Worked solutions on contour integrals of a rational function and on Cauchy estimates with their equality case.
§9.3Estimated reading time: 35 min
This short page revises the Cauchy formula and contains worked solutions.
Conventions
D(a,r)={z:∣z−a∣<r} is the open disk and D(a,r)={z:∣z−a∣≤r} is the closed disk.
∣z−a∣=r denotes the circle, positively oriented (counterclockwise, traversed once) unless a question expressly says it is reversed. Reversing the orientation multiplies a contour integral by −1.
"Holomorphic on a neighborhood of D(a,r)" means holomorphic on some open set containing the closed disk. The boundary circle is included.
The results used
Theorem 9.3.1
(Cauchy integral formula with derivatives)
Let h be holomorphic on an open set containing D(a,r). Then for every z0 with ∣z0−a∣<r and every integer n≥0,
Take h≡1, which is entire. For m=1 the formula with n=0 reads 1=2πi1∮∣z−a∣=rz−z0dz. For m≥2 take n=m−1≥1: the left side is h(m−1)(z0)=0, because every derivative of a constant is 0, so the integral vanishes.
On-contour poles. A genuine pole on one of these circles makes the parametrized integrand undefined at that point and unbounded nearby, so the usual contour integral is undefined. Check cancellations first: a removable singularity can instead be filled by its holomorphic extension. This set does not replace a divergent integral by a principal value, which is a separate construction.
Counterexample mode
A pole of the numerator inside the circle
Claim: if g is holomorphic on D(0,1) except at the center, then ∮∣z∣=1z−z0g(z)dz=2πig(z0) for ∣z0∣<1. Take g(z)=1/z and z0=21. The claimed value is 2πi⋅2=4πi. But
z(z−21)1=2(z−211−z1),
and both 21 and 0 lie inside ∣z∣=1, so the standard integrals give 2(2πi−2πi)=0=4πi. The formula needs g holomorphic on an open set containing the whole closed disk, center included: a pole of g inside the circle contributes a term of its own.
Roadmap
Topic
Main skill
Key tools
Contour integrals
Evaluating ∮F for a rational F with two poles, over contours that enclose different poles; reversed orientation; an on-contour case
Splitting F as (holomorphic part)/(singular factor), partial fractions, Cauchy formula for h and h′
Cauchy estimates
Deriving Cauchy estimates, optimizing them on a concrete function, and deciding exactly when equality can hold
Derivative formula, the ML-type estimate, the equality case of the triangle inequality for integrals
Checklist before you write any Cauchy-formula step
Locate every singularity of the integrand, including singularities of the "numerator" part you intend to call h.
Check for cancellations explicitly. Factor the numerator and confirm whether it vanishes at any pole. Never assume a pole cancels.
Name the closed disk on which h must be holomorphic, and check that every singularity of h lies strictly outside it. A singularity on the boundary is not allowed.
Check the point z0 lies strictly inside the circle.
Check the orientation. Insert a factor −1 for a reversed circle.
Check whether any genuine pole lies on the contour. If one does, stop: the usual integral is undefined. A removable singularity must first be filled by its holomorphic extension.
For Cauchy estimates, state the open set of holomorphy, the closed disk, and the number M with ∣f∣≤M on the circle.
Sanity-check the answer. Do sums over small circles match the value over a big circle? Is a bound at least as large as the exact value?
Common mistakes
Common mistake
(Using a circle that encloses the singularity)
For f(z)=1/(3−z) it is tempting to use any radius, say R=6. On ∣z∣=6 one has ∣3−z∣≥3, so M=1/3, and the estimate with n=1 would claim ∣f′(0)∣≤61/3=181. But f′(0)=91>181. The hypothesis fails: f is not holomorphic at 3, which lies inside D(0,6). The estimate holds only for 0<R<3.
Common mistake
(Exponent and factorial in the derivative formula)
The denominator (z−z0)n+1 fixes the order: (z−1)2 means n=1, whereas (z−1)3 means n=2, and then the factor n!2πi must not be forgotten. For example, ∮∣z∣=2(z−1)3ezdz=2!2πih′′(1)=πie for h=ez, not 2πie.
Quick checks
Quick check
(Evaluate ∮∣z∣=2(z−1)2ezdz.)
Identify h, z0 and n first.
Solution · Answer
Here h(z)=ez is entire, z0=1 lies inside the circle, and (z−1)2=(z−z0)n+1 gives n=1. So the integral is 2πih′(1)=2πie.
Quick check
(Apply the Cauchy estimate to f(z)=ez at a=0 on ∣z∣=R with R=n (n≥1). What bound results for ∣f(n)(0)∣, and is it attained?)
First find the maximum of ∣ez∣ on the circle.
Solution · Answer
On ∣z∣=R we have ∣ez∣=eRez≤eR, with equality at z=R. So M=eR and the bound is n!eR/Rn; at R=n it equals n!en/nn (this radius minimizes eR/Rn, since R−nlnR has derivative 1−n/R). The exact value is f(n)(0)=1. Because en=k∑nk/k!>nn/n!, the bound exceeds 1; for n=1 it is e. It is not attained, in agreement with the equality criterion below, since ez is not a monomial.
Contour integrals of a rational function
Let
F(z)=(z−1)2(z+3)z2.
(i) Identify the singularities of F, and confirm that no cancellation occurs.
Evaluate each of the following exactly. If an integral does not exist in the usual sense, say so and explain why.
(ii)∮∣z∣=2F(z)dz
(iii)∮∣z+3∣=1F(z)dz
(iv)∮∣z∣=4F(z)dz, where the circle is traversed clockwise (reversed orientation). Use an explicit algebraic decomposition of F.
(v)∮∣z−6∣=2F(z)dz
(vi)∮∣z∣=3F(z)dz
(vii) Check that your answers to (ii), (iii) and (iv) are consistent.
Solutions: contour integrals
(i) Singularities. The denominator (z−1)2(z+3) vanishes only at z=1 (to order 2) and at z=−3 (to order 1).
The numerator z2 vanishes only at z=0. In particular:
at z=1 the numerator equals 1=0;
at z=−3 the numerator equals 9=0.
So no cancellation occurs. F has a genuine double pole at 1 and a genuine simple pole at −3, and F is holomorphic on C∖{1,−3}.
Where the two poles sit relative to each contour:
Contour
z=1
z=−3
∣z∣=2
inside (∣1∣=1<2)
outside (∣−3∣=3>2)
∣z+3∣=1
outside (∣1+3∣=4>1)
inside (center)
∣z∣=4
inside
inside (3<4)
∣z−6∣=2
outside (∣1−6∣=5>2)
outside (∣−3−6∣=9>2)
∣z∣=3
inside
on the contour
Worked example 9.3.3
(Only the double pole is enclosed)
Write
F(z)=(z−1)2h(z),h(z)=z+3z2.
The numerator h is itself rational and has its own pole at −3. That pole is the one to watch. Since ∣−3∣=3>2, it lies outside D(0,2). In fact h is holomorphic on the open disk D(0,3), which contains D(0,2).
The point z0=1 satisfies ∣1∣<2. The derivative formula with n=1 gives
∮∣z∣=2(z−1)2h(z)dz=2πih′(1).
By the quotient rule,
h′(z)=(z+3)22z(z+3)−z2=(z+3)2z2+6z,h′(1)=167.
Therefore
∮∣z∣=2F(z)dz=2πi⋅167=87πi.
(iii) Over ∣z+3∣=1: only the simple pole at −3 is enclosed.
Write
F(z)=z+3k(z),k(z)=(z−1)2z2.
The only singularity of k is z=1, and ∣1−(−3)∣=4>1. So k is holomorphic on D(−3,4), which is an open set containing D(−3,1).
The center z0=−3 lies inside the circle. Cauchy's formula with n=0 gives
∮∣z+3∣=1F(z)dz=2πik(−3)=2πi⋅169=89πi.
Worked example 9.3.4
(Both poles enclosed, clockwise: partial fractions)
No single factorization F=h/(z−z0)m works here. Any such h would still carry the other pole inside D(0,4). Instead, decompose F algebraically by partial fractions:
(z−1)2(z+3)z2=z−1A+(z−1)2B+z+3C.
Finding the coefficients.
C: multiply by (z+3) and set z=−3. This gives C=(−4)29=169.
B: multiply by (z−1)2 and set z=1. This gives B=41.
A: multiply by z and let z→∞. The left side tends to 1 and the right side tends to A+C. So A=1−169=167.
Verification by recombining. Compute A(z−1)(z+3)+B(z+3)+C(z−1)2 term by term:
167(z2+2z−3)+41(z+3)+169(z2−2z+1).
Collect powers of z:
z2: 167+169=1;
z: 1614+164−1618=0;
constant: −1621+1612+169=0.
The result is z2, so the decomposition is correct.
Notice that A=h′(1) from part (ii). This is no coincidence: it is the same computation seen from a different angle.
Integrating each part, positive orientation first. Each term has the form (z−z0)m1 with numerator 1. The constant 1 is entire, so it is holomorphic on a neighborhood of D(0,4). Both z0=1 and z0=−3 lie strictly inside ∣z∣=4. By the standard integrals:
∮∣z∣=4Fdz=167(2πi)+41⋅0+169(2πi)=2πi.
Reversing the orientation. The clockwise circle gives the negative of this value:
∮∣z∣=4,clockwiseF(z)dz=−2πi.
(v) Over ∣z−6∣=2: no pole is enclosed.
The closed disk D(6,2) lies in {Rez≥4}, so it contains neither 1 nor −3. Hence F is holomorphic on an open set containing D(6,2).
One can quote Cauchy's theorem. Alternatively, use the integral formula directly. Set g(z)=(z−6)F(z). Then g is holomorphic on the same neighborhood, and F(z)=z−6g(z). Since z0=6 is the center,
∮∣z−6∣=2Fdz=2πig(6)=2πi⋅0.
So
∮∣z−6∣=2F(z)dz=0.
Worked example 9.3.5
(A pole on the contour)
The point −3 satisfies ∣−3∣=3, so it lies on the circle. Near z=−3,
∣F(z)∣=∣z−1∣2∣z+3∣∣z∣2→∞.
So F is undefined at a point of the contour and unbounded near it. It is not continuous on the contour, and the usual contour integral ∮∣z∣=3Fdz is undefined. We do not assign it a principal value.
Two tempting arguments fail here.
Tempting but wrong: "1 is inside, so the answer is 2πih′(1)." The numerator h(z)=z2/(z+3) from part (ii) has its pole at −3, which lies on the boundary of D(0,3). So h is not holomorphic on any neighborhood of that closed disk, and the hypothesis of the derivative formula fails.
Tempting but wrong: "half of the pole at −3 counts." That reasoning belongs to principal-value theory, which is not the usual contour integral.
(vii) Consistency check.
The two small-circle answers add up to the big-circle answer:
87πi+89πi=2πi=∮∣z∣=4,counterclockwiseFdz.
This agrees with (iv). The total over a circle enclosing both poles equals the sum of the contributions of the two poles, and reversing orientation gives −2πi.
Cauchy estimates and their equality case
(i) Cauchy estimates. Let f be holomorphic on an open set U containing D(a,R), where R>0. Suppose ∣f(z)∣≤M for all z with ∣z−a∣=R. Prove that for every integer n≥0,
∣f(n)(a)∣≤Rnn!M.
(ii) A concrete bound. Let f(z)=3−z1 and a=0.
(a) State precisely which radii R are admissible in (i).
(b) For each admissible R, find the best constant M(R).
(c) Write down the resulting bound for ∣f(n)(0)∣, and optimize it over R.
(d) Compare the optimized bound with the exact value of f(n)(0). Is equality ever attained for n≥1?
(iii) Off-center bound. For the same f, use a suitable disk to show that ∣f′(z0)∣≤1 whenever ∣z0∣≤1. Then compare with the exact value.
(iv) The equality case. Under the hypotheses of (i), with M taken to be the maximum of ∣f∣ on the circle, prove the following:
∣f(n)(a)∣=Rnn!Mif and only iff(z)=c(z−a)n on D(a,R)for some constant c with ∣c∣=M/Rn.
(v) Testing equality. Decide whether equality holds in the Cauchy estimate in each case, and explain using (iv):
(a) p(z)=z2+z on D(0,1), for n=1 and for n=2;
(b) q(z)=3z2 on D(0,2), for n=1 and for n=2.
Solutions: Cauchy estimates
(i) Statement and proof.
Theorem 9.3.6
(Cauchy estimates)
Let f be holomorphic on an open set containing D(a,R) and suppose ∣f∣≤M on ∣z−a∣=R. Then ∣f(n)(a)∣≤n!M/Rn for every n≥0.
Proof
(Cauchy estimate from the derivative formula)
The hypotheses of the derivative formula hold with h=f, z0=a and r=R. So
f(n)(a)=2πin!∮∣z−a∣=R(z−a)n+1f(z)dz.
Parametrize the circle by z=a+Reit for 0≤t≤2π, so that dz=iReitdt. Substituting,
(a) Admissible radii. The function f is holomorphic on U=C∖{3}. The closed disk D(0,R) lies in U exactly when 3∈/D(0,R), that is, when R<3. When R<3, U is an open set containing D(0,R).
So the admissible radii are 0<R<3. The radius R=3 is not admissible: the singularity 3 lies on the circle, f is unbounded there, and f is not holomorphic on any neighborhood of D(0,3).
(b) Best constant on the circle. Take ∣z∣=R<3. By the reverse triangle inequality,
∣3−z∣≥3−∣z∣=3−R>0,
with equality at z=R. Hence
∣z∣=Rmax∣f(z)∣=3−R1=:M(R).
(c) The bound and its optimization. By (i), for every R∈(0,3),
∣f(n)(0)∣≤Bn(R):=Rn(3−R)n!.
Take n≥1. To make Bn(R) as small as possible, maximize ψ(R)=Rn(3−R) on (0,3). Differentiate:
ψ′(R)=nRn−1(3−R)−Rn=Rn−1(3n−(n+1)R).
This is positive for R<R∗ and negative for R>R∗, where
R∗=n+13n∈(0,3).
So ψ attains its maximum at R∗, with value
ψ(R∗)=(n+13n)nn+13=(n+1)n+13n+1nn.
The optimized bound is therefore
∣f(n)(0)∣≤Bn(R∗)=3n+1nnn!(n+1)n+1(n≥1).
For n=0, handle the endpoint separately: B0(R)=1/(3−R) is strictly increasing on (0,3). Its infimum is 1/3 as R↓0, but no admissible positive radius attains that infimum. Taking this limit in the estimates gives ∣f(0)∣≤1/3, which equals the exact value. At every fixed admissible radius, however, B0(R)>1/3.
(d) Comparison with the exact value. By induction, f(n)(z)=(3−z)n+1n!. To see this, differentiate n!(3−z)−(n+1): the result is (n+1)!(3−z)−(n+2). Hence
f(n)(0)=3n+1n!.
The ratio of the optimized bound to the exact value is
f(n)(0)Bn(R∗)=nn(n+1)n+1=(n+1)(1+n1)n.
By Bernoulli's inequality, (1+n1)n≥1+n⋅n1=2. Also n+1≥2. So the ratio is at least 4 for every n≥1.
Some sample values:
n
R∗
optimized bound
exact f(n)(0)
ratio
1
3/2
4/9
1/9
4
2
2
1/2
2/27
27/4
Conclusion: equality is never attained for n≥1, for any admissible R. Even at the optimal radius the bound is too large by a factor of at least 4, and the excess factor grows roughly like e(n+1).
Part (iv) explains why: f is not of the form czn on any disk centered at 0.
(iii) An off-center bound.
Fix z0 with ∣z0∣≤1. Use the disk D(z0,1). Every point z of this disk satisfies ∣z∣≤∣z0∣+1≤2<3, so D(z0,1) is contained in the open set C∖{3}, on which f is holomorphic.
On the circle ∣z−z0∣=1 we have ∣z∣≤2, so
∣3−z∣≥3−∣z∣≥1,hence ∣f(z)∣≤1.
So M=1 is a valid bound on this circle. Part (i) with n=1 and R=1 gives
∣f′(z0)∣≤11!⋅1=1.
Exact comparison. Since ∣3−z0∣≥2, the exact derivative satisfies
∣f′(z0)∣=∣3−z0∣21≤41.
The estimate is valid but not sharp.
Note that the hypotheses had to be checked separately for each z0: the disk changes with z0, and the bound M is only valid on that particular circle.
(iv) The equality case.
Theorem 9.3.7
(Equality in the Cauchy estimates)
Let f be as above with M=∣z−a∣=Rmax∣f∣. Then ∣f(n)(a)∣=n!M/Rn if and only if f(z)=c(z−a)n on D(a,R) for a constant c with ∣c∣=M/Rn.
Proof
(Equality forces a monomial)
Assume M=∣z−a∣=Rmax∣f(z)∣, and use the notation of (∗).
(⇐) Monomials give equality. Suppose f(z)=c(z−a)n on D(a,R) with ∣c∣=M/Rn. On the circle ∣f∣=∣c∣Rn=M, so this M is indeed the maximum. Also f(n)(a)=n!c. Hence
∣f(n)(a)∣=n!∣c∣=Rnn!M.
(⇒) Equality forces a monomial.
Trivial case. If M=0, then f≡0 on the circle. By Cauchy's formula f≡0 on D(a,R), which is the case c=0.
Main case, M>0. Suppose ∣f(n)(a)∣=n!M/Rn. Then both inequalities in the proof of (i) are equalities:
∫02πφdt=∫02π∣φ∣dt=2πM.
Step 1: ∣φ∣ is constant. The function M−∣φ(t)∣ is continuous and nonnegative, and its integral is 0. So it vanishes identically: ∣φ(t)∣=M for all t.
Step 2: φ is constant. Let w=∫02πφdt. Then ∣w∣=2πM>0, so we can write w=∣w∣eiθ. Now
∫02πRe(e−iθφ(t))dt=Re(e−iθw)=∣w∣=∫02π∣φ(t)∣dt.
The function ∣φ∣−Re(e−iθφ) is continuous and nonnegative, and by the line above its integral is 0. So it vanishes identically, and Re(e−iθφ(t))=∣φ(t)∣ for all t.
A complex number whose real part equals its modulus is a nonnegative real number. Hence e−iθφ(t)=∣φ(t)∣=M, that is,
φ(t)≡Meiθ.
Step 3: identify f on the circle. From the definition of φ,
f(a+Reit)=Meiθeint=c(Reit)n,c=RnMeiθ.
So f(z)=c(z−a)n for every z with ∣z−a∣=R.
Step 4: extend to the disk. Let g(z)=f(z)−c(z−a)n. It is holomorphic on U and vanishes on the circle. For ∣z0−a∣<R, Cauchy's formula gives
g(z0)=2πi1∮∣z−a∣=Rz−z0g(z)dz=0.
Thus g≡0 on D(a,R), and f(z)=c(z−a)n there, with ∣c∣=M/Rn. No conclusion about other connected components of U is needed. ■
Proof X-ray
Where the equality proof gets its rigidity
The estimate loses information at exactly two places: the triangle inequality for integrals, ∣∫φ∣≤∫∣φ∣, and the pointwise bound ∣φ∣≤M. Equality forces both to be sharp, and each yields one rigid conclusion.
Sharpness of ∣φ∣≤M (Step 1) uses continuity: a continuous nonnegative function with integral 0 vanishes identically, so ∣φ∣≡M pointwise and not merely on average. Here M must be the true maximum: if M exceeded max∣f∣, then ∫∣φ∣≤2πmax∣f∣<2πM and equality could never hold.
Sharpness of the triangle inequality (Step 2) is the key move. Multiplying by the unimodular number e−iθ turns the complex equality ∣w∣=∫∣φ∣ into a statement about real parts, ∫(∣φ∣−Re(e−iθφ))dt=0, to which the same vanishing argument applies. All values of φ then have the same argument and, with constant modulus, φ is constant.
Constant φ gives f(a+Reit)=c(Reit)n: the circle values agree with c(z−a)n (Step 3). Holomorphy has not been used since the estimate itself.
Step 4 is where holomorphy on the whole disk returns: the Cauchy formula recovers f inside from its circle values, so f−c(z−a)n vanishes on the disk. This is the only step that controls the interior; values on the circle alone would not.
This confirms the conclusion of (ii)(d): 1/(3−z) is not a monomial czn, so it can never achieve equality.
Concept lensAnalytic
Equality read off the Taylor coefficients
Write f(z)=k≥0∑ck(z−a)k, so that cn=f(n)(a)/n! and the estimate says ∣cn∣Rn≤M. Parseval's identity on the circle gives
k≥0∑∣ck∣2R2k=2π1∫02πf(a+Reit)2dt≤M2.
So ∣cn∣Rn≤M again, and equality forces ck=0 for every k=n: all the mean square of f on the circle sits in one coefficient, and f=cn(z−a)n.
(v) Testing equality.
(a) p(z)=z2+z on D(0,1). The polynomial p is entire, so every closed disk is allowed.
On ∣z∣=1 we have ∣p(z)∣=∣z∣∣z+1∣=∣z+1∣≤2, with equality at z=1. So M=2.
n
Cauchy bound n!M/Rn
exact p(n)(0)
equality?
1
2
p′(0)=1
no
2
4
p′′(0)=2
no
This is consistent with (iv): p is a sum of two monomials, not a single monomial czn, so equality is impossible for every n.
(b) q(z)=3z2 on D(0,2). The polynomial q is entire. On ∣z∣=2 we have ∣q∣=12, so M=12.
n
Cauchy bound n!M/Rn
exact q(n)(0)
equality?
1
12/2=6
q′(0)=0
no
2
2⋅12/4=6
q′′(0)=6
yes
This is also consistent with (iv). Here q=cz2 with ∣c∣=3=12/22, so equality holds for n=2. For n=1, q is not of the form cz1, so the inequality is strict.
Lesson. Equality in a Cauchy estimate is a rigid condition. It occurs for the n-th derivative exactly when f is the single monomial c(z−a)n on the disk, with ∣c∣=M/Rn. If the restriction of f to this disk is not that monomial, the estimate is strict.