LIBRARY OF THE
UNIVERSITY OF ILLINOIS
AT URBANA-CHAMPAIGN
510 .fc*
TS
0.
iii. The error p_(t, e) = u - Au in approximating the partial
derivatives by finite differences is controlled at t. such
J
that AAt. I |a(t . , e) I I < e/2 where At . = t . n - t . . In
J — J 2 j j+1 j
practice, this is too much to expect. At the end of this
section, we discuss the effect of relaxing this requirement
to make it more realistic.
Let A. denote A(t., e). In the event that A( t . , e) 4 A(t , e) define
J J J J -*-
A(t" e) = A(t e) = A
Let v denote the exact solution to
v. = XA(t, e)v (2)
— t —
We obtain approximations w. to v. = v(t., e) by numerical integration using
tJ J J
the trapezoidal rule
' *i+i = v -i + 1/2 Xit J !A(t j^ + "Vi'^n 1
= w + 1/2 XAt A(t )[w + v ] (3)
J J J J J •*-
Define t. to be the local truncation error of the trapezoidal method in (3).
-J
That is, if w, = v., then t. = v ! _ - w... Assume for the moment that we
-J -J -J -J+l -J+l
can control the stepsizes {At.} such that for all j £ , x. L. = e/2. Later,
J ' ' - J ' ' 2
we will examine the effect of slightly weakening this assumption. Let
z = v - ¥ be the difference between the exact and numerical solutions to (2).
We will now derive a bound for z L. , then address ourselves to the question
1 '— n' ' 2
of bounding |u - vj | , and lastly combine the result of the two analyses to
arrive at an estimate of the global error | |u - w| | for the program.
In solving (2) by the trapezoidal rule, each time a step is taken,
the current global error is multiplied by exp(At.XA.), and a new local
J J
truncation error is added. We thus have
z.,_, = exp(At .XA. )z. + t. (h)
-J+l . J J -J "J
Assuming that w. - V-. then z = 0, and applying (k) , we have
"0 — - "0 —
n-1 n-1
z = i 1 - i • TLi exp(At .XA. )]t.
Since the A. ' s are symmetric negative definite, I exp(At .XA. ) I I _ $ 1. Using
l ' ' J l 2
this fact and the assumption that It.IL = e/2, we obtain
N^ll 2 « °Jo e/2 " ne/2 (5)
We now examine |u - vj | Q . Using the above definition of o_(t, e),
we can rewrite (l) as
u. = XA(t, e)u + Xa(t, e) (6)
— t — —
Subtracting (2) and integrating, we have
t t
|u(t ) - v(t )|L $ |/ n exp[X/ n A(s', e)ds']Xo(s, e)ds|L
— nn^Q s — d
We again use the fact that A(s', e) being symmetric negative definite implies
t
||exp[X/ n A(s', e)ds']|| * x to S et
s c -
t t
|u(t ) - v(t )| L $ / n I |exp[X/ n A(s», e)ds']||l |Xa(s, e) | Lds
— n— n^Q s d — d
t
$ / n ||Xa(s, e)||_ds
d
We approximate / J | Xo_( s , e) | | ds by At . X | |o_(t . , e)|| . Assuming that the
t ^ ^
J
error in this approximation is small enough to "be neglected, we can use the
assumption that XAt.||o_(t, e ) | | <: e/2 to get
J <-
n-1 t A +1
u(t ) - v(t I < e/2 .1. f J ds/At. = ne/2 (T)
'- n — n '2 J=0 t J
J
Finally, we combine (5) and (7) to get a bound for |u - w| | . Using the
triangle inequality, we get
|u - w| | $ ne (8)
Since we are using the trapezoidal rule to integrate (2) numerically, it is
reasonable to expect that Ix-llp = 0(AtT > ). But, since | t_. | | = e/2, it
3 1/3
follows that At. = 0(e) and hence, that At. = 0(e ). If we now assume that
J J
-1/3
At. = 0(l/n) (as it ought to be), then we have n = 0(e ). Substituting
J
this relation in (8), we get
|u - w|| 2 $ ke 2/3 (9)
for some constant k.
In practice, we cannot expect a program to actually control o_(t., e)
J
and x_. as strictly as was assumed in the above analysis. We will now look
at the consequences of reducing these expectations to a realizable level.
Since u is smooth and A(t, e) is piecewise smooth in t, we can express the
truncation error of the trapezoidal method at t . , as a power series in At.
J+l J
00
x. = e x. .At:
-j 1=3 -tj
Instead of requiring that ||t_JIo = £ /2 5 we require that I|t At.|| $ e/2.
In practice, we do not know t_ exactly, but we will assume that we can
estimate it accurately enough so that the following conclusions are not
altered. Since | t_ A"t . | L $ e/2, we may infer that At. = 0(e 1/3), and
-30 J u
hence
| 1 t_ | | = | |t At^| I + o(At^) * e/2 + o(e) (10)
The (at least) piecewise smoothness of u and A(t, e) allows us to
00 "1
similarly write a(t, e) on [t., t., n ) as a(t, e) = .1 a.(t, e)Ax . We
- j o+l - T=p. -l
will no longer purport to control o_ such that XAt . | |£.(t., e)|| < e/2.
J J *-
Instead, we only require that
P i, ,
XAt. o (t., e)Ax J L $ e/2 (ll)
'-p. 2
J
Once again, in practice, we only have an estimate of a (t., e), but we will
assume that the estimate is sufficiently accurate for the discussion to
1/3 2/3p l
remain unaltered. Since At. = 0(e ) , it follows from (ll) that Ax = 0(e J )
and hence,
1+2/3P,
At. | |a(t, e)| | S e/2 + o(e J ) (12)
J
We can therefore, still expect the results of the previous analysis to hold,
even when the requirements are less stringent.
3. Programming Application
A version of the numerical procedure analyzed in the previous section
was implemented in a FORTRAN IV program. This section "begins by giving some
of the particulars of its operation. For a more detailed account of its
functioning, see the block diagram which has been included in the appendix.
The program was written specifically to solve the heat equation for various
initial conditions. It approximates the second spatial derivatives using
centered differences and uses the trapezoidal rule to integrate with respect
to time. It was numerically verified that all of the centered difference
formulas used give rise to symmetric matrices A( t , e) which are negative
J
definite.
Prior to attempting an integration step, an estimate of the local
spatial truncation error is made using centered differences on the current
numerical solution to estimate the magnitude of the higher order derivative
which appears in the leading term of the Taylor expansion for the error.
If the estimated error exceeds e/2 , the current centered difference formula
for the second derivatives is replaced by another centered difference
formula which is two orders of accuracy in Ax higher. The local spatial
truncation error for the new formula is estimated, and if its magnitude
still exceeds e/2, the process repeats with the centered difference formula
that is higher in order by yet two more powers of Ax. This continues until
either a formula is found for which the estimated error is less than e/2 in
magnitude or the order of the formula reaches a limit specified in the
initialization section of the program.
If, on the other hand, the estimated spatial truncation error is much
less than e/2, an estimate is made of what the spatial truncation error would
8
be if the current centered difference formula were replaced by a centered
difference formula whose accuracy is two orders lower. If the estimated
error for this lower order formula still is less than e/2 in magnitude, the
lower order formula replaces the current formula.
An integration step in t is then taken, and the temporal truncation
error is estimated using step-doubling. This procedure involves taking a
step of size At and comparing the result with the solution obtained by
taking two steps of size At/2. If the estimated error exceeds e/2, At is
adjusted downward, and the integration step is reattempted with the adjusted
At. If, instead the estimated error is less than e/2, the step is accepted,
and At may possibly be increased for the next step.
The program was used to solve the heat equation
u, = Xu
t xx
with the diffusivity constant X = 1/10 subject to the initial condition
u(x, 0) = f(x)
We assumed f(x) was periodic with period 2 and antisymmetric with respect to
x = 0, ±1, ±2, ... . Hence, we can restrict our attention to the interval
[0, l]. This reduces the run time and also permits us to use centered
differences at all mesh points in [0,l], Under these assumptions, with
boundary conditions
u(0, t) = u(l, t) = 0,
the solution is
oo 2 2
u(x, t) = Z- b exp(-n tt Xt) sin nnx
n=l n
where b = 2 / f(x) sin nirx dx
n
The following initial conditions were used:
Case 1. f(x) = sin ttx,
b = 1, n = 1,
n
0, n > 1.
Case 2. f(x) =1 < x < 1.
b = U/(m0 , n odd,
0, n even.
Here u(x, 0) is a square-wave.
Case 3. f(x) - kx, $ x $ .25,
1, .25 S x * .75,
h - kx, .75 $ x $ l.
b = 8 (sin mr/U + sin 3nir A ) , n odd,
n —
nu
0, n even.
Here u(x, 0) is an isoceles trapezoid on [0, l].
Case U. f(x) = x/a, < x £ a,
= 1 - (x - a)/(l - a), a { x $' 1.
2 2
b = 2 sin (mrp)/[n tt p(l -a)].
. Case Ua. p = 1/2
Case Ub. p = 7/10
Case Uc. p = 9/10
Case Ud. p = 999/1000
Here u(x, 0) on [0, l] is a peak with its summit at x = a. As a
approaches 1, the slope to the left of p slowly decreases, while
10
the slope to the right of y becomes steeper at an ever increasing
rate. For y close to 1, u(x, 0) is very nearly a sawtooth wave.
_2
The program was run for each of the above cases with e's of 10 ,
-3 -8 -9
10 , . . . , 10 except for CASE 1 where it was also run for e = 10 and
e = 10 . For all cases except for CASE 1, the integration was begun at
t = 1/10 rather than at t = in order to eliminate the difficulties arising
from the discontinuities at zero. The initial number of points in the
centered difference approximation to the second derivative was three. The
initial settings for At and the number of internal mesh points were .01 and
19 respectively. Integration was halted as soon as t exceeded 2.
11
k. Results
The results of the tests described above are summarized in the tables
which appear in this section. An explanation of each column that appears in
the table is given below:
e_ - On each integration step, the magnitude of the estimated spatial and
temporal truncation errors are both required to be less than or equal
to e/2.
#_ OATT.fl to RKSTEP - Each time that the subroutine RKSTEP is called, an
integration is performed using the trapezoidal rule. Since the
program uses step doubling, each time a step is attempted, the number
of calls to RKSTEP is increased by three.
#_ CALLS times #_ POINTS - Each time RKSTEP is called, the current number of
internal mesh points is added to this number. It thus provides a
rough measure of the work done by the program.
LOG 10 DIFFERENCE in WORK - The difference between the log (base 10) of the
entry from the current row of the tf CALLS TIMES # POINTS column and
the log (base 10) of the entry from the preceding row in that column.
This may give a better idea of the rate at which the work done by this
program increases for each factor of ten decrease in e.
INTERNAL MESH POINTS - The number of equispaced points {x. } (exclusive of
x^ = and x ,_ = l) at which the solution is computed. If the
n+1
program has difficulty in taking the first step successfully, it
may attempt to raise the number of points .
INITIAL COUPLING - The number of points used in the centered difference
approximation to the second derivative when the first successful
step was taken.
12
FINAL COUPLING - The number of points used in the centered difference
approximation to the second derivative when the upper limit of
integration was reached.
13
e
H W
« &
O E
^ K
M CC
< O
M K
K ft;
s « w
a -p
M <
o
M
C3
O
o
En
W Eh
CO g
O
o s S
H H O
op*
q E s
t-3 l*H H
H
Q
CO
CO CO Eh
H H S
J S H
W
o s
Ph
i — 1 H
o
on
CM
ON
t—
CO
H
Pn
13
LTN
H
ir-
CM
en
on
o w
o
CM
on
-*
-=t
-=f
3 Pn
H
■«
CO
CO
EH
t-3 CO
a
ON
LTN
on
IT)
LTN
H
_ht
3g
H
CO
VO
CO
On
on
on
t-
O
LTN
VO
o
LTN
ON
On
CM
O H
Ph
H
CM
VO
CO
H
B
H
LTN
Sfc
sfc
a
Ph
LTN
H
-=f
.si-
On
On
t-
3g
EH
-3"
LTN
CO
t-
VO
CO
O
CO
in
en
t—
C—
o
£
H
*
H
CM
on
-5f
LTN
VO
t-
u
H
W
H
w
H
H
W
H
H
H
H
H
H
H
CM
Ps
15
x < o
O H
Fn
-P
<
O
H
E
p
o
o
o
H
h3
o
o
En
en
EH
H
o
PL,
o
ro
ro
H
h-J
CQ
S
sal
g
H
O
o
H
H
Ph
^te
3fc
CO
Pn
i-l
M
3
o
H
CO
on
C\J
on
o
CO
o
on
H
LT\
LTN
VO
t-
o
I
1
vo
1
VO
vo
1
vo
1
on
i
-=r
oo
on
on
I
J-
J-
CM
I
LTN
I
W
o
H
CM
LTN
I
w
o
on
i
w
On
o\
on
on
i
w
CVJ
on
CO
I
vo
On
J-
I
LTN
H
CVJ
l/N
I
W
o
t-
-3-
o
CM
on
on
CM
c—
vo
O
a.
on
o o
o\
H
ON
ON
ON
ON
CM
CM
LTN
I
w
LTN
I
w
CM
O
D—
H
CO
CM
H
LT\
H
LTN
LT\
On
LTN
CO
VO
rH
c—
H
CM
on
H
H
H
vo
I
VO
H
CM
VO
I
VO
H
CM
H
vo o
vo On
H O
o o
t—
vo
on
j- H vo on vo ltn
VO On ON ON CM J"
o h cm on j- j-
LT\
H
on
on
H
CM
H
H
rH
on
VO
CM
vo
CM
CM
-3-
ON
on
j-
vo
en
CO
On
ON
rH
CM
LT\
CM
H
^t
LTN
VO
C—
3
H
H
H
on
g
CO
Eh
P
CO
on
EH
16
s -p
M <1
W
H
W
O
o
K CO
W g
En S
EH
W
O
o as
H pa
o
Ph ^
W s
W M
H
Q
a
CO
CO E-t
i-5 co a
d @ M
< U o
O M PL,
en
H
OJ
OJ
no
H
CO
j*
LTN
t~
LTN
t-
o
on
NO
NO
NO
NO
oo oo .j-
I I i
WWW
un co j-
H H oo
I
w
On
H
ir\ no
I I
W W
O CO
J- CO
00 OO
I I
w w
O OJ
-=r cm
-3-
l
w
CO
1"
w
On
H
ltn
I
W
o
M3
I
w
CO
CO
LTN
CM
LTN
CM
On
H
O
On
O
t—
O
H
CM
O
OO
ON
ON
ON
-3- CM NO
00 CO O
O H 00
LfN 00
t- H
On
H
I
CM
W
oo
w
-3-
LiA
w
w
NO
w
NO
I
w
ON
NO
I
w
On
H
-3-
oo
CO
o
o
On
ON
CM
CO
NO
H
H
CM
CM
o-i
O ON t—
On H ltn
00 -J" _=f
t—
ON
t—
LTN
t-
NO
O
LTN
LTN
CO
-H-
CO
H
OO
o
CO
H
CM
NO
ON
o
OO
H
LTN
o
OO
00
NO
CM
NO
LfN
o
H
CM
LTN
CM
H
I
w
LTN
•
II
•H
IS
Ph
a3
W
o
Ph
o
w
CO
En
5
CO
w
Ph-
w
<; s k
S M K
fin W
s « w
13 -P
H <3
1*1
3 h
H
S O
H O
En
5
CO
H
O
PL,
O S ffi
h g o
o E s
J fe H
m
a
CO
3
s
H
O
o
H
LL,
^fc
5»fc
ro
a<
h-1
m
!5|
O
EH
H
CO
o
IT
OO
t-
H
CO
O
-d-
H
oo
vo
l#\
t—
t-
o
1
CM
1
VO
1
VO
1
VO
1
VO
1
ro oo
I I
W H
t- H
CO
CM
ltn
LTN
I
W
ON
J"
OO
VO
I
W
oo
I
w
o
oo
I
w
oo
CM
-3-
i
w
oo
CM
I
ON
LTN
I
w
o\
oo
vo
I
w
vo
o
LTN
CM
CM
CM
t-
O
CM
O
O
ON
o
CM
OO
VO
ON
00
t—
ON
H
H
On
H
ON
ON
On
OO
CM
On
CM
VO
o
CM
-=t
H
CM
oo
_=r
H
H
H
^
VO
I
w
oo
LTN
VO
I
w
OO
LTN
H
o
c—
VO
oo
C— VO LTN t—
0O H CM CM
CM -3" _=T -3"
H
OO
On
-=t
VO
u^
LTN
ON
CM
O
ON
On
LTN
OO
t-
LTN
ON
o
H
_=T
H
H
CM
00
vo
CM
ON
\s\
r-\
CO
LTN
ro
-4"
VO
00
ON
H
OO
H
CM
VO
OO
H
vo I~-
I I
H H
II
-P
Pm
O
CO
&H
w
Eh
18
H
< P2
CM
-=)•
00
CO
_=r
H
H H
55 O
J-
-=r
^t
ir\
CO
CO
K
55 H K
H
00
\o
vo
VO
vo
hIhK
P£
+
I
1
1
1
1
hJ Ph
^
H
*^!
O
^
Ph
o
-=r
00
1
^*
-H-
LfN
VO
VO
>$^
H
1
Ph
Ph
w
1
pq
1
Ph
1
S 5?
Ph
O
o
CM
t—
H
-=f
o
S M
Ph
VD
oo
O
oo
O
CM
oo
ph
W
on
H
V£>
rH
OO
vo
H
oo
00
J"
J"
LfN
vo
vo
K
Ph
I
I
1
1
1
I
1
38
£
H
w
W
Ph
Ph
W
H
LTN
OJ
CM
t—
H
-=t-
o
S K
W
C\J
CM
O
OO
O
CM
00
W
t—
CM
VO
H
00
VX)
H
3
VO
^H-
C—
vo
O
H
CM
O
t-
CO
o
55 -P
LTN
-h"
CM
ON
J"
<-\
H
H <
CO
CM
CM
o
o
o
o
P>H
O
S
33
on
LTN
LTN
t-
t—
t-
t-
Hh£
h o
o
o
%n
H J
LTN
t-
ON
H
3
H
H
^£
H
H
H
s o
M O
3
g w
e
Ph CO
55
ON
On
ON
O
LTN
CM
O
w g
H
H
H
H
CM
CM
00
_=*
Eh S
55
£
M
8
o s
£
H H
O
VO
CO
O
VO
00
OO
Ph
Es
LTN
00
00
H
_3"
CM
O Ph
O
CM
oo
-H-
_=r
J-
J Ph
R
M
fi
CO
CO
EH
>-3 CO
55
H
c—
00
On
J"
LTN
ON
H
LTN
CM
CO
H
J-
LTN
c—
5 < Q
< a k
S M K
Pn W
3
a -P
M <
o
a
M
§
D
o
u
19
CO
ro
rH
t-
rH
ON
lo
.*
OJ
On
LTN
t-
rH
oo
C—
ion
NO
NO
i
w
OO
NO
OO
CO
w
LTN
OJ
i
w
00
NO
US
J-
LA
w
W
00
CO
o
t—
H
C\J
no
p4
H
OJ
no
OO
I
_H/
On
NO
ro
i
w
H
OJ
OJ
_H/
I
00
no
LA
I
w
o
H
LA
I
w
OO
t-
OJ
NO
I
w
H
OJ
no
o
LA
OJ
OJ
OO
CO
CO
o
_H/
o
OJ
o
en
NO
w
o
CO
NO
I
Pm
o
ro
On
o
On
o
o
t—
o
^n
M 1-3
LA
t-
H
H
H
rH
H
EH Ph
r-l
H
H
rH
H
H P
a O
H O
3
CO
a w
EH
K CO
a
On
ON
On
OJ
CO
NO
LTN
W g
M
rH
H
H
OJ
OJ
OO
_=r
s
H
g
o a
pq
H w
o
OO
NO
OJ
-=*
LTN
NO
pq
^
LA
On
o
OJ
ro
OJ
CO W
O P«H
O
H
-d-
-=f
-=r
_=r
a
j E
H
M
Q
J CO
CO
EH
a
ON
LTN
LT\
c—
H
OO
^t
53 H
H
CO
NO
^*
OO
On
NO
OJ
o
LTN
NO
o
NO
On
o
CO
O M
ft
H
OJ
NO
On
o
EH
H
LTN
=Jfc
=»fc
CO
Ph
t-q
H
US
H
H
t-
OJ
OJ
OJ
3£
EH
J"
LTN
CO
t-
t—
On
Os
CO
H
OO
t-
NO
O
K
rH
=tfc
H
OJ
OO
-=r
us
NO
t—
(J
W
w
W
H
pq
W
w
H
H
rH
H
H
H
H
CO
On
On
On
-P
•H
o
p^
o
(in
CO
EH
CO
3
20
FINAL At - The stepsize when the upper limit of integration was reached.
MAX ERROR EVER - The magnitude of the worst error in any individual component
of the numerical solution vector that was ever found. The numerical
solution was compared to the exact solution on each step that t
equalled or exceeded an integral multiple of 2/10.
MAX FINAL ERROR - The magnitude of the worst error among the components of the
final numerical solution vector.
LOG 10 DIFFERENCE IN MAX FINAL ERROR - This column is to MAX FINAL ERROR as
LOG 10 DIFFERENCE IN WORK is to § CALLS TIMES # POINTS.
Results indicate that the idea of controlling the global error through
separate control of the local spatial and temporal discretization errors is
indeed quite feasible. Of particular interest is the LOG 10 DIFFERENCE IN MAX
FINAL ERROR columns. These show that for each factor of ten decrease in e,
2/3
the global error decreased by a factor which appeared to approach (10)
This is as predicted in the previous section.
The LOG 10 DIFFERENCE IN WORK columns show that as e was successively
decreased by factors of ten, the amount of work done increased by factors
1/3 1/2
which tended to stabilize in the range of (10) to (10) with the exact
number depending on the initial condition. For CASE 1, where it was never
found necessary to increase the number of internal mesh points, the number was
1/3
precisely (10) . This is what would be expected with the trapezoidal rule.
Since the local truncation error is proportional to the cube of the stepsize,
and the program attempts to keep the local errors on the order of e, if the
-1/3
mesh is fixed, the number of integration steps performed should be 0(e ).
Combining the relationships between e and the global error and between
e and the amount of work done by the program, we can get a relationship
21
between the global error and the work. Denoting the global error by e and
the work by q, we can write
2/3
limit e(e) = c e
E -*■
limit q(e) = c e~ m 1/3 ^ m $ 1/2
e -*■
and hence,
limit e(q) = c q P -2 £ p $ -k/3
e -> J
where c. , c_ , and c are constants.
While this may not seem like a particularly good return of accuracy
for the computational effort expended, it compares quite favorably with most
of the methods commonly used to solve partial differential equations. The
well-known Crank-Nicolson method which for the heat equation gives rise to
the system
n+1 n . ,_. , / .2 n „2 n+l N
u. = u. + 1/2 Ar( 6 u. + 6 u. )
i i x l x i
where u. = u(x. , t), r = At/Ax , and 6 is the standard 3-point centered
11 x
3 2
difference, has a local truncation error of 0(At + At Ax ) . The Douglas
formula is the most accurate formula involving the same six points to approx-
imate u (x., t ._). It generates the system
xx i n+1
n+1 _ n , , x 2 n , x 2 n+lx 1 ,,.2 n x 2 n+l x
u. = u. + 1/2 Xr[& u. + 6 u. ) + r-r X( 6 u. - 6 u. )
i i xi x l 12 xi xi
for the heat equation which results in a local error- of 0(At + At Ax ) . If,
as At approaches zero, r is kept fixed, the local error is 0(At ) . The global
2
error should therefore be 0(At ). This means that
limit e(At) = 0(At 2 )
At ->
just as for the trapezoidal rule.
In advancing the solution one time step, Douglas' method must solve a
22
tridiagonal system. The amount of work necessary to solve such a system is
proportional to the number of equations in the system and hence to the number
of internal mesh points. But since the number of internal mesh points is
always within one of Ax , we may take the work proportional to Ax . The
number of steps that must be taken and hence, the number of times the system
must be solved is proportional to At . Hence, for Douglas' method, Ax At
may be taken as a measure of work analogous to the measure # CALLS TIMES #
POINTS described above for the program. If, as before, r is assumed to remain
fixed as At varies, for the Douglas method we have
limit q(At) = 0(Ax _1 At" 1 ) = 0(At~ 3 ' 2 )
At ■*■
and hence,
limit e(q) = 0(q /3 )
At +
which is no better than the worst case observed for the program.
23
These include X,
e, At . , At
mm max
H me:.h points
max
and coupling
c
Sturt
3
Read in
various
parameters
ffor programs
Coupling *
Coupling + S
T.etting
starting
values, etc.
Various
.initializations)
Coupling is tne
number of points
in the difference
approximation to
the second
spatial derivative
Coupling * '',
Uart « n
Xo is the magnitude
of the largest
component in the
estimated spatial
truncation error
Calculate
Start
Set up A
matrix for
current
coupling
'Calculate LU
' decompositions
for I
T x '
and T +
fit ,
£
C
Take step
of size At
and store
result in
varray YWH0LE y
Stop
3
V
2U
n mesh points *
Biech points
Calculate HSAT,
the smallest num-
ber of internal
mesh points r. ,t.
Xito < e/2
mesh points
NSAT
Subscripts max
or "min" indicate
maximum or
minimum per-
missible values
<3
Take 2 steps
of size At/2
and store final |
result in
array YNEW
Calculate t(At)
= 1/ 3*Max | YNEW( i ) -r.mOLEC i )
i
T * T + it
YES
,/Print message^
\ and STOP J
SAT
Set it
SAT
Max(4t SAT' A W At/5)
25
Print computed
solution in
YNEW, the
exact solution, t
and the errors
printout
time *
printout
time + .2
YOLD ■*■ YNEW
Calculate
FACTO R#l= factor
by which At
could be in-
creased and rse/2
•still hold
I
FACTO PJl «-
.8 * FACTO R#l
FACTO R#l
FACTO R#l
FACT0PJ1
Min(FACTOR#l, 5)
Max(FACTOH#l, l)
Min( FACTORS!, At
max
/At)
Calculate
At \o
\
F
FACTOR #2 «-
e/(2 At Xt)
26
Calculate
At Xa for
coupling - 2
Coupling ■*-
coupling - 2
FACTOR «-
Min( FACTO R#l, FACT0R#2 * .8)
ir
At +
At y FACTOR
\
1
c
FormAEC-427 U.S. ATOMIC ENERGY COMMISSION
< 6/68 > UNIVERSITY-TYPE CONTRACTOR'S RECOMMENDATION FOR
AECM3201 DISPOSITION OF SCIENTIF:C AND TECHNICAL DOCUMENT
( See Instructions on Reverm Side )
1. AEC REPORT NO.
COO-2383-0025
2. TITLE
AUTOMATIC INTEGRATION OF THE HEAT EQUATION
3. TYPE OF DOCUMENT (Check one):
tf l a. Scientific and technical report
] b. Conference paper not to be published in a journal:
Title of conference
Date of conference _^_
Exact location of conference _
Sponsoring organization
□ c. Other (Specify)
4. RECOMMENDED ANNOUNCEMENT AND DISTRIBUTION (Check one):
ET) a. AEC's normal announcement and distribution procedures may be followed.
~2 b. Make available only within AEC and to AEC contractors and other U.S. Government agencies and their contractors.
3 c. Make no announcement or distrubution.
5. REASON FOR RECOMMENDED RESTRICTIONS:
6. SUBMITTED BY: NAME AND POSITION (Please print or type)
C. W. Gear
Professor and Principal Investigator
Organization
University of Illinois
Department of Computer Science
Urbana, IL 6l801
Signature
^i\j2l t0 Cft-r
Date
October 3, 1975
FOR AEC USE ONLY
7. AEC CONTRACT ADMINISTRATOR'S COMMENTS, IF ANY, ON ABOVE ANNOUNCEMENT AND DISTRIBUTION
RECOMMENDATION:
8. PATENT CLEARANCE:
I I a. AEC patent clearance has been granted by responsible AEC patent group.
I I b. Report has been sent to responsible AEC patent group for clearance.
I I c. Patent clearance not required.
BIBLIOGRAPHIC DATA
SHEET
1. Report No.
UIUCDCS-R-75-75 1 *
3. Ret ipienl '■• Ai i i isii in No.
4. Tn It- and Subt it 1 1-
AUTOMATIC INTEGRATION OF THE HEAT EQUATION
5. Report i). iti-
October 1975
7. Autrtor(s)
Michael H. Ostrar
8. Performing Organization Rept.
No " UIUCDCS-R-75-75 1 *
9. Performing Organization Name and Address
Department of Computer Science
University of Illinois at Urbana-Champaign
Urbana, IL 6l801
10. Project/Task/Work Unit Nc
11. Contract /Grant No.
US AEC AT (11-1) 2 383
12. Sponsoring Organization Name and Address
US AEC Chicago Operations Office
9800 South Cass Avenue
Argonne, IL 60^39
13. Type of Keport & Period
Covered
14.
15. Supplementary Notes
16. Abstracts
A scheme for automatically integrating the heat equation is discussed.
A program based on this scheme is explained, and the results of integrating the
heat equation for a number of different initial conditions are presented and
interpreted.
17. Key Words and Document Analysis. 17a. Descriptors
17b. Identifiers /Open-Ended Terms
17c. COSATI Field/Group
18. Availability Statement
Unlimited
19. Security Class (This
Report)
UNCLASSIFIED
20. Security Class (Thi
Page
UNCLASSIFIED
21. No. of Pages
33
22. Price
FORM NTIS-35 ( 10-701
USCOMM-DC 40329-P7 1
*««
% 1*
^
i&
wMm^MMmuuDtiuiuumi