Newmarks method and discrete

energy
applied to resistive MHD equation
CHIBA, Fumihiro

and KAKO, Takashi

The University of Electro-Communications
Chofu, Tokyo, Japan
Abstract
For the second order time evolution equation with a general dissi-
pation term, we introduce a recurrence relation of Newmarks method.
Deriving an energy inequality from this relation, we obtain the stabil-
ity and the convergence theories of Newmarks method. Next we take
up resistive MHD equation as an application of Newmarks method.
We introduce a discrete energy of the solution derived from the above
energy inequality. We investigate this quantity using numerical ex-
periments.
1 Introduction of Newmarks method
We recall the basic ideas of Newmarks method[7],[10] for the second order
time evolution equation in the d-dimensional Euclidean space R
d
. Let M, C
and K be dd real symmetric matrices on R
d
which are constant in time, and
let f(t) be a given R
d
-valued function on [0, ). We use the usual Euclidean
scalar product (, ) and the corresponding norm | | in R
d
. Throughout this
paper, we assume that M is positive denite:
(Mx, x) > 0, x ,= 0, (1)

chibaf@mac.com, http://math.digi2.jp/
1
which is equivalent to the condition:
(Mx, x) m(x, x), x, (2)
with a positive constant m. The identity matrix on R
d
is denoted by I. We
consider the approximation method for the following initial value problem of
the second order time evolution equation for an R
d
-valued function u(t):
M
d
2
u
dt
2
(t) +C
du
dt
(t) + Ku(t) = f(t), u(0) = u
0
,
du
dt
(0) = v
0
. (3)
Let a
n
, v
n
and u
n
be approximations of (d
2
/dt
2
)u(t), (d/dt)u(t) and u(t)
respectively at t = n with a positive time step and an integer n, and let
f
n
= f(n). Then Newmarks method for (3) is dened through the following
relations:

Ma
n
+ Cv
n
+Ku
n
= f
n
,
u
n+1
= u
n
+ v
n
+
1
2

2
a
n
+
2
(a
n+1
a
n
),
v
n+1
= v
n
+a
n
+(a
n+1
a
n
).
(4)
The rst relation corresponds to the equation (3), and the second and the
third relations correspond to the Taylor expansions of u(t + ) and v(t + )
at t = n respectively. Here, and are positive tuning parameters of the
method.
An interpretation of the parameter related to the acceleration (d
2
/dt
2
)u(t)
can be seen in [7]: The case = 1/6 corresponds to the approximation where
the acceleration is a linear function of t in each time interval; the case = 1/4
corresponds to the approximation of the acceleration to be a constant func-
tion during the time interval which is equal to the mean value of the initial
and nal values of acceleration; the case = 1/8 corresponds to the case
where the acceleration is a step function with the initial value for the rst
half of each time interval and the nal value for the second half of the interval.
It is often the case that is equal to 1/2.
2 Iteration scheme of Newmarks method
Based on the formulas in (4), Newmarks method generates the approxima-
tion sequence u
n
, n = 0, 1, 2, . . . , N by the following iteration scheme:
Step 1. For n = 0, compute a
0
from the initial data u
0
and v
0
:
a
0
= M
1
(f
0
Cv
0
Ku
0
).
2
Step 2. Compute a
n+1
from f
n+1
, u
n
, v
n
and a
n
solving a linear equation:
a
n+1
= (M +C +
2
K)
1
[Ku
n
(C +K)v
n
+( 1)C + (
1
2
)
2
Ka
n
+f
n+1
].
Step 3. Compute u
n+1
from u
n
, v
n
, a
n
and a
n+1
:
u
n+1
= u
n
+v
n
+ (
1
2
)
2
a
n
+
2
a
n+1
.
Step 4. Compute v
n+1
from v
n
, a
n
and a
n+1
:
v
n+1
= v
n
+ (1 )a
n
+a
n+1
.
Step 5. Replace n by n + 1, and return to Step 2.
Here, Step 1 is nothing but the rst relation of (4) with n = 0. The expression
of a
n+1
in Step 2 is obtained by eliminating u
n+1
and v
n+1
from the second
and the third equalities in (4) together with the rst equality in (4) with n
replaced by n + 1:
Ma
n+1
+Cv
n+1
+ Ku
n+1
= f
n+1
.
Step 3 and Step 4 are from the second and the third relations of (4).
3 Recurrence relation of Newmarks method
Newmarks method (4) for the second order equation (3) is reformulated as
follows in the recurrence relation([1] and [11]). Eliminating v
n
and a
n
from
(4) by a series of lengthy computations, we have
(M +
2
K)D

u
n
+CD

u
n
+(1 )C +(
1
2
)KD

u
n
+Ku
n
= I +(
1
2
)D

+
2
D

f
n
,
(5)
where

u
n
= (u
n+1
u
n
)/,
D

u
n
= (u
n
u
n1
)/,
D

u
n
= (D

u
n
D

u
n
)/.
We conrmed this result by a formula manipulation software NCAlgebra [4]
on Mathematica[12]. Especially, in the case = 1/2, we have:
(M +
2
K)D

u
n
+
1
2
C(D

+ D

)u
n
+Ku
n
= (I +
2
D

)f
n
. (6)
These recurrence relations are useful for the stability and error analyses of
Newmarks method. See [6] and [9] for the case with C 0.
3
4 Derivation of an energy inequality
Taking a scalar product of (5) and (D

+D

)u
n
, we can derive an energy in-
equality for Newmarks method. From this inequality, we obtain the stability
conditions for Newmarks method.
From now on, let T be a xed positive number and N be a positive
integer, and dene := T/N and n = 0, 1, 2, . . . , N. Let C be nonnegative:
(Cx, x) 0, K be positive denite: (Kx, x) k(x, x) with k > 0, and m > 0
be the smallest eigenvalue of M. We also assume that
1
2
. We have the
following theorem for the energy inequality.
Theorem 1 Let u
n

N
n=0
be a sequence generated by the scheme in Section
2. Then for a positive constant
0
determined as below, there exists a positive
constant C
0
such that the inequality:
|M
1/2
D

u
n
|
2
+
2

1
2
(
1
2
)
1
2
|K
1/2
D

u
n
|
2
+ (1
1
2
)|K
1/2
u
n
|
2
C
0
, (7)
holds for all
0
and an arbitrary positive constant , where
0
is deter-
mined either as

0
> 0 when

2
, (8)
or as
0 <
0
<

m
(
1
2
)|K|
when 0 <

2
. (9)
Remark 1 The constant C
0
is dened as follows:
C
0
:=
(
1
1
2

0
)
1
[
w
0
+
T

2
{
(4 + 2) sup
0tT
|f(t)|
}
2
]
exp
(

(
1
1
2

0
)
1
T
)
, (10)
where
w
0
= |M
1/2
D

u
0
|
2
+
2
{

1
2
(

1
2
)}
|K
1/2
D

u
0
|
2
+(KD

u
0
, u
0
) +|K
1/2
u
0
|
2
+(
1
2
)|C
1/2
D

u
0
|
2
,
(11)
and in (11) is dened as:
0 < m, in the case (8), (12)
or
0 < m
2
0
(
1
2

)
|K|, in the case (9). (13)
4
Proof of Theorem 1. Using (5), we derive an energy inequality as
follows (see [1] for the case = 1/2). Rearranging (5), we have
(M +
2
K)D

u
n
+
1
2
C(D

+D

)u
n
+(
1
2
)C(D

D

)u
n
+(
1
2
)KD

u
n
+Ku
n
= g
n
,
(14)
where
g
n
:= I +(
1
2
)D

+
2
D

f
n
. (15)
We take a scalar product of (14) and (D

+ D

)u
n
. Since C is nonnegative,
we have
((M +
2
K)D

u
n
, (D

+D

)u
n
) + (
1
2
)(C(D

D

)u
n
, (D

+D

)u
n
)
+(
1
2
)(KD

u
n
, (D

+D

)u
n
) + (Ku
n
, (D

+D

)u
n
) (g
n
, (D

+D

)u
n
)
= (
1
2
C(D

+D

)u
n
, (D

+D

)u
n
) 0.
(16)
Using the assumption for M, C and K, we obtain the following lemma:
Lemma 1 When we put
w
n
:= ((M +
2
K)D

u
n
, D

u
n
) + (Ku
n+1
, u
n
)
+(
1
2
)(CD

u
n
, D

u
n
)
1
2

2
(
1
2
)(KD

u
n
, D

u
n
),
we have
w
n
w
n1
+(g
n
, D

u
n
+D

u
n1
). (17)
Proof. Multiplying to both side of (16) and using the relation:
D

= D

D

,
we have
((M +
2
K)(D

D

)u
n
, (D

+D

)u
n
) +(
1
2
)(C(D

D

)u
n
, (D

+D

)u
n
)
+
1
2

2
(
1
2
)(K(D

+D

)u
n
, (D

+D

)u
n
)
1
2

2
(
1
2
)(K(D

D

)u
n
, (D

+D

)u
n
)
+(Ku
n
, (D

+D

)u
n
) (g
n
, (D

+D

)u
n
)
0.
(18)
Using the identities:
((D

D

)u
n
, (D

+D

)u
n
) = (D

u
n
, D

u
n
) (D

u
n1
, D

u
n1
),
and
D

u
n
=
1
2
(D

+D

)u
n

1
2
(D

D

)u
n
,
5
we have
((M +
2
K)D

u
n
, D

u
n
) ((M +
2
K)D

u
n1
, D

u
n1
) + (
1
2
)(CD

u
n
, u
n
)
(
1
2
)(CD

u
n1
, u
n1
) +
1
2

2
(
1
2
)(K(D

+D

)u
n
, (D

+ D

)u
n
)
+
1
2

2
(
1
2
)(KD

u
n
, D

u
n
)
1
2

2
(
1
2
)(KD
tau
u
n1
, D

u
n1
) +(Ku
n
, (D

+D

)u
n
)
(g
n
, (D

+D

)u
n
) 0.
(19)
Furthermore, using the relation:
(Ku
n
, (D

+D

)u
n
) = (Ku
n
, u
n+1
u
n1
) = (Ku
n+1
, u
n
) (Ku
n
, u
n1
),
we have
((M +
2
K)D

u
n
, D

u
n
) ((M +
2
K)D

u
n1
, D

u
n1
)
+(
1
2
)(CD

u
n
, u
n
) (
1
2
)(CD

u
n1
, u
n1
)
+
1
2

2
(
1
2
)(KD

u
n
, D

u
n
)
1
2

2
(
1
2
)(KD

u
n1
, D

u
n1
)
+(Ku
n+1
, u
n
) (Ku
n
, u
n1
) +
1
2

2
(
1
2
)(K(D

+D

)u
n
, (D

+D

)u
n
)
(g
n
, (D

+D

)u
n
).
(20)
We neglect the nonnegative term
1
2

2
(
1
2
)(K(D

+ D

)u
n
, (D

+ D

)u
n
)
in the left-hand side and obtain (17).
Proof of Theorem 1 (continued). Using the equality:
(Ku
n+1
, u
n
) = (KD

u
n
, u
n
) + (Ku
n
, u
n
),
we can modify the expression of w
n
as
w
n
= |M
1/2
D

u
n
|
2
+
2

1
2
(
1
2
)|K
1/2
D

u
n
|
2
+(KD

u
n
, u
n
) +|K
1/2
u
n
|
2
+(
1
2
)|C
1/2
D

u
n
|
2
,
(21)
Summing up (17) from n = 1 to n, we obtain
w
n
w
0
+
n

i=1
(g
i
, D

u
i
+ D

u
i1
). (22)
To modify (22), we show the next lemma.
Lemma 2 Under either the condition
(8) and 0 < m, (23)
or
(9) and
0
and 0 < m
2
0
(
1
2
)|K
1/2
|
2
, (24)
we have
|D

u
i
|
2
w
i
. (25)
6
Proof. Let be a positive number. Using the assumption that M and
K are positive denite, C is nonnegative and 1/2, we have
w
i
|D

u
i
|
2
|M
1/2
D

u
i
|
2
+
2
(
1
2
)|K
1/2
D

u
i
|
2
+
1
4
|K
1/2
(D

u
i
+ 2u
i
)|
2
|D

u
i
|
2
(m)|D

u
i
|
2
+
2
(
1
2
)|K
1/2
D

u
i
|
2
+
1
4
|K
1/2
(D

u
i
+ 2u
i
)|
2
.
If and satises the condition (8) and m, then w
i
|D

u
i
|
2
becomes
nonnegative for any i.
On the other hand, if /2 < 0, then we have
w
i
|D

u
i
|
2
m|D

u
i
|
2

2
(
1
2
)|K
1/2
|
2
|D

u
i
|
2
|D

u
i
|
2
= m
2
(
1
2
)|K
1/2
|
2
|D

u
i
|
2
.
If
0
satises the condition (9) and
0
then
m
2
(
1
2
)|K
1/2
|
2
> m
2
0
(
1
2
)|K
1/2
|
2
.
Hence, if m
2
0
(
1
2
)|K
1/2
|
2
, then w
i
|D

u
i
|
2
0 for any i.
Thus we obtain (25).
Using Lemma 2, we have the next lemma.
Lemma 3 Under either the condition
(8) and 0 < minm, 2/
0
, (26)
or
(9) and
0
and 0 < minm
2
0
(
1
2
)|K
1/2
|
2
, 2/
0
, (27)
we have, for
0
,
w
n
C
0
, n = 0, 1, 2, . . . , N 1. (28)
7
Proof. Using (25), we modify (22) as follows:
w
n
w
0
+
n

i=1
(g
i
, D

u
i
+D

u
i1
)
w
0
+
n

i=1
|g
i
|(|D

u
i
| +|D

u
i1
|)
w
0
+
n

i=1

2
|g
i
|
2
+ (

2
2
|D

u
i
|
2
+

2
2
|D

u
i1
|
2
)
w
0
+

2
n

i=1
(w
i
+w
i1
) +

2
n

i=1
|g
i
|
2
w
0
+

2
w
n
+
n1

i=0
w
i
+

2
N1

i=1
|g
i
|
2
.
If f(t) is bounded on [0, T], then, for i 1, we have
|g
i
| = |I +(
1
2
)D

+
2
D

f
i
|
= |f
i
+ (
1
2
)(f
i
f
i1
) + (f
i+1
2f
i
+ f
i1
)| (4 + 2) sup
0tT
|f(t)|.
Hence we have
w
n
w
0
+

2
w
n
+
n1

i=0
w
i
+

2
N1

i=1
(4 + 2) sup
0tT
|f(t)|
2
w
0
+

0
2
w
n
+
n1

i=0
w
i
+
T

2
(4 + 2) sup
0tT
|f(t)|
2
.
Since 0 < 1
0
/2 from (26) or (27), we have
w
n
(1
1
2

0
)
1
[w
0
+
n1

i=0
w
i
+
T

2
(4 + 2) sup
0tT
|f(t)|
2
].
Using the Gronwall inequality, we then obtain
w
n
(1
1
2

0
)
1
[w
0
+
T

2
(4 + 2) sup
0tT
|f(t)|
2
] exp((1
1
2

0
)
1
T),
where we use N = T. Thus, we can dene C
0
as follows
C
0
:= (1
1
2

0
)
1
[w
0
+
T

2
(4 + 2) sup
0tT
|f(t)|
2
] exp((1
1
2

0
)
1
T), (29)
8
where
w
0
= |M
1/2
D

u
0
|
2
+
2

1
2
(
1
2
)|K
1/2
D

u
0
|
2
+(KD

u
0
, u
0
) +|K
1/2
u
0
|
2
+(
1
2
)|C
1/2
D

u
0
|
2
.
(30)
Proof of Theorem 1 (continued). Lastly, under either the condition
(8) or (9), using the following inequality with an arbitrary :
(K
1/2
D

u
n
, K
1/2
u
n
) |K
1/2
D

u
n
|

|K
1/2
u
n
|

1
2

2
|K
1/2
D

u
n
|
2
+
1

|K
1/2
u
n
|
2
,
we obtain the energy inequality (7) from (28) in Lemma 3. From (22) we
have
If f
n
= g
n
= 0, then C
0
= w
0
.
Remark 2 When C = 0, = 1/2 and f(t) 0, we obtain
((M+
2
K)D

u
n
, D

u
n
)+(Ku
n+1
, u
n
) = ((M+
2
K)D

u
0
, D

u
0
)+(Ku
1
, u
0
).
Namely, in this case, Newmarks method conserves energy.
Remark 3 The constant C
0
in the theorem depends primarily on the coe-
cient matrices M, C and K, and secondarily on the tuning parameters and
and it also depends on the initial values u
0
and v
0
and the inhomogeneous
term f, and nally on
0
in the way described above.
Remark 4 In the case C is negative or K is not positive, we consider v(t) :=
e
t
u(t) instead of u(t). By choice of appropriate > 0, we may obtain new
matrices satisfying the condition in Theorem 1 as coecients of equation of
v(t).
5 Stability and convergence conditions for New-
marks method
In this section, using the energy inequality (7) we derive stability conditions
for Newmarks method. With respect to a parameter , we divide the sta-
bility condition into two cases and lead to the next theorem.
9
5.1 Stability theorem
Theorem 2 Let M and K be positive denite and C be nonnegative. Let
m and k be the smallest eigenvalues of M and K. Assume 1/2. Then,
Newmarks method for (3) in a time interval [0, T] is stable in the following
two cases with respect to :
Case 1: If
1
2
< , then with
0
> 0 and given in (12) and (13) we have,
for <
0
and N = T/,
|u
n
|

(42+1)
(42)k
C
0
, n = 0, 1, 2, . . . , N. (31)
Case 2: If 0
1
2
, then for
0
satisfying
0 <
0
<

m
(
1
2
)K
, (32)
we have for
0
and N = T/
|u
n
| |u
0
| +

C
0
m
2
0
(
1
2
)K
n, n = 0, 1, 2, . . . , N.
(33)
Here, in both cases, C
0
is given by (10) together with (11).
Proof. First, we treat Case 1. We consider the following condition for
the coecients in (7):

1
2
(
1
2
)
1
2
= 0 and 1
1
2
> 0.
This implies that

1
2
(
1
2
) =
1
2
>
1
4
,
and hence we have
>
1
2

1
4
.
Conversely, if we assume that > /2 1/4 and put = 2 ( 1/2),
then we have 1 1/2 > 0. Using (7), we have
k|u
n
|
2
|K
1/2
u
n
|
2
(1
1
2
)
1
C
0
.
Hence we have the stability estimate (31).
10
Next, we consider Case 2. Putting = 1/2 in (7), we have
C
0
|M
1/2
D

u
n
|
2
+
2
(
1
2
)|K
1/2
D

u
n
|
2
m|D

u
n
|
2
+
2
(
1
2
)|K
1/2
D

u
n
|
2
.
Noticing the condition /2 > 0, we have
m
2
(
1
2
)|K
1/2
|
2
|D

u
n
|
2
C
0
.
Under the condition:

0
<

m
(
1
2
)K
1/2

2
,
the coecient in the left-hand side becomes positive and we have
|D

u
n
|
2

C
0
m
2
(
1
2
)K
1/2

2

C
0
m
2
0
(
1
2
)K
1/2

2
.
Thus we have (33) in the same way as in Case 1.
Remark 5 H. Fujii investigated in [2] and [3] the stability condition for the
Rayleigh damping case with C = aK + bM, a, b R.
5.2 Convergence theorem
Using the recurrence relation (5) and the stability theorem, we can show in
this section the convergence of Newmarks method together with its conver-
gence order.
Let u(t) be the solution of (3) and u
n
(n = 0, 1, 2, . . . , N) be the solution
of Newmarks method for (3). We assume that M and K are positive denite
and C is nonnegative.
The discretization error e
n
is dened as e
n
:= u(n) u
n
. Then we have
the following theorem.
Theorem 3 We assume that f C
2
([0, T]), then we have the estimate
|e
n
| = O(
l
), where

l = 2 for =
1
2
,
l = 1 for >
1
2
.
Proof. To prove this theorem, we rst show the following two lemmas.
11
Lemma 4 At the mesh points t = n with n = 0, 1, we have
e
0
= 0, e
1
= O(
3
).
Proof. Since u
0
= u(0), we have e
0
= u(0) u
0
= 0. Next we estimate
e
1
. From Step 3 in the iteration scheme in Section 2, we have
u
1
= u
0
+v
0
+ (
1
2
)
2
a
0
+
2
a
1
,
where
v
0
=
du
dt
(0), a
0
=
d
2
u
dt
2
(0)
and
a
1
= (M +C +
2
K)
1
[f
1
(C +K)v
0
Ku
0
+( 1)C + (
1
2
)
2
Ka
0
].
On the other hand, by Taylors theorem, we have
u() = u
0
+v
0
+
1
2

2
a
0
+O(
3
) = u
0
+v
0
+ (
1
2
)
2
a
0
+
2
a
0
+O(
3
).
So, we obtain
e
1
= u() u
1
=
2
a
0
+ O(
3
)
2
a
1
=
2
(a
0
a
1
) +O(
3
),
and from the relation
a
0
= M
1
(f
0
Cv
0
Ku
0
),
and the above expression of a
1
, we have
a
1
= (M
1
+O()) [(f
0
+O()) (Cv
0
+O()) Ku
0
+O()]
= M
1
(f
0
Cv
0
Ku
0
) + O().
Thus we obtain e
1
= O(
3
).
At the mesh points t = n, n = 0, 1, 2, . . . , N, we have u
n
= u(n) e
n
and f
n
= f(n). Using the recurrence relation (5), we can prove the next
technical lemma.
Lemma 5 Dene p
n
as
p
n
:= (M +
2
K)D

e
n
+ [CD

+ (1 )CD

+(
1
2
)KD

]e
n
+Ke
n
. (34)
Then it is expressed as follows:
p
n
= (
1
2
)M
d
3
u
dt
3
(n) +O(
2
). (35)
12
Proof. Substituting u(n) e
n
for u
n
in the recurrence relation (5), we
have
p
n
= (M +
2
K)D

e
n
+ [CD

+ (1 )CD

+(
1
2
)KD

]e
n
+Ke
n
= (M +
2
K)D

u(n) + [CD

+ (1 )CD

+(
1
2
)KD

]u(n) +Ku(n)
I +(
1
2
)D

+
2
D

f(n).
Using the expressions
D

u(n) = u((n + 1)) u(n)/, D

f(n) = f(n) f((n 1))/,
D

u(n) = u((n + 1)) 2u(n) +u((n 1))/
2
, etc.,
we rewrite the right hand side of the above formula. Applying Taylors
theorem to u((n + 1)), f((n 1)), etc. at t = n, we have
p
n
= M
d
2
u
dt
2
(n) +C
du
dt
(n) +Ku(n) f(n)
+(
1
2
)(C
d
2
u
dt
2
(n) +K
du
dt
(n)
df
dt
(n)) +O(
2
).
Using the equalities:
M
d
2
u
d
2
t
(n) + C
du
dt
(n) + Ku(n) f(n) = 0,
and
C
d
2
u
dt
2
(n) +K
du
dt
(n)
df
dt
(n) = M
d
3
u
dt
3
(n),
we obtain (35).
Proof of Theorem 3 (continued). Using Lemma 4, 5 and the stability theorem
we can obtain the estimate of |e
n
|. To apply Theorem 2 to (34) we consider
a modication of (10). If we look the proof of Theorem 1 again, we can
replace (4 + 2) sup
0tT
|f(t)|
2
with sup
1nN1
|p
n
|
2
. Then we have in
this case
C
0
= (1
1
2

0
)
1
w
0
+
T

2
sup
1nN1
|p
n
|
2
exp((1
1
2

0
)
1
T).
Hence we have from (31) or (33)
|e
n
| C
1

C
0
+|e
0
| = C
1

C
0
,
where C
1
is independent of u
0
, v
0
and p
n
. By Lemma 4 and (11), where u
0
is replaced by e
0
, we have
w
0
= |M
1/2
D

e
0
|
2
+
2

1
2
(
1
2
)|K
1/2
D

e
0
|
2
+(KD

e
0
, e
0
) +|K
1/2
e
0
|
2
+(
1
2
)|C
1/2
D

e
0
|
2
= |M
1/2 1

e
1
|
2
+
2

1
2
(
1
2
)|K
1/2 1

e
1
|
2
+(
1
2
)|C
1/2 1

e
1
|
2
= O(
4
).
13
From this we obtain with another constant C
2
C
0
C
2
O(
4
) + sup
1nN1
|p
n
|
2
.
On the other hand, from Lemma 5, we have

sup
1nN1
|p
n
|
2
= O(
4
) for =
1
2
,
sup
1nN1
|p
n
|
2
= O(
2
) for >
1
2
.
Thus we obtain the results.
5.3 Energy inequality and discrete energy
Related to an energy inequality, we dene the following quantity.:
w
n
:= (MD

u
n
, D

u
n
) +
(

1
2
)
(CD

u
n
, D

u
n
)
+
2
{

1
2
(

1
2
)}
(KD

u
n
, D

u
n
) + (Ku
n+1
, u
n
)
= ((M +
(

1
2
)
C +
2
(

1
2
(

1
2
))
K)D

u
n
, D

u
n
) + (Ku
n+1
, u
n
)
= (/D

u
n
, D

u
n
) + (Ku
n+1
, u
n
),
(36)
where
/ := M +
(

1
2
)
C +
2
(

1
2
(

1
2
))
K M.
We call w
n
a discrete energy. We also dene other quantities w

n
and w

n
:
w

n
:= (MD

u
n
, D

u
n
) + (Ku
n
, u
n
), (37)
and
w

n
:= (Mv
n
, v
n
) + (Ku
n
, u
n
), (38)
where v
n
is a velocity given in Newmarks method together with u
n
. These
are discrete analogues of the continuous energy:
c(t) :=
(
M
d
dt
u(t),
d
dt
u(t)
)
+ (Ku(t), u(t)). (39)
The dierence quotient D

u
n
approximates the derivative of u at t =
n + /2 with the second order in . So the average of (/D

u
n
, D

u
n
)
14
and (/D

u
n1
, D

u
n1
) is a second order approximation of (M
d
dt
u,
d
dt
u) at
t = n. The average of (Ku
n+1
, u
n
) and (Ku
n
, u
n1
) is represented as follows:
1
2
(Ku
n+1
, u
n
) + (Ku
n
, u
n1
) = (Ku
n
,
1
2
u
n+1
+u
n1
),
where we use the assumption that K is symmetric. Since
1
2
u
n+1
+ u
n1

is a second order approximation in of u

h
at t = n,
1
2
(Ku
n+1
, u
n
) +
(Ku
n
, u
n1
) is a second order approximation of (Ku, u) at t = n. Now
we dene the following quantity W
n
as a second order approximation in of
(39) at t = n:
W
n
:=
{
w
0
for n = 0,
1
2
(w
n
+ w
n1
) for n 1.
5.4 Convergence theorem of W
n
Theorem 4 For a xed T > 0, the convergence order of c(T) W
N
with
T = N is estimated as follows:
[c(T) W
N
[ =
{
O(
2
) for =
1
2
,
O() for >
1
2
.
Proof. Using e
n
= u(n)u
n
in Theorem 3, we replace u
n
= u(n)e
n
in c(n) W
n
and using the Taylor expansion of u(t) and the fact e
n
=
O(
l
), l = 1 or 2, we obtain
c(n) W
n
=
1
2
2
(M(e
n+1
e
n
), e
n+1
e
n
) + (M(e
n
e
n1
), e
n
e
n1
)
+
1

(
M(e
n+1
e
n1
),
du
dt
(n)
)
+O(
l
).
So, we consider the estimation of e
n+1
e
n
and e
n+1
e
n1
. We dene the
quantities
E
n
=
1

(e
n+1
e
n1
) for n 1,
and, using p
n
dened in Lemma 5, we put
P
n
= (M +
2
K)D

E
n
+ [CD

+ (1 )CD

+(
1
2
)KD

]E
n
+KE
n
=
1

(p
n+1
p
n1
) for n 1,
(40)
By the similar calculation in the proof of Theorem 3, we have
E
1
= O(
2
) and E
2
= O(
2
),
15
and also we get
P
n
= (2 1)
(
d
2
u
dt
2
(T) +C
d
3
u
dt
3
(T)
)
+O(
2
).
Hence we have P
n
= O(
3
) when = 1/2 and P
n
= O(
2
) when > 1/2. In
the same way as the proof of Theorem 3, we have the estimate of E
n
|E
n
|

O(
4
) + O
(
sup
0nN1
|P
n
|
2
)
.
In the same way as for E
n
, we have the estimate of e
n+1
e
n
|e
n+1
e
n
| =
{
O(
3
) for = 1/2,
O(
2
) for > 1/2.
Thus we obtain the following estimation:
[c(T) W
N
[ O(E
N
) =
{
O(
2
) for = 1/2,
O() for > 1/2.
6 Application to resistive MHD equation
6.1 Simplied resistive MHD equation
We investigate the behavior of solution to resistive MHD equation(see [8],[5]
for details). Let (t, x) := B
x
(t, x) be the x-component of magnetic eld at
time t which we assume depends only on xvariable in space direction and pe-
riodic with period one, and hence the function on RS, where S =R/N(the
unit circle). Then satises the following second order equation in t if we
neglect a weak interaction term along magnetic lines:
{

2
t
2


t
+H(x)
2
= 0,
=

2
x
2
,
(41)
where is a nonnegative constant called resistivity, and H(x) is a smooth
real periodic function which represents the Alfven frequency at x. We call
the problem (41) a model equation of magnetic line for the solution and the
16
resistive linearized MHD. Based on Newmarks scheme, we compute the time
evolution of the discrete approximations of energy dened in the formulas in
(36), (37) and (38).
For general discussion, we formulate the model problem to the abstract
equation for u(t):

2
t
2
u +L

t
u +Hu = 0, (42)
and we consider this equaiton in a Hilbert space H. We assume that L
is a nonnegative self-adjoint operator and the inverse operator (L + 1)
1
is
compact in H and H is a bounded operator. We refomulate the equation
(42) to the rst order system:
d
dt
(
u
u

)
= F

(
u
u

)
, F

:=
(
0 I
H L
)
. (43)
We dene the domain of F

as follows:
T(F

) := HT(L).
The operator F

is the generator of a C
0
-semigroup and its resolvent is given
as follows:
(F

)
1
=
(

1

+
1

D
1

H D
1

D
1

H D
1

)
, (44)
where D

()(= D

for short) is dened as follows:

D

() :=
2
+L +H, T(D

) = T(L). (45)
Since H is bounded and (L + 1)
1
is compact, D

has a compact inverse for

sucient large positive . Hence, (44) is well dened and has its essential
spectrum 0, .
We consider (43) on H H. In the weak form, (43) becomes to nd
v, w HH such that

t
(v, ) + (w, )(w, ) (Hv, ) (Lw, ), for all , H. (46)
We factorize L in the third term of the left-hand side and adopt the repre-
sentation:
(Lv, ) =
(
L
1/2
v, L
1/2

)
.
17
Then we can nally formulate the equation in 1 1 with 1 = T
(
L
1/2
)
to
nd v, w 1 1 such that

t
(v, ) + (w, ) = (w, ) (Hv, ) (L
1/2
w, L
1/2
), for all , 1.
(47)
6.2 Numerical method for the model problem
The weak formulation (47) introduces the following discrete approximation.
We takes as an approximate space 1
h
1
h
, where 1
h
is a nite dimensional
approximate subspace of 1. Then the nite element approximation problem
is to restrict the problem on 1
h
to nd v
h
, w
h
1
h
1
h
such that

t
(v
h
,
h
)+(w
h
,
h
) = (w
h
,
h
)(Hv
h
,
h
)(L
1/2
w
h
, L
1/2

h
), for all
h
,
h
1
h
.
(48)
Since L = with T(L) = H
2
(S), we have
T
(
L
1/2
)
= 1 = H
1
(S),
We dene a nite dimensional subspace 1
h
of 1 as a set of piecewise linear
continuous functions for the equipartition of S = [0, 1) with N mesh points.
Namely, we take h = 1/N and dene 1
h
as
1
h
:= u
h
: u
h
is periodic with period 1 and continuous,
and linear on [k h, (k + 1)h], k = 0, 1, . . . , N 1. (49)
6.3 Numerical tests of Newmarks method applied to
the model MHD equa-tion
We investigate numerically the convergence order of Newmarks method. We
calculate the numerical solution at t = T by Newmarks method with a time
step
p
= T/2
p
, p = 1, 2, 3, . . .. We put H(x) = 2 + sin 2x and h = 1/256.
Let us denote by u
n,
be the numerical solution with a time step at t = n
of (41), we assume the convergence order as follows:

u
2
16
,
16
u
2
p
,p

L
2
(0,1)
= C
1

l
p
, (50)
where we use u
2
16
,
16
as the ne enough approximation to the exact solution at
t = T(= 10). We compute numerically the convergence order l based on the
18
above assumption (50). In these calculations, we set = 0.001, 0.0001, 0, =
1/6, = 1/2, 0.51, 2/3, 3/4 and p = 6, 7, . . . , 16. We use the least square
method for estimating l. Table 1 shows the estimation of the convergence
order l for each . The results are consistent with the theoretical orders in
Theorem 3.
Table 1: Convergence order l for various
1/2 0.51 2/3 3/4
= 0.001 2.0371 1.2517 1.0149 0.9841
= 0.0001 2.0331 1.2019 1.0158 0.9873
= 0 2.0302 1.1995 1.0165 0.9884
6.4 Numerical tests of discrete energy
We estimate the convergence order of the quantities w
n
, w

n
, w

n
and W
n
. We
calculate these quantities at t = T with a time step
p
= T/2
p
, p = 1, 2, 3, . . ..
We put = 1/6, = 1/2 and H(x) = 2 + sin 2x. In the case of w
n
, let us
denote by w
n,
the numerical solution with a time step at t = n, and we
assume the relation of convergence order as follows:

w
2
16
,
16
w
2
p
,p

= C
2

l
p
, (51)
where we use w
2
16
,
16
on behalf of the exact solution at t = T(= 10) as its ne
enough approximation to the exact solution. We estimate the convergence
order l based on the above assumption (51). In these calculations, we put
= 0.001, 0.0001, 0, h = 1/256 and p = 6, 7, . . . , 16. We use the least square
method for estimating l. In other cases of w

n
, w

n
and W
n
, we estimate l
similarly. Table 2 shows the estimation of l in each case. It shows that W
n
and w
n
have very good performance.
Table 2: Estimates of l at h = 1/256
w
n
w

n
w

n
W
n
= 0.001 1.0774 1.1038 2.0311 2.1457
= 0.0001 1.3514 1.0940 2.0422 2.0265
= 0 2.0242 1.0480 2.0182 2.0242
Figure 1, 2, 3 and 4 show the time evolution behaviors of w
n
, w

n
, w

n
and
W
n
.
19
References
[1] Chiba, F. and Kako, T., On the stability of Newmarks method,
Research Institute of Mathematical Sciences, Kyoto University, RIMS
Koky uroku, 1040(1998), pp. 39-44.
[2] Fujii, H., Finite-Element Galerkin Method for Mixed Initial-Boundary
Value Problems in Elasticity Theory, (Center for Numerical Analysis,
The University of Texas at Austin, 1971).
[3] Fujii, H., A Note on Finite Element Approximation of Evolution Equa-
tions, Research Institute of Mathematical Sciences, Kyoto University,
RIMS Koky uroku, 202(1974), pp. 96117.
[4] Helton, J. W. and Miller, R. L., NCAlgebra, (Department
of Mathematics, University of California San Diego, 1994),
http://math.ucsd.edu/ncalg/.
[5] Kako, T., Spectral and numerical analysis of resistive linearized mag-
netohydrodynamic operators, In Series on Applied Mathematics Vol.5,
Proceedings of the First China-Japan Seminar on Numerical Mathemat-
ics, Shi Z-C. and T. Ushijima T.(eds), Kinokuniya,(1993), pp. 60-66.
[6] Matsuki, M. and Ushijima, T., Error estimation of Newmark method
for conservative second order linear evolution, Proc. Japan Acad Ser.
A.,69(1993),219-223.
[7] Newmark, N. M., A method of computation for structural dynamics,
Proceedings of the American Society of Civil Engineers, Journal of the
Engineering Mechanics Division. 85 EM 3(1959), pp. 67-94.
[8] Pao, Y. and Kerner, W., Analytic theory of stable resistive magneto-
hydrodynamic modes, Phys. Fluids.,28,1, pp. 287-293(1985).
[9] Raviart, P. A. and Thomas, J. M., Introduction `a lAnalyse Numerique
des Equations aux Derivees Partielles, Masson, Paris, 1983.
[10] Wood, W, L., Practical Time-Stepping Schemes, (Clarendon Press Ox-
ford, 1990).
20
[11] Wood, W. L., A further look at Newmark, Houbolt, etc., time-stepping
formulate, International Journal for Numerical Methods in Engineer-
ing, 20(1984), pp. 1009-1017.
[12] Wolfram, S., The Mathematica Book, Third Edition, (Addison-Wesley,
1998).
21
20 40 60 80 100
t
0.5
1
1.5
2
2.5
3
w
n
1
20 40 60 80 100
t
0.5
1
1.5
2
2.5
3
w
n

2
20 40 60 80 100
t
0.5
1
1.5
2
2.5
3
w
n

3
20 40 60 80 100
t
0.5
1
1.5
2
2.5
3
W
n
4
Figure 1: The case = 0.5: Frame 1=w
n
, Frame 2=w

n
, Frame 3 =w

n
and
Frame 4=W
n
20 40 60 80 100
t
0.5
1
1.5
2
2.5
3
Wn vs wn
Figure 2: The case = 0.5: the lower one is W
n
, and another one is w

n
20 40 60 80 100
t
0.5
1
1.5
2
2.5
3
w
n
1
20 40 60 80 100
t
0.5
1
1.5
2
2.5
3
w
n

2
20 40 60 80 100
t
0.5
1
1.5
2
2.5
3
w
n

3
20 40 60 80 100
t
0.5
1
1.5
2
2.5
3
W
n
4
Figure 3: The case = 0.125: Frame 1=w
n
, Frame 2=w

n
, Frame 3 =w

n
and
Frame 4=W
n
20 40 60 80 100
t
0.5
1
1.5
2
2.5
3
wn vs Wn
Figure 4: The case = 0.125: the lower one is W
n
, and another one is w

n
22