bessel boundary-value profile [Hankel/ODE]

Solve the radial ODE on 0 ≤ r ≤ R:

1
urr + ur − λ2 u = 0, u(0) finite, u( R) = 1,
r
········································································································································
72
 @ integra land

for λ > 0.
Solution. The finite-at-0 solution is u(r ) = C I0 (λr ). Enforce u( R) = 1:
I0 (λr )
u (r ) = .
I0 (λR)

Problem 204: An STFT “windowed spectrum” identity [STFT↔Fourier]

Show that for fixed x,

Vg f ( x, ω ) = F { f (t) g(t − x )}(ω ),

and interpret |Vg f ( x, ω )|2 as a local spectrum.

Solution. This is immediate from the definition:
Z
Vg f ( x, ω ) = f (t) g(t − x )e−iωt dt = F { f (·) g(· − x )}(ω ).

Thus |Vg f ( x, ω )|2 is the squared magnitude of the Fourier transform of the
windowed signal f (t) g(t − x ), i.e. the local (time-indexed) spectrum.

Problem 205: Fourier proof of a classical decay estimate [Fourier]

Assume f ∈ L1 (R) and f ′ exists with f ′ ∈ L1 (R). Show that

∥ f ′ ∥1
|F { f }(ω )| ≤ ( ω ̸ = 0).
|ω |

Solution. Integrate by parts:

1 1
Z Z
−iωt
F { f }(ω ) = f (t)e dt = f ′ (t)e−iωt dt = F { f ′ }(ω ),
iω iω
where boundary terms vanish since f ∈ L1 and f ′ ∈ L1 . Hence |F f (ω )| ≤
1
R ′ ∥ f ′ ∥1
|ω |
| f ( t )| dt = |ω |
.

········································································································································
73