Necessary conditions for Convergence

Kritika
February 2021

1 Introduction
We will discuss about the necessary condition of convergence of Linear multi
step method. first we will define some basic concept which will be needed to
find necessary condition.
Definition: Given a sequence of equally spaced mesh points xn with step
size h, we consider the general linear k-step method
k
X k
X
αj yn+j = h βj f (xn+j , yn+j ) (1)
j=0 j=0

where the coefficients α0 , α1 , ..., αk and β0 , β1 , ..., βk are real coefficients such
that αk 6= 0 and both α0 , β0 are not equal to zero.
Zero Stable : A linear multi-step method is zero-stable for any ordinary
differential equation of the form y 0 = f (x, y), y(x0 ) = y0 where f satisfies the
Lipschitz condition, if and only if its first characteristic polynomial has zeros
inside the closed unit disc, with any which lie on the unit circle being simple.
This condition often called as root condition.
Where first characteristic polynomial of linear multi-step method is define
as
Xk
ρ(z) = αj z j (2)
j=0

and the second characteristic polynomial is define as

k
X
σ(z) = βj z j (3)
j=0

Consistency The numerical scheme (1) is said to be consistent with the

differential equation y 0 = f (x, y), y(x0 ) = y0 if the truncation error is such that
for any  > 0 there exists h() for which
|Tn | <  f or 0 < h < h() (4)
and any (k + 1) points (xn , y(xn )), ..., (xn+k , y(xn+k )) on any solution curve in
R of the initial value problem.

1
 where Tn is define as
1
Tn = [C0 y(xn ) + C1 hy 0 (xn ) + ...] (5)
hσ(1)
Since method is consistent i.e. Tn → 0 as h → 0 thus equation (1) is
consistent if C0 = 0 and C1 = 0 where C0 and C1 are defined as
k
X
C0 = αj = ρ(1) (6)
j=0

k
X k
X
C1 = jαj − βj = ρ0 (1) − σ(1) (7)
j=1 j=0

Convergence : The linear multistep method (1) is said to be convergent if,

for all initial value problems y 0 = f (x, y) y(x0 ) = y0 subject to the hypotheses
of Picard’s Theorem, we have that
lim yn = y(x) (8)
h→0, nh=x−x0

holds for all x ∈ [x0 , XM ] and for all solutions (yn )∞ n=0 of the difference equa-
tion (1) with consistent starting conditions, i.e. with starting conditions ys =
ηs (h), s = 0, 1, ..., k − 1, for which lim ηs (h) = y0 , s = 0, 1, ..., k − 1 .
h→0

2 Necessary conditions for Convergence

Theorem 1 A necessary condition for the convergence of the linear multi-step
method (1) is that it be zero-stable.
Proof : Let us suppose that the linear multi-step method (1) is convergent;
we wish to show that it is zero stable
whether or not a method is zero-stable can be determined by merely consid-
ering its behaviour when applied to the trivial differential equation. Thus we
will consider intial value problem,

y 0 = 0, y(0) = 0
on the interval x ∈ [0, XM ], whose solution is trivial i.e. y(x) = 0. Applying (1)
to this problem yields.
αk yn+k + αk−1 yn+k−1 + ... + α0 yn = 0 (9)
Since the method is assumed to be convergent, for any x ¿ 0,we have that,
lim yn = 0 (10)
h→0, nh=x

For all solutions of (9) satisfying, ys = ηs (h), s = 0, 1, ..., k − 1 where,

lim ηs (h) = 0, s = 0, 1, ..., k − 1 (11)
h→0

2
Let z = reiθ be a root of the first characteristic polynomial ρ(z); r ≥ 0; 0 ≤
θ < 2π
k
X
i.e. ρ(z) = αj z j = 0
j=0

k
X
ρ(z) = αj rj eijθ = 0 (12)
j=0

Claim : The number yn = hrn cos nθ satisfy equation (9) and (11)
→ substitute yn in equation (9) we get,

αk hrn+k cos (n + k)θ + αk−1 hrn+k−1 cos (n + k − 1)θ + ... + α0 hrn cos nθ = 0
eiθ +e−iθ
we know that cos θ = 2 , substitute in above equation

hrn inθ
[e (αk rk eikθ +αk−1 rk−1 ei(k−1)θ +...+α0 )+e−inθ (αk rk e−ikθ +αk−1 rk−1 e−i(k−1)θ +...+α0 )] = 0
2
hrn inθ
(e ρ(z) + e−inθ ρ(z̄)) = 0
2
Since z satisfies characteristic polynomial from (12), then its conjugate also
satisfies equation (12) Hence LHS of above equation is equal to RHS. Thus it
yn satisfies equation (9).
We can clearly see that yn = hrn cos nθ satisfies equation (11).
Thus our claim is true.
Now if θ 6= 0 and θ 6= π then

yn2 − yn−1 yn+1 h2 r2n (cos2 nθ − cos (n − 1)θ cos (n − 1)θ)

=
sin2 θ sin2 θ
h2 r2n (cos2 nθ − cos2 nθ cos2 θ + sin2 nθ sin2 θ)
=
sin2 θ
h2 r2n (cos2 nθ sin2 θ + sin2 nθsin2 θ)
=
sin2 θ
= h2 r2n
Since the left-hand side of this identity converges to 0 as h → 0, n → ∞, nh = x
the same must be true of the right-hand side; therefore,
x
lim ( )2 r2n = 0
n→∞ n
This implies that r ≤ 1. for θ = 0 and θ = π yn → 0 ,if r ≤ 1 In other words,
we have proved that any root of the first characteristic polynomial of (9) lies in
the closed unit disc.

3
 Next we prove that any root of the first characteristic polynomial of (9) that
lies on the unit circle must be simple.
Assume, instead, that z = reiθ , is a multiple root of ρ(z),with |z| = 1 (and
therefore r = 1) i.e. ρ0 (z) = 0 and 0 ≤ θ < θ
k
X
ρ0 (z) = jαj z j−1 = 0 (13)
j=1

We shall prove below that this contradicts our assumption that the method
(9) is convergent.
Let yn = h1/2 nrn cos nθ, substitute yn into equation (9),

αk h1/2 (n+k)rn+k cos (n + k)θ+αk−1 h1/2 rn+k−1 (n+k−1) cos (n + k − 1)θ+...+α0 nh1/2 rn cos nθ = 0
eiθ +e−iθ
we know that cos θ = 2 , substitute in above equation

h1/2 rn inθ
[e (αk (n+k)rk eikθ +...+α0 n)+e−inθ (αk (n+k)rk e−ikθ +...+α0 n)] = 0
2
h1/2 rn inθ
[e (nρ(z) + rρ0 (z)) + e−inθ (nρ(z̄) + rρ0 (z̄))] = 0
2
Hence, from equation (12) and (13), LHS is equal to RHS. i.e. yn = nh1/2 rn cos nθ
is solution of (9).
It also satisfy equation(11) for,

|ηs (h)| = |ys | ≤ h1/2 s ≤ h1/2 k − 1, s = 0, 1, ..., k − 1.

If θ = 0 and θ = π we have,

|yn | = x1/2 n1/2 rn

we can deduce from that,

lim |yn | = ∞
n→∞

which contradict equation(10). Since, here yn is independent of h.

Now, if θ 6= 0 and θ 6= π. Let zn = n−1 h−1/2 yn = h1/2 x−1 yn = rn cos nθ

zn2 − zn−1 zn+1 cos2 nθ − cos (n + 1)θ cos (n − 1)θ

2 = r2n = r2n (14)
sin θ sin2 θ
Since by (10) lim zn = 0 as zn is independent of h.
n→∞
it follows that the left-hand side of (14) converges to 0 as n → ∞ But then
the same must be true of the right-hand side of (14);however, the right-hand side
of (14) cannot converges to 0 as n → ∞ since |r| = 1. Thus again it contradict.
Hence if first characteristic root lie on unit circle then it is simple.
i.e. if LMM converges then it satisfies root condition means it is zero stable.
Theorem 2 A necessary condition for the convergence of the linear multi-step
method (9) is that it is consistent.

4
 Proof : Let us suppose that the linear multi-step method (1)is convergent;
we wish to show that it is then consistent.
i.e. from definition of Consistency we have to show that C0 = 0 and
C1 = 0.
First we show that C0 = 0. We consider initial value problem y 0 = 0, y(0) =
1, whose exact solution is y(x)=1.On the interval [0, XM ], XM > 0. Applying
(1)to this IVP we get,
αk yn+k + αk−1 yn+k−1 + ... + α0 yn = 0 (15)
We chose ys = 1, s = 0, 1, ..., k − 1. Since method is convergent.we deduce
that,
lim yn = 1 (16)
h→0, nh=x

Since in the present case yn is independent of the choice of h,(16) is equivalent

to saying that
lim yn = 1 (17)
n→∞

Passing to the limit n → ∞ in (15), we deduce that,

αk + αk−1 + ... + α0 = 0 (18)
Hence by definition of C0 , we get
C0 = 0
In order to show that C1 = 0,
we now consider the initial value problem y 0 = 1, y(0) = 0,on the interval
[0, XM ], XM > 0 whose solution is y(x)=1. Applying (1) on this problem. we
get,
αk yn+k + αk−1 yn+k−1 + ... + α0 yn = h(βk + βk−1 + ... + β0 ) (19)
Where XM − x0 = XM − 0 = N h and 1 ≤ n ≤ N − k For a convergent method
every solution of (19) satisfying,
lim ηs (h) = 0, s = 0, 1, ..., k − 1 (20)
h→0

where ys = ηs (h), s = 0, 1, ..., k − 1, must also satisfy,

lim yn = x (21)
h→0, nh=x

Since according to the previous theorem zero-stability is necessary for conver-

gence, we may take it for granted that the first characteristic polynomial ρ(z) of
the method does not have a multiple root on the unit circle. |z| = 1 ,therefore
ρ0 (z) = kαk + (k − 1)αk−1 + ... + 2α2 + α1 6= 0 (22)
Let us define a sequence (yn )∞
n=0 by yn = Knh. where,

βk + βk−1 + ...β0
K= (23)
kαk + ... + 2α2 + α1

5
this sequence clearly satisfies (20)and is the solution of (19) since, substitute
the value of yn in equation (19).

Kh(αk (n + k) + αk−1 (n + k − 1) + ... + α0 n = h(βk + ... + β0 )

using condition C0 = 0,we get

Kh(αk k + αk−1 (k − 1) + ... + α0 n = h(βk + ... + β0 )

by definition of K, above equality is true.

Thus from equation (21), we have

x = y(x) = lim yn = lim Knh = Kx

h→0, nh=x h→0, nh=x

Thus, K=1 substitute value of K in equation (23), we get

C1 = (kαk + ... + 2α2 + α1 ) − (βk + βk−1 + ... + β0 ) = 0

Thus C1 = 0.
Hence LMM is consistent if it is convergent.

6
Lemmas Required for The Study of Sufficient Conditions for
Convergence of Linear Multi-step Methods

Abhinav Das, Student in Department of Mathematics,

South Asian University

Introduction
We consider a linear multi-step method
k
X k
X
αj yn+j = h βj f (xn+j , yn+j ) (1)
j=0 j=0

and we study two lemmas required for the study of sufficient conditions for the convergence of the
linear multi-step method (1). We have already discussed its zero-stability, consistency, and some
necessary conditions required for the convergence of the method but we discuss these properties very
briefly again here as preliminaries and we discuss the lemmas in detail.

Preliminaries
1. Zero-Stability: A linear k-step method for the ordinary differential equation y 0 = f (x, y) is
said to be zero-stable if there exist a constant K such that for any two sequences (yn ) and (ŷn )
which have been generated by the same formulae but different initial value y0 , y1 , ....., yk−1 and
ŷ0 , ŷ1 , ....., ŷk−1 , respectively, we have
|yn − ŷn | ≤ K max {|y0 − ŷ0 | , |y1 − ŷ1 | , ....., |yk−1 − ŷk−1 |} (2)
xn −x0
for xn ≤ XM and as h tends to 0. Where, XM ∈ [x0 , XM ] and h = n

• Theorem: Let us consider a ordinary differential equation of the form y 0 = f (x, y) such
that f satisfies the lipschitz condition
|f (x, y) − f (x, z)| ≤ L |y − z| (3)
where (x, y) and (x, z) lie in the rectangle R = {(x, y) : x0 ≤ x ≤ XM , |y − y0 | ≤ YM }
where XM > x0 and YM are constants. Here L is called lipschitz constant. Then a linear
multi-step method is zero-stable if and only if its first characteristics polynomial
k
X
ρ(z) = αj z j (4)
j=0

has zeros inside the closed unit disc, with any which lie on the unit circle are simple.
2. Consistency: A linear method (1) is said to be consistent with the differential equation y 0 =
f (x, y) if truncation error Tn given by
Pk 0
j=0 [αj y (xn+j ) − hβy (xn+j )]
Tn = Pk (5)
h j=0 βj

1
 is such that for any  > 0 there exists h() for which |Tn < | for 0 < h < h() and any (k + 1)
points (xn , y(xn )) , ....., (xn+k , y(xn+k )) on any solution curve in R of the initial value problem
y 0 = f (x, y), y(x0 ) = y0 .
3. Convergence: The linear multi-step method (1) is said to be convergent if the initial value
problem y 0 = f (x, y), y(x0 ) = y0 where f is continuous in rectangle R and satisfies the
lipschitz condition, we have
lim yn = y(x) (6)
h→0,nh=x−x0

N
holds ∀ x ∈ [x0 , XM ] and for all solutions {yn }n=0 of the difference equation (1) with consistent
starting conditions, i.e. with starting conditions ys = ηs (h), s = 0, 1, ...., k − 1, for which
limh→0 ηs (h), s = 0, 1, ....., k − 1.
• Necessary conditions for convergence
– Theorem: A necessary condition for the convergence of the linear multi-step method
(1) is that it be zero-stable.
– Theorem: A necessary condition for the convergence of the linear multi-step method
(1) is that it be consistent.
4. Holomorphic Function: A complex valued function f (z) is called holomorphic function if it is
differentiable with respect to z at every point of domain of the function.
5. Meromorphic Function: A function f is meromorphic in its doamin if every point of domain
has a neighbourhood such that either f or 1/f is holomorphic in it.
6. Zero and Pole: A zero of a meromorphic function f is a complex number z such that f (z) = 0.
A pole of f is a zero of 1/f .
7. (P1) Theorem : If a ∈ Ω where Ω is a region and f ∈ H(Ω − {a}), one of the following three
cases must occur
(a) f has removable singularity at a.
(b) There are
Pmcomplex numbers c1 , ...., cm where m is positive integer and cm 6= 0 such that
ck
f (z) = k=1 (z−a) k has removable singularity at a.

(c) If r > 0 and D(a; r) ⊂ Ω then f (D0 (a; r)) is dense in the plane.

2
Lemma 1:
Statement:
Suppose that all roots of the polynomial ρ(z) = αk z k + .... + α1 z + a0 lie in the closed unit disk
|z| ≤ 1 and those which lie on the unit circle |z| = 1 are simple. Assume further that the numbers
γl , l = 0, 1, 2, ...., are defined by
1
= γ0 + γ1 z + γ2 z 2 + ..... (7)
αk + .... + α1 z k−1 + α0 z k
Then, Γ ≡ supl≥0 |γl | < ∞.

Proof:
Given that ρ(z) = αk z k + .... + α1 z + α0

∴ ρ( z1 ) = αk z1k + .... + α1 z1 + α0

⇒ z k ρ( z1 ) = αk + .... + α1 z k−1 + α0 z k

We define ρ̂(z) = z k ρ( z1 ) and according to the statement of the lemma 1, the roots of the poly-
nomial ρ(z) = αk z k + .... + α1 z + α0 lie in the closed unit disk |z| ≤ 1 and the roots which lie on the
unit circle |z| = 1 are simple. We can see from the definition of poles and zeros of the polynomial,
1
that the function ρ̂(z) is holomorphic in the open disc |z| < 1. Since the roots z1 , z2 , ...., zm of ρ(z)
1
on |z| = 1 are simple, the poles of ρ̂(z) are also simple. Now we consider a theorem whose statement
will be useful for our proof. [Let G be a region let f : G → C be analytic function. Suppose
lim supz→a |f (z)| ≤ M ∀δ∞ G f or some M > 0 then |f (z)| ≤ M ∀z ∈ G ] Using this theorem and
theorem p1, now we preoceed and there exist constants A1 , A2 , ...., Am such that
1 A1 Am
f (z) = − − .... − (using theorem p1) (8)
ρ̂(z) z − z11 z − z1m
is holomorphic for |z| < 1 |f (z)| ≤ M ∀ |z| ≤ 1.
The Taylor’s series expansion of of f (z) at z = 0 is given by
1 2 00 1 n (n)
f (z) = f (0) + zf 0 (0) + z f (0) + ..... + z f (0) + ...... (9)
2! n!
We now use the Cauchy’s estimate [ If f is a holomorphic function in the open disc D(a; r), center
a and radius r, and |f (z)| ≤ M ∀z ∈ D(a; r) then |f (a) (a)| ≤ M ( Rnn! ), n = 0, 1, 2, ....] and see that
the coefficient of Taylor’s expansion of f at z = 0 forms a bounded sequence. As
∞
Ai X
− = A i zil z l , i = 1, 2, ...., m (10)
z − z1i i=0
1
and |z| ≤ 1, we get (using 8), the coefficient in the Taylor’s series expansion of ρ̂(z) form a
bounded sequence. That is supl≥0 |γl | < ∞, l = 0, 1, 2, ....,
This completes the proof.

3
Lemma 2:
Statement:
Let the difference equation

αk em+k + αk−1 em+k−1 + ..... + α0 e0 = h(βk,m em+k

(11)
+βk−1,m em+k−1 + .... + β0,m em ) + λm

be given.
Suppose that all roots of the polynomial ρ(z) = αk z k + .... + α1 z + α0 lie in the closed unit disk
|z| ≤ 1 and those which lie on the unit circle |z| = 1 are simple. Let B ∗ and Λ denote non-negative
constants and β a positive constant such that

|βk,n | + |βk−1,n | + |βk−2,n | ..... + |β0,n | ≤ B ∗

(12)
|βk,n | ≤ β, |λn | ≤ Λ, n = 0, 1, 2, ...., N,

and let 0 ≤ h < |αk | β −1 . Then every solution of the difference equation (4) for which

|es | ≤ E, s = 0, 1, ...., k − 1, (13)

satisfies
|en | ≤ K ∗ exp(nhL∗ ), n = 0, 1, 2, ...., N (14)
where
Γ
L∗ = Γ∗ B ∗ , K ∗ = Γ( ∗) (N A + AEK) , Γ∗ =   (15)
−1
1 − h |αk | β

Γ is same as in lemma 1 and

A = |αk | + |αk−1 | + .... + |α0 | (16)

Proof:
For a fixed k we consider the numbers γl , l = 0, 1, 2, ..., n − k, as defined in the lemma 1. Multiplying
both side of the equation (5) for m = n − k − l by γl , l = 0, 1, 2, ..., n − k and take the sum of the
resulting equations. We denote the sum by Sn and after some simplifications in that sum we get

Sn = (αk en + αk−1 en−1 + .... + α0 en−k ) γ0

+ (αk en−1 + αk−1 en−2 + .... + α0 en−k−1 ) γ1 + ....
+ (αk ek + αk−1 ek−1 + .... + α0 e0 ) γn−k
∴ Sn = αk γ0 en + (αk γ1 + αk−1 γ0 ) en−1 + .... (17)
+ (αk γn−k + αk−1 γn−k−1 + .... + α0 γn−2k ) ek
+ (αk−1 γn−k + .... + α0 γn−2k+1 ) ek−1 + ....
+ αk0 γn−k e0

We define γl = 0 for l < 0 and see that


 1 f or l=0
αk γl + αk−1 γl−1 + .... + α0 γl−k = (18)
0 f or l>0


we have seen that

Sn = en + (αk−1 γn−k + .... + α0 γn−2k+1 ) ek−1 + .... + α0 γn−k e0 (19)

4
By manipulating similarly the right-hand side in the sum, we find that,

en + (αk−1 γn−k + .... + α0 γn−2k+1 ) ek−1 + .... + α0 γn−k e0

= h[βk,n−k γ0 en + (βk−1,n−1 γ0 + βk,n−k−1 γ1 ) en−1 + ....
(20)
+ (β0,n−k γ0 + ..... + βk,n−2k γk ) en−k + ..... + β0,0 γn−k e0 ]
+ (λn−k γ0 + λn−k−1 γ1 + .... + λ0 γn−k )

Thus, by our assumption and noting that by (12) γ0 = αk−1 ,

n−1
X
|en | ≤ hβ αk−1 |en | + hΓB ∗ |em | + N ΓΛ + AΓEk. (21)
m=0

Hence,
n−1
X
1 − hβ αk−1 |en | ≤ hΓB ∗


|em | + N ΓΛ + AΓEk. (22)
m=0

From the definition of L∗ and K ∗ as stated earlier we can rewrite the last inequality as follows:
n−1
X
|en | ≤ K ∗ + hL∗ |em | , n = 0, 1, 2, ......, N (23)
m=0

The final estimate is deduced from (17) by induction. First, we note that by (11), AΓ ≥ 1. Conse-
quently, K ∗ ≥ ΓAEK ≥ EK ≥ E. Now, letting n = 1 in (17),

|e1 | ≤ K ∗ + hL∗ |e0 | ≤ K ∗ + hL∗ E ≤ K ∗ (1 + hL∗ ) (24)

Repeating this, we get

m
|em | ≤ K ∗ (1 + hL∗ ) , m = 0, 1, 2, ...., k − 1 (25)

Now we suppose that the inequality has already been shown to hold for m = 0, 1, 2, ....., n − 1, where
n ≥ k; we shall prove that it then also holds for m = n, which will complete the induction. Indeed,
from (17) that
n
(1 + hL∗ ) − 1
|en | ≤ K ∗ + hL∗ K ∗ = K ∗ (1 + hL∗ ) (26)
hL∗
∗
Further, as 1 + hL∗ ≤ ehL we have from (19) that
∗
|en | ≤ K ∗ ehL n , n = 0, 1, 2, ......, N (27)

This completes the proof of lemma 2.

Note that (17) ⇒ (21) is usually referred to as the Discrete Gronwall Lemma
Lemma 2 is used prove that the zero stability and consistency, which is necessary condition for
the convergence of the multi-step methods are also the sufficient condition for the convergence.

5
Advanced Numerical Techniques for Differential
Equations (Theorem 9 and Theorem 10)

Nabaraj Adhikari
February 6, 2021
 1

Definition 0.1. A function f (x, y) is said to satisfies Lipschitz condition in the

second component on the rectangle R if there exist L > 0 such that

(1) |f (x, y) − f (x, z)| ≤ L|y − z| ∀(x, y), (x, z) ∈ R

Lemma 0.2 (2). Suppose that all roots of the polynomial ρ(z) = αk z k + · · · + α0 lie
in the closed unit disk |z| ≤ 1 and those which lie on the unit circle |z| = 1 are
simple. Let B ∗ and L denote non-negative constants and β a positive constant such
that
|βk,n | + |βk−1,n | + · · · + |β0,n | ≤ B ∗ , |βk,n | ≤ β
, |λn | ≤ Γ, n = 0, 1, · · · , N and let 0 ≤ h ≤ |αk |β −1 . Then every solution of
αk em+k + αk+1 em+k−1 +· +α0 e0 = h(βk,m em+k + βk−1,m em+k−1 + · · · + β0,m em ) + λm
for which
|es | ≤ E
for s = 0, 1, · · · k − 1 satisfies |en | ≤ exp(nhL∗ ), n = 0, 1, · · · , N where L∗ = Γ∗ B ∗ ,
K ∗ = Γ∗ (N Λ + AEk), Γ∗ = (1−h|αΓk |−1 B) , Γ = sup |γl | < ∞, A = |αk | + |αk−1 | +
1
· · · + |α0 | and γl = αk +···+α 0z
k.

Theorem 0.3. For a linear multi-step method that is consistent with the ordinary
differential equation y 0 = f (x, y), y(x0 ) = y0 where f is assumed to satisfy a Lipschitz
condition, and starting with consistent starting conditions, zero-stability is sufficient
for convergence.

Proof. Let us define

δ = δ(h) = max |ηs (h) − y(a + sh)|

0≤s≤k−1

given that the starting conditions ys = ηs (h), s = 0, · · · , k − 1 are assumed to be

consistent, we have that lim δ(h) = 0. We have to prove that
n→0

lim yn = y(x)
n→∞

where nh = x−x0 for all x in the interval [x0 , XM ]. We begin the proof by estimating
the truncation error:

k
1 X
Tn = [ αj y(xn+j ) − hβj y 0 (xn+j )]
hσ(1) j=0

As y 0 ∈ [x0 , XM ], it makes sense to define , for  ≥ 0, the function

χ() = max
∗
|y 0 (x∗ ) − y 0 (x)|
|x −x|≤
 2

for x, x∗ ∈ [x0 , XM ]. For s= 0, 1, · · · , k − 1, we can write

y 0 (xn+s ) = y 0 (xn ) + θs χ(sh)

where |θs | ≤ 1. Furthermore, by the Mean Value Theorem there exist ξs ∈ (xn , xn+s )
such that
y(xn+s ) = y(xn ) + shy 0 (ξs )
Thus
y(xm+s ) = y(xm ) + sh[y 0 (xm ) + θs0 χ(sh)]
where |θs0 | ≤ 1

|σ(1)Tn | ≤ |h−1 (α1 + α2 + · · · + αk )y(xn )

+(α1 + 2α2 + · · · + kαk )y 0 (xn )| − |(β0 + β1 + · · ·
+ βk )y 0 (xn )|
+ (|α1 | + 2|α2 | + · · · )
+ k|αk |)|χ(kh)| + (|β0 | + |β1 | + · · · + |βk |)|χ(kh)|

Since the method has been assumed consistent, the first, second and third terms on
the right-had side cancel, giving

|σ(1)Tn | ≤ (|α1 | + 2|α2 | + · · · )

+ k|αk |)|χ(kh)| + (|β0 | + |β1 | + · · · + |βk |)|χ(kh)|

Thus

(2) |σ(1)Tn | ≤ K|χ(kh)|

where K = (|α1 | + 2|α2 | + · · · + k|αk |) + (|β0 | + |β1 | + · · · + |βk |) we conclude that

the global error em = y(xm ) − ym satisfies αk em+k + · · · + α0 e0 = h(βk gm+k em+k +
· · · + β0 gm em ) + σ(1)Tn h where
(
[f (xm , y(xm )) − f (xm , ym )]/em , em 6= 0
gm =
0 , em = 0

By virtue of (2), we then have that αk em+k + · · · + α0 e0 = h(βk gm+k em+k + · · · +

β0 gm em ) + θKχ(kh)h. As f is assumed to satisfy the Lipschitz condition (1) we
have that |gm | ≤ L, m = 0, 1, · · · , On applying Lemma 2 with E = δ(h), Λ =
Kχ(kh)h, N = XMh−x0 , B ∗ = BL where B = |β0 | + |β1 | + · · · + |βk |, we find that

(3) |en | ≤ Γ∗ [Akδ(h) + (XM − x0 )Kχ(kh) exp[(xn − x0 )LΓ∗ B]]

 3

where |A| = |α0 | + |α1 | + · · · + |αk | and Γ∗ = Γ

1−h|α−1
k βk |L

Now, y 0 is a continuous function on the closed interval [x0 , XM ]; therefore it

is uniformly continuous on [x0 , XM ]. Thus, χ(kh) → 0 as h → 0; also, by virtue of
the assumed consistency of the starting values, δ(h) → 0 as h → 0. Passing to the
limit h → 0 in (3), we deduce that for x − x0 = nh

lim |en | = 0 equivalently lim |y(x) − y0 | = 0

n→∞ n→∞

so the method is convergent.

Theorem 0.4. For a linear multi-step method that is consistent with the ordinary
differential equation y 0 = f (x, y), y(x0 ) = y0 where f is assumed to satisfy a Lipschitz
condition, and starting with consistent initial data, zero-stability is necessary and
sufficient for convergence. Moreover if the solution y(x) has continuous derivative
of order (p + 1) and truncation error O(hp ), then the global error en = y(xn ) − yn is
also O(hp ).
 Maximum order of a zero-stable linear
multi-step method

February 7, 2021

1 IVP
The first order differential equation of the form

y 0 = f (x, y) (1)

with initial condition

y(x0 ) = y0 (2)

is called an initial value problem.

For a given sequence of equally spaced mesh points (xn ) with step size h, the
general linear k - step method is given by-

k
X k
X
αj yn+j = h βj f (xn+j , yn+j ) (3)
j=0 j=0

where the coefficients α0 , α1 , . . . ,αk and β0 , β1 , . . . ,βk are real constants.

Definition A linear k-step method for the ordinary differential equation
y 0 = f (x, y) is said to be zero-stable if there exists a constant K such that
for any two sequences (yn ) and (yˆn ) which have been generated by the same
formulae but different initial data y0 , y1 , . . . ,yk−1 and yˆ0 , ŷ1 , . . . , ŷk−1

1
respectively, we have

|y − yˆn | ≤ K max{|y0 − yˆ0 |, |y1 − ŷ1 |, . . . , |yk−1 − ŷk−1 |} (4)

for xn ≤ XM and h tends to 0.

For the linear k-step method, we consider its first and second characteristics
polynomial, respectively
k
X
ρ(z) = αj z j (5)
j=0

k
X
σ(z) = βj z j (6)
j=0

where αk 6= 0, α0 2 + β0 2 6= 0
The truncation error of (3) is defined as follows:
Pk
j=0 [αj y(xn+j ) − hβj y 0 (xn+j )]
Tn = (7)
h kj=0 βj
P

Definition The numerical scheme (3) is said to be consistent with the dif-
ferential equation (1) if the truncation error defined by (7) is such that for
any  > 0 there exists h() for which
Tn <  for 0 < h < h() and any (k + 1) points (xn , y(xn )), ...(xn+k , y(xn+k ))
on any solution curve in R of the IVP(1 − 2).
now let us suppose that the solution to the differential equation equation
is sufficiently smooth and let us expand y(xn+j ) and y 0 (xn+j ) into a Taylor
series about the point xn and substitute these expansion into the numerator
of (7) to obtain

1 
C0 y(xn ) + C1 hy 0 (xn ) + C2 h2 y 00 + . . .

Tn = (8)
hσ(1)
where

2
 k
X
C0 = αj (9)
j=0
k
X k
X
C1 = jαj − βj (10)
j=0 j=0
k k
X j2 X
C2 = αj − jβj (11)
j=0
2! j=0

similarly

k k
X jq X j q−1
Cq = αj − βj (12)
j=0
q! j=0
(q − 1)!
for consistency we need that Tn → 0 as h → 0 and this requires that C0 = 0
and C1 = 0
In terms of characteristics polynomials this consistency requirement can be
restated in compact form as ρ(1) = 0 and ρ0 (1) = σ(1) 6= 0.
Definition The numerical method (4) is said to be of order of accuracy p if
and only if C0 = C1 = · · · = Cp = 0 and Cp+1 6= 0
in this case,

Cp+1 p (p+1)
Tn = hy (xn ) + O(hp+1 ) (13)
σ(1)
the number Cp+1 (6= 0) is called the error constant of the method.

Theorem 1. A linear multi-step method is zero stable for any differential

equation of the form (1) where f satisfies the Lipschitz condition if and only
if its first characteristics polynomial has zeros inside the closed unit disc,
with any which lie on the unit circle being simple.

Definition The roots of the first characteristic polynomial lie in the

closed unit disc and those on the unit circle are simple is called root condition.

3
Theorem 2. A linear multi-step method is zero stable for any differential
equation of the form (1) where f is assumed to satisfies the Lipschitz condi-
tion, and starting with consistent initial data, zero-stability is necessary and
sufficient for con- vergence. Moreover if the solution y(x) has continuous
derivative of order (p + 1) and truncation error O(hp ), then the global error
en = y(xn ) − yn is also O(hp )

Let us suppose that we have already chosen the coefficients αj , j = 0, .

. . , k, of the k-step method (3). The question we shall be concerned with
in this section is how to choose the coefficients βj , j = 0, . . . , k, so that
the order of the resulting method (3) is as high as possible.
In view of Theorem 2 we shall only be interested in consistent methods, so
it is natural to assume that the first and second characteristic polynomials
ρ(z) and σ(z) associated with (3) satisfy ρ(1) = 0, ρ0 (1) − σ(1) = 0, with
σ(1) 6= 0
Now consider the function φ of the complex variable z defined by

ρ(z)
φ(z) = − σ(z) (14)
logz
the function log(z) appearing in the denominator is made single-valued by
cutting the complex plane along the half plane Re ≤ 0
Now we begin our analysis with the following fundamental lemma.
Lemma 3 Suppose that p is a positive integer, the linear multi-step method
(4) with stability polynomials ρ(z) and σ(z) is of order of accuracy p if and
only if the function φ(z) defined by (14) has a zero of multiplicity p at z = 1
Proof : Let us suppose that the k-step method (3) for the numerical ap-
proximation of the initial value problem (1 − 2) is of order p.
Then, for any sufficiently smooth function f (x, y), the resulting solution to
(1 − 2) yields a truncation error of the form:

Cp+1 p (p+1)
Tn = hy (xn ) + O(hp+1 ) (15)
σ(1)

4
as h → 0, Cp+1 6= 0, xn = x0 + nh
In particular , for the initial value problem
y 0 = y, y(0) = 1
we get,
enh
Tn = [ρ(eh ) − hσ(eh )] (16)
hσ(1)
Substituting ρ(eh ) and σ(eh ) we get

enh Cp+1 p
Tn = [ρ(eh ) − hσ(eh )] = enh h + O(hp+1 ) (17)
hσ(1) σ(1)

as h → 0, Cp+1 6= 0. Thus the function

f (h) = h1 [ρ(eh ) − hσ(eh )]

is holomorphic in a neighbourhood of h = 0 and has a zero of order p at
h = 0. The function z = eh is a bijective mapping of a neighbourhood of
h = 0 onto a neighbourhood of z = 1. Therefore φ(z) is holomorphic in a
neighbourhood of z = 1 and has a zero of multiplicity p at z = 1.
Conversely, Suppose that φ(z) has a zero of multiplicity p at z = 1. Then
f (h) = φ(eh ) is a holomorphic function in the vicinity of h = 0 and has a
zero of multiplicity p at h = 0. Therefore,
g(h) = kj=0 (α =j −hβj)ejh
P

has a zero of multiplicity (p + 1) at h = 0,implying that g(0) = g 0 () = ... =

g p (0) = 0, but g p+1 (0) 6= 0, first
g(0) = 0 = kj=0 αj = C0
P

Now, by successive differentiation of g with respect to h,

g 0 (0) = 0 = kj=0 (jαj − βj ) = C1
P

Pk
g 0 (0) = 0 = j=0 (jαj − βj ) = C 1

00 Pk 2
g (0) = 0 = j=0 (j αj − 2jβj ) = 2C2

5
 000 Pk 3
g (0) = 0 = j=0 (j αj − 3j 2 βj ) = 6C3
Similarly,
Pk
g p (0) = 0 = j=0 (j
p
αj − pj p−1 βj ) = p!Cp

From here we can see that C0 = C1 = C2 = ... = Cp = 0; since g p+1 6= 0 we

have that Cp+1 6= 0, consequently (3) is of order of accuracy p.

Now we will use this lemma in the next theorem to supply a lower bound
for the maximum order of a linear multistep method with prescribed first
stability polynomial ρ(z).

Theorem 3. Suppose that ρ(z) is a polynomial of degree k such that ρ(1) = 0

and ρ0 (1) 6= 0 and let k̂ be an integer such that 0 ≤ k̂ ≤ k. then there exists a
unique polynomial σ(z) of degree k̂ such that ρ0 (1) − σ(1) = 0 and the order
of the multi-step method associated with ρ(z) and σ(z) is ≥ k̂ + 1.

ρ(z)
Proof : Since the function log(z)
is holomorphic in the neighbourhood of
z = 1 it can be expanded into a convergent Taylor series:

ρ(z)
log(z)
= C0 + C1 (z − 1) + C2 (z − 1)2 + ...
On multiplying both sides by log z, we get
ρ(z)
log(z)
log(z) = log(z)[C0 + C1 (z − 1) + C2 (z − 1)2 + ...]
On differentiating and deducing in the nbhd of z = 1, we deduce
C0 = ρ0 (1)(6= 0)

Now let’s define

σ(z) = c0 + c1 (z − 1) + ... + ck̂ (z − 1)k̂
when z = 1, clearly we have,

6
σ(1) = c0 = ρ0 (1)(6= 0).
ρ(z)
With this definition φ(z) = log(z)
− σ(z) = Ck̂+1 (z − 1)k̂+1 + ...,
and therefore φ(z) has a zero at z = 1 of multiplicity not less than k̂ + 1. By
lemma 3 the linear k-step method associated with ρ(z) and σ(z) is of order
> k̂ + 1.
The uniqueness of σ(z) possessing the desired properties follows from the
uniqueness of the Taylor series expansion of φ(z) about the point z = 1.
We note in connection with this theorem that for most methods of practical
interest either k̂ = k − 1 resulting in an explicit method or k̂ = k correspond-
ing to an implicit method.

7
 Example 5 and Theorem 12
Advanced Numerical Techniques for Differential Equations
Course Instructor: Prof. Kapil Sharma

Submitted By:
Probhat Pachung
Research Scholar
(1st semester)
South Asian University
Department of Mathematics
(2020-2021)
 Outline

1. Notation and Preliminaries.

2. Example 5
3. Theorem 12
4. References
1.Notation and Preliminaries:
The first order differential equation of the form

y 0 = f (x, y ) (1)

with initial condition

y (x0 ) = y0 (2)
is called an initial value problem.
We have already discussed by considering three consecutive points
xn−1 , xn = xn−1 + h, xn+1 = xn−1 + 2h, integrating the differential
equation between xn−1 and xn+1 and applying Simpson’s rule to
approximate the resulting integral yields
Z xn+1
y (xn+1 ) = y (xn−1 ) + f (x, y (x))dx
xn−1
1
≈ y (xn−1 ) + [f (xn−1 , y (xn−1 )) + 4f (xn , y (xxn )) + f (xn+1 , y (xn+
3
which leads to the method
1
yn+1 = y (xn−1 )+ [f (xn−1 , y (xn−1 ))+4f (xn , y (xxn ))+f (xn+1 , y (xn+1 ))]
3
(3)
Given a sequence of equally spaced mesh points (xn ) with step size
h, the general linear k -step method is given by-
k
X k
X
αj yn+j = h βj f (xn+j , yn+j ) (4)
j=0 j=0

where the coefficients α0 , α1 , . . . ,αk and β0 , β1 , . . . ,βk are real

constants.
Definition: A linear k-step method for the ordinary differential
equation y 0 = f (x, y ) is said to be zero-stable if there exists a
constant K such that for any two sequences (yn ) and (yˆn ) which
have been generated by the same formulae but different initial data
y0 , y1 , . . . ,yk−1 and yˆ0 , ŷ1 , . . . , ŷk−1 respectively, we have
|y − yˆn | ≤ K max{|y0 − yˆ0 |, |y1 − ŷ1 |, . . . , |yk−1 − ŷk−1 |} (5)
for xn ≤ XM and h tends to 0.
Given the linear k-step method, we consider its first and second
characteristics polynomial, respectively

ρ(z) = kj=0 αj z j
P

σ(z) = kj=0 βj z j
P
The truncation error of (4) is defined as follows:
Pk 0 (x
j=0 [αj y (xn+j ) − hβj y n+j )]
Tn = (6)
h kj=0 βj
P

Definition: The numerical scheme (4) is said to be consistent with

the differential equation (1) if the truncation error defined by (6) is
such that for any  > 0 there exists h() for which
|Tn | <  for 0 < h < h()
and any (k + 1) points (xn , y (xn )), ...(xn+k , y (xn+k )) on any
solution curve in R of the IVP(1 − 2).
now let us suppose that the solution to the differential equation
equation is sufficiently smooth and let us expand y (xn+j ) and
y 0 (xn+j ) into a Taylor series about the point xn and substitute these
expansion into the numerator of (6) to obtain
1 
C0 y (xn ) + C1 hy 0 (xn ) + C2 h2 y 00 + . . .

Tn = (7)
hσ(1)
where
k
X
C0 = αj
j=0

k
X k
X
C1 = jαj − βj
j=0 j=0
 k k
X j2 X
C2 = αj − jβj
2!
j=0 j=0

etc.

k k
X jq X j q−1
Cq = αj − βj
q! (q − 1)!
j=0 j=0

for consistency we need that Tn → 0 as h → 0 and this requires

that C0 = 0 and C1 = 0
In terms of characteristics polynomials this consistency requirement
can be restated in compact form as ρ(1) = 0 and ρ0 (1) = σ(1) 6= 0.
Definition: The numerical method (4) is said to be of order of
accuracy p if and only if C0 = C1 = · · · = Cp = 0 and Cp+1 6= 0
in this case,

Cp+1 p (p+1)
Tn = h y (xn ) + O(hp+1 )
σ(1)
the number Cp+1 (6= 0) is called the error constant of the method.
Theorem: A linear multi-step method is zero stable for any
differential equation of the form (1) where f satisfies the Lipschitz
condition if and only if its first characteristics polynomial has zeros
inside the closed unit disc, with any which lie on the unit circle
being simple.
Definition: The roots of the first characteristic polynomial lie in
the closed unit disc and those on the unit circle are simple is called
root condition.
now consider the function φ of the complex variable z defined by

ρ(z)
φ(z) = − σ(z) (8)
logz
the function log (z) appearing in the denominator is made
single-valued by cutting the complex plane along the half plane
Re ≤ 0
Lemma: Suppose that p is a positive integer, the linear multi-step
method (4) with stability polynomials ρ(z) and σ(z) is of order of
accuracy p if and only if the function φ(z) defined by (8) has a zero
of multiplicity p at z = 1
The lower bound for the maximum order of a linear multi-step
method with prescribed first stability polynomial ρ(z) is given by
the following theorem.
Theorem: Suppose that ρ(z) is a polynomial of degree k such that
ρ(1) = 0 and ρ0 (1) 6= 0 and let k̂ be an integer such that
0 ≤ k̂ ≤ k. then there exists a unique polynomial σ(z) of degree k̂
such that ρ0 (1) − σ(1) = 0 and the order of the multi-step method
associated with ρ(z) and σ(z) is ≥ k̂ + 1.
2.Example 5:
[This is the example which gives you the idea of the upper
bound for the order of the zero-stable linear k− step
method.]
Consider a linear two step method with ρ(z) = (z − 1)(z − λ)
the method will be zero stable as long as λ ∈ [−1, 1).
[ WHY?; By the mean zero stability the roots of ρ(z) lies inside the
closed unit disc and those roots which are in boundary should be
simple but here the roots of ρ(z) are 1,λ so if λ = −1 then we have
no problem in zero stability but if λ = 1 then 1 is a root of ρ(z)
with multiplicity 2 which is on boundary but not simple so we can’t
consider λ = 1 that’s why the interval where λ belongs to is
[−1, 1).]
 ρ(z)
Consider a Taylor series expansion of the function log (z) about the
point z = 1.

ρ(z) (z − 1)(1 − λ + (z − 1))

= 2 3 4
log (z) (z − 1) − (z−1)
2 + (z−1)
3 − (z−1)
4 + ...
z − 1 (z − 1)2 (z − 1)3

= (1 − λ + (z − 1)) {1 − + −
2 3 4
+ O((z − 1)4 )}−1
z − 1 (z − 1)2 (z − 1)3

= (1 − λ + (z − 1)) 1 − { + −
2 3 4
−1
+ O((z − 1)4 )}
 z − 1 (z − 1)2 (z − 1)3
 
4
= (1 − λ + (z − 1)) 1 + − + + 0((z − 1) )
2 3 4
3−λ 5+λ 1+λ
=1−λ+ (z − 1) + (z − 1)2 − (z − 1)3
2 12 24
+ O((z − 1)4 )
A two-step method of maximum order is obtained by selecting
 
3−λ 5+λ ρ(z)
σ(z) = 1 − λ + (z − 1) + (z − 1)2 ∵ φ(z) = − σ(z)
2 12 logz
1 + 5λ 2 − 2λ 5+λ 2
=− + z+ z
12 3 12
If λ 6= −1, the resulting method is of the third order with error
constant
1+λ
C4 = −
24
whereas if λ = −1, the method is of fourth order. In the former
case, the method is
 
5+λ 2 − 2λ 1 + 5λ
yn+2 −(1+λ)yn+1 +λyn = h fn+2 + fn+1 − fn
12 3 12

with λ a parameter contained in the interval(−1, 1).

Calculation: From (4) we have the linear k− step method is
k
X k
X
αj yn+j = h βj f (xn+j , yn+j )
j=0 j=0

here in our case k = 2,

2
X 2
X
αj yn+j = h βj f (xn+j , yn+j )
j=0 j=0

⇒ α0 yn + α1 yn+1 + α2 yn+2 = h [β0 f (xn , yn ) + β1 f (xn+1 , yn+1 )

+ β2 f (xn+2 , yn+2 )
(9)

now if i able to determined the unknown co-efficient i.e.

α0 , α1 , α2 , β0 , β1 , β2 and put in (9) then i will the get the required
method.
so
What are α0 , α1 , α2 , β0 , β1 , β2 ?
αi ’s are obtained by comparing co-efficients of the like powers of z
the actual definition of ρ(z) and the given ρ(z).
2
X
ρ(z) = αj zj = α0 + α1 z + α2 z 2
j=0

and
ρ(z) = (z − 1)(z − λ)
= z 2 − (1 + λ)z + λ

so comparing both we have

α0 = λ, α1 = −(1 + λ), α2 = 1
Similarly βi ’s are obtained from the following two equations:
2
X
σ(z) = βj z j = β0 + β1 z + β2 z 2
j=0

and
1 + 5λ 2 − 2λ 5+λ 2
σ(z) = − + z+ z
12 3 12
so by comparing the co-efficients of the like powers of z we have
1 + 5λ 2 − 2λ 5+λ
β0 = − , β1 = , β2 =
12 3 12
By substituting all αi ’s and βi ’s value (i = 1, 2) in (9) we get the
required method.
Similarly in the later case, the method has the form
h
yn+2 − yn = (fn+2 + 4fn+1 + fn ).
3
Now by inspection,the linear k-step method (4) has 2k + 2
co-efficients
αj , βj , j = 0, 1, ..., k of which αk is taken to be 1 by normalisation.
This leaves us with 2k + 1 free parameters if the method is implicit
and 2k free parameters if the method is explicit(given that in the
later case βk is fixed to have value 0 because presence of βk
implies that the method is implicit).
According to (7), if the method is required to have order p, p + 1
linear relationships C0 = 0, ..., Cp = 0 including αj , βj , j = 0, ..., k,
must be satisfied. Thus in the case of the implicit method, we can
impose p + 1 = 2k + 1 linear constants C0 = 0, ..., C2k+1 = 0 to
determine the unknown constants, yielding a method of order
p = 2k.
Similarly, in the case of an explicit method, the highest order we can
expect is p = 2k − 1.
Unfortunately there is no guarantee that such method will be zero
stable. Indeed, in a paper published in 1956 Dahlquist proved
that there is no consistent zero-stable k-step method which
is of order > k + 2. Therefore, the maximum order 2k, 2k − 1
cannot be attained without violating the condition of zero-stability.
We formalise these facts in the next theorem.
3.Theorem 12 Theorem There is no zero-stable k-step method
whose order exceeds k + 1 if k is odd or k + 2 if k is even.
proof: Consider a linear k-step method (4) with associated first and
second stability polynomials ρ and σ. Further, consider the
transformation

ζ −1
ζ ∈ C 7→ ∈C
ζ +1

which maps :

I The open unit disc |ζ| < 1 of the ζ-plane onto the open
half-plane Rz < 0 of the z−plane.

I The circle |ζ| = 1 is mapped onto the imaginary axis Rz = 0.

I The point ζ = 1 onto z = 0 and

I the point ζ = −1 onto ζ = ∞.

It is clear that the functions r and s defined by
 k  
1−z 1+z
r (z) = ρ ,
2 1−z

 k  
1−z 1+z
s(z) = σ
2 1−z

are in fact polynomials, deg (r ) ≤ k and deg (s) ≤ k.

I If ρ(ζ)) has a root of multiplicity p, 0 ≤ p ≤ k; at
ζ0 −1
ζ = ζ0 6= −1,then r (z) has a root of multiplicity p at z = ζ0 +1 .
Calculation:
[
since ζ0 is a root of ρ(ζ) with multiplicity p. so we can write

ρ(ζ0 ) = 0

now r (z) = 0 if
 
1+z
ζ0 =
1−z
ζ0 − 1
=⇒ z =
ζ0 + 1
]

I If ρ(ζ) has a root of multiplicity p ≥ 1,0 ≤ p ≤ k, at ζ = −1,

then r (z) is of order degree k − p.
Calculation:
[
Since ζ = −1 is a root of ρ(ζ) with multiplicity p. so (ζ + 1)p is a
one factor of ρ(ζ). this means since ρ(ζ) is of degree ≤ k so ρ(ζ) is
of the form:
ρ(ζ) = (ζ + 1)p × [deg (polynomial) ≤ (k − p)]
   p
1+ζ 1+ζ
=⇒ ρ = + 1 × [deg (polynomial) ≤ (k − p)]
1−ζ 1−ζ
 p
2
= × [deg (polynomial) ≤ (k − p)]
1−ζ

now it is clear that the degree of r (z) is k − p when ζ = −1 is a

root of multiplicity p.
]
Since, by assumption, the method is zero-stable,
I ζ = 1 is a simple root of ρ(ζ).
consequently,
I z = 0 is a simple root of r (z) = 0.
Thus

r (z) = a1 z + a2 z 2 + · · · + ak z k , aa 6= 0, aj ∈ R.
It can be assumed, without loss of generality, that a1 > 0. since by
zero-stability all roots of ρ(ζ) are contained in the closed unit disc,
we deduce that all roots of r (z) have real parts ≤ 0. Therefore, all
coefficients aj , j = 1, . . . , k of r (z) are nonnegative.
Now let us consider the function
 k  
1−z 1+z 1
q(z) = φ = r (z) − s(z).
2 1−z log 1+z
1−z
Calculation:
[
we have

ρ(z)
φ(z) = − σ(z)
log (z)
 
1+z
1+z ρ 1−z

1+z

=⇒ φ( )=   −σ
1−z log 1+z 1−z
1−z
   
1+z

1−z
k
1+z

1−z
k ρ 1−z
 
1+z 
=⇒ × φ( )= ×   −σ
2 1−z 2 log 1+z 1−z
1−z
1
=⇒ q(z) =   × r (z) − s(z)
1+z
log 1−z

]
The function q(z) has a zero of multiplicity p at z = 0 if and only if
φ(ζ) defined by (8) has a zero of multiplicity p at ζ = 1;
Calculation:(proof )
[
z = 0 is a root of q(z) with multiplicity p.so

q(0) = 0
 k  
1−0 1+0
⇐⇒ ×φ =0
2 1−0
⇐⇒ φ(1) = 0

iff 1 is a root of φ(ζ) with multiplicity p.

hence proved.
]
according to Lemma this is equivalent to the linear k− step method
associated with ρ(ζ) and σ(z) has order p then

s(z) = b0 + b1 z + b2 z 2 + · · · + bp−1 z p−1 .

where

z r (z)
1+z z
= b0 + b1 z + b2 z 2 + . . .
log 1−z
As the degrees of s(z) is ≤ k, the existence of a consistent
zero-stable k− step linear multi-step method of order p > k + 1(or
p > k + 2) now looking for the possibility that

bk+1 = · · · = bp−1 = 0. or (bk+2 = · · · = bp−1 = 0)

Let us consider whether this is possible.
We denote by c0 ,c1 ,c2 ,. . . the coefficients in the Taylor series
expansion of the function
z
1+z
log 1−z

namely,
z
= c0 + c2 z 2 + c4 z 4 + . . .
log 1+z
1−z

Then, adopting the notational convention that av = 0 for v > k, we

have that
 z
  = (c0 + c1 z 2 + c4 z 4 + . . . )
1+z
log 1−z
r (z) z r (z)
×  = × (c0 + c1 z 2 + c4 z 4 + . . . )
z 1+z
log 1−z z

(a1 z + a2 z 2 + · · · + ak z k )
b0 + b1 z + b2 z 2 + . . . = × (c0 + c2 z 2 + c4 z 4 +
z
so by comparing co-efficients of like powers of z from both sides we
have
 b0 = c0 a1
b1 = c0 a2 .
etc.
b2v = c0 a2v +1 + c2 a2v −1 + · · · + c2v a1 , v = 1, 2, . . .

now it is clear that c2v < 0, v = 1, 2, . . .

(1) If k is an odd number, then since av = 0 for v > k, we have that

bk+1 = c2 ak + c4 ak−2 + · · · + ck+1 a1 .

since a1 > 0 and no av is negetive, it follows that bk+1 < 0.
(2) If k is an even number, then

bk+1 = c2 ak + c4 ak−2 + · · · + ck a2 .
since c2v < 0, v = 1, 2, . . . and aµ ≥ 0,µ = 2, 3, . . . , k, we deduce
that bk+1 = 0 if and only if
 a2 = a4 = · · · = ak = 0,
i.e., when r (z) is an odd function of z. This together with the fact
that all roots of r (z) have real parts ≤ 0, implies that all roots of
r (z) must have real part equal to zero. Consequently, all roots of
ρ(ζ) lie on |ζ| = 1. Since ak = 0,the degree of r (z) is k − 1 and
therefore −1 is a (simple) root of ρ(ζ).
As c2v < 0,aµ≥0 and a1 > 0, it follows that

bk+2 = c4 ak+1 + c6 ak−3 + · · · + ck+2 a1 < 0.

showing that bk+2 6=0.

Thus
I if a k− step method is zero-stable and k is odd then bk+1 6= 0,
whereas
I if k is even then bk+2 6= 0.
This proves that there is no zero-stable k− step method whose
order exceeds k + 1 if k is odd or k + 2 is k is even.
Hence proved.
4.Reference:
1. Numerical Solution of Ordinary Differential Equations : Endre
Siili; Mathematical Institute, University of Oxford
Thank You