Reprinted fror:n November 1982, Vol.

104, Journal of Engineering for Industry

0. Masory
Research Engineer,
Gould Laboratories,
Rolling Meadow, Il l. 60008,
Assoc. Mem. ASME
Rteference-Word Circular
Interpolators for CNC Systems
Y. Koren
Goebel Professor,
Department of Mechanical
Engineering,
Cenler for Robotics
and Integrated Manufacturing,
University of Michigan.
Ann Arbor, Mich. 48109
Mem.ASME
Interpolation techniques for CNC manufacturing systems are either of the
Reference-Pulse or Reference-Word type. Reference-Pulse interpolators have been
discussed in a previous paper [2]. The present paper provides an analysis of a
number of Reference-Word circular interpolation techniques that are
us.efu/ m CNC systems. These interpolators junction in an on-line iterative mode
and generate binary words which are supplied as references to the control/oops in
CNC sys.tems. various methods are evaluated in terms of ac-
cu,racy, maxzmum permztted radzus, and number of required iterations to complete
a circular arc.
Introduction
The present paper is the last in a series of three papers
dealing with the design concepts of computer numerical
control (CNC) systems [1, 2]. The first pape:r [1] presented
two principal structures of control systems denoted as
Reference-Pulse and Sampled-Data systems. With the first
technique the computer produces a sequenc:e of reference
pulses for each axis of motion, each pulse gene:rating a motion
of one basic length-unit (BLU) of axis travel. These pulses can
actuate a stepping motor in an open-loop system, or be fed as
a reference to a closed-loop hardware system. With the
Sampled-Data technique, the control loop is closed by soft-
ware through the computer itself, as shown in Fig. 1. The
control program compares a reference binary word with the
feedback signal to determine the position error. The error is
fed at fixed time intervals to a digital-to-analog converter
(DAC), which in turn supplies a voltage proportional to the
required axis velocity.
Both of these techniques require distinct: interpolation
routines to generate their corresponding references: pulses or
binary words. The interpolator routine coordinates the
motion along the machine axes, which are separately driven,
to generate the required machining pa!h. The machining path
is usually obtained from a combination of linear and circular
segments, and accordingly the control program contains both
linear and circular interpolation subroutines. Tlhe interpolator
routine functions in an on-line mode and is usually written in
assembly language in order to improve computing speed.
Although linear interpolators can be easily written in
assembly language, difficulties arise in assembly language
programming of circular interpolators which necessitates the
on-line solution of a second order equation.
The previous paper [2] in this series was concerned with the
evaluation of circular interpolators for Reference-Pulse CNC
systems. All Reference-Pulse interpolators are based on an
iterative technique controlled by an interrupt dock. At each
Contributed by the Production Engineering Division for publication in the
JOURNAL OF ENOINEERINO FOR INDUSTRY and presented at. the 1981 Winter
Annual Meeti ng, November 15- 17, 1981, Washington, D.C., of the AMERJCAN
SOCIETY OF MECIIANICAL ENGINEERS. Revised manuscript received at ASME
Headquarters June 15,1982.
400 I Vol. 104. NOVFMRFR 1Q.R?
Fig. 1 A singleaxis sampleddata system
interrupt, a single iteration of the interpolation routine is
executed, which in turn can provide an output pulse that
advances the corresponding machine axis by one BLU.
Therefore, the maximum attainable feedrate, or axis velocity
in BLUs/s, is inversely proportional to the execution time of a
single iteration. It has been shown that the appropriate
Reference-Pulse interpolation technique for machine tool
systems is the software DDA [3], since the uniformity of the
feedrate along the circular path is obtained only with this
method. Using a PDP-11 as the CNC controller, an in-
terpolation iteration time of 52 p.5 has been obtained [2],
which allows a maximum velocity of 19,230 BLUs/s. A BLU
might be on the order of 10 Jtm in a typical machine tool
system, resulting in a maximum feedrate of about 11 m/min,
which is adequate for machining applications. However,
modern CNC systems employ a single axial drive motor for
both idle motions and machining, and in these systems a
velocity of 11 m/min is considered to be too low for idle
motion.
In contrast, the maximum velocity in Sampled-Data
systems is not limited by the computer. In these systems the
circular interpolation is based upon approximating a circle
with straight line segments. For each segment the interpolator
generates a reference word proportional to the local axial
velocities which are transmitted to the corresponding software
loop-comparators of the control axes. The present paper
evaluates Reference-Word circular interpolator techniques
which can be applied to manufacturing systems permitting the
simultaneous operation of two axes-of-motion.
The Concept of Reference-Word Interpolators
In circular interpolation, the simultaneous motion of two
y
~ ~ U L ~ ~ ~ ~ X
Fig. 2 Definitions for the referenceworcl technique
axes generates a circular arc at a constant t:angential velocity,
or feedrate, V
0
The axial velocities satisfy the following
equations:
where
Vx(t) = V
0
sin O(t)
Vy (t) = V
0
cos O(t)
O(t) =(V
0
1R)t
a n d ~ is the radius of the circular arc.
(1)
The velocity components Vx and Vy are: computed by the
circular interpolator and are supplied as r-eference inputs to
the computer closed-loops. The circle generated in this case is
actually comprised of straight line segments . At the beginning
of each segment the references are supplied by the in-
terpolator and the end of the segment is located with the aid of
a feedback signal. Increasing the number of these segments
improves the accuracy of the generated cilrcle, but increases
the number of iterations, thus requiring more computer time.
The optimal number of segments is the smallest one which
maintains the path error within the required limit of one BLU.
Each iteration of the algorithm corresponds to an angle a,
as illustrated in Fig. 2. The choice of the a:ngle a depends on
the interpolation method. All of these methods employ the
difference equation:
cosO(i+ 1) = AcosO(i) - BsinO(i) (2)
sinO(i+ 1) = AsinO(i) +BcosO(i)
where the coefficients A and B are given by:
A=cosa B=sina (3)
and
O(i+ l)=O(i) +a (4)
The corresponding segment is terminated at the point X(i + 1),
Y(i + I) which is approximated by:
X(i+ 1) = R(i)cosO(i + 1) (5)
Y(i + 1) = R(i) sinO(i+ 1)
Substituting equation (2) into (5) yields:
X(i+ l) = AX(i) -BY(i) (6)
Y(i+ 1) =AY(i) +BX(i)
Equation (6) is the basic relationship which permits the
calculation of a successive point based on the present one.
The main differences between the variious interpolation
methods are in the determination of a and in approximation
of the coefficients A and Bin equation (3). Once the angle a is
chosen, the interpolator routine proceeds, lfor each iteration,
as follows:
1 At each point X(i), Y(i) the interpolator calculates the
Journal of Engineering for Industry
coordinates of the successive point X(i + 1), Y(i + 1) ac-
cording to equation (6). The segment lengths are:
DX(i) = X(i + l) - X(i)=(A- l)X(i)-BY(i) (7)
DY(i) = Y(i + l) - Y(i)==(A - l)Y(i) + BX(i)
and the corresponding velocities are:
where
Vx (i) = VoDX(i)IDS(i)
Vy (i) = V
0
DY(i)IDS(i)
(8)
2 The values obtained from equation (7) are the in-
cremental positions, and those of equation (8) are the
velocities, or the reference-words, for t he present segment.
These values are supplied to the software control loops.
3 Upon completion of the segment, the routine increments
the coordinates X(i) and Y(l) .
Since the angle a is relatively small,, the chord length DS can
be approximated by its arc length Ra and the calculation of
the velocities Vx and Vy can be simplified to:
Vx(i) ==K DX(i) (9)
Vy (i) =K DY(i)
where K = V
0
/ Ra. The parameter K is a constant, which is
calculated only once for each circle, and consequently t he
reference words are actually proportional to the segment
lengths.
Approximation of a circle by straight line segments causes
two types of errors as illustrated in Fig. 3:
(a) Radial error ER, due to the truncation effects:
ER(i) = R(i) - R=,lX
2
[i)_.f._([} -R (10)
where R is the required radius.
(b) Chord height error EH defined by:
EH (i) =R - R (i) cos(a/2) (II)
Equation (10) is explicitly obtained using the approximated
values of A and B, which causes the truncation error in t he
radius. Since the algorithms use an iterative technique, this
error is accumulated with time. The effect of the truncation
error ER at the ith iteration can be approximated by (see
Appendix 1):
ER(i) = i(C- l)R
(12)
where
C=.JAT+lfi
Equation (12) shows how the truncation error increases with
the number of iterations. The largest circular arc produced by
a single instruction in NC systems is a quarter of a circle. The
compatible number of iterations is:
N = 1rl2a (13)
Upon completing this number of steps the error ER reaches its
maximum value:
ERmax =(7r/2a)(C-1)R
(14)
By contrast, the error EH is not cumulative. Using the for-
mula
cos(a/2) = .J(l + cosa)/2 = .J(l + .Af7i (15)
in equation (11), yields:
EH(i) =R-R(i) .J(l +A)/2. (16)
The angle a is selected so that either the error ER or EH
NOVEMBER 1982, Vol. 104/401
reaches a maximum value of one BLU. In addition to the
angle ex, the interpolator is given the coordinates of the initial
point [X(O), Y(O)] and the required feedrate V
0
These
parameters are obtained from the part-program for each
circular arc.
Several interpolators applying the reference-word method
are now discussed and their relative features are compared.
Euler Method
The Euler method is based upon a first-order Taylor series
expansion to equation (3) which provides the simplest ap-
proximation:
A = I ; B=cx (17)
The basic equations of the algorithm are obtained by sub-
stituting for A and Bin equation (7):
DX(i) = - aY(i)
DY(i) = cxX(i) (18)
Substitution of equation (18) in (9) gives the required
velocities:
Vx(i) = -(Vo!R)Y(i)
Vy (i) = (V
0
1R)X(i)
(19)
Since the series expansion is truncated after the frrst term, the
accuracy of the Euler method depends mainly on this trun-
cation error ER rather than on the chord error EH. The
maximum error is obtained by substituting equation (17) into
(14):
ERmax - 1)R (20)
For small a, equation (20) can be approximated by:
ERmax =(?r/4)aR (21)
which allows for calculating the value of a. To satisfy the
accuracy requirement, the maximum permitted error is one
BLU and the corresponding angle is:
a=4! (?rR) (22)
Consequently, the number of iterations required to Complete a
quarter of a circle is obtained from equation (13):
(23)
Improved Euler Method
The improved Euler method (IEM) [3] also us.es the ap-
proximation (17) but it applies the calculated value of X(i +I)
to determine the present value of Y(i + 1). With this
modification, equation (6) becomes:
X(i+ l)=AX(i) - BY(i)
Y(i +I) =A Y(i) + BX(i+ 1)
(24)
and with substitution of A and B from (17) the main
equations of t he (IEM) algorithm are obtained:
X(i+l) = X(i)-aY(i) (25)
Y(i+ I) = Y(i) +cxX(i+ I)
The segment lengths as defined in equation (7) are:
DX(i) = -aY(i) (26)
DY(i) =cxX(i+ I)
The improvement as compared with the previous method is
illustrated by deriving the explicit expression for Y(i+ 1) in the
second equation of (25):
X(i+ I) X(i) - aY(i) (27)
Y(i+ I) = (l-cx
2
)Y(i) +cxX(i)
402/ Vol. 104, NOVEMBER 1982
Table 1 Maximum radius in IEM
n[bits] Q p Rm!BLUs]
15 8 7 255
23 13 10 4095
31 16 15 32767
The actual average A which is used in equation (27) is given
by:
1 I
-[I+ (1 - a
2
)] =I- - a
2
2 2
(28)
rather than the value of I which is used for approximation of
cosa in the Euler method. Therefore, a better approximation
of the coefficient A is achieved with the IEM without any
additional computational effort.
The error ER reaches its maximum value for particular
iteration that satisfies the condition (see Appendix 2):
X(i) = (1 +a) Y(i) (29)
or at angle of 8 = 1rl 4 and the corresponding error is:
a
ERmax = 4 R
(30)
Since ER should not exceed one unit
a=41R (31)
The number of iterations required to produce a quarter of a
circle as obtained from equation (13) is:
N = 1rRI8 (32)
which is less than with the Euler method by a factor of 1r (see
equation (23)).
Although the algorithms of the Euler and Improved Euler
methods are relatively simple, they cannot be implemented
with basic assembler instructions such as ADD and SHIFT.
The computational difficulties arise in calculating the angle a
and in performing multiplications cxX and a Y. With the Euler
method, part of the difficulty in calculating the angle a is with
using the factor ?f (equation (22)), which is not needed with
the IEM (equation (31)). The second difficulty can be over-
come by expressing the angle a in powers of 2, i.e., a=2-m,
thereby permitting the use of SHIFT operations rather than
multiplications. In this case the angle a is calculated by:
a= 2- m s4/R
or
2m '
2
C!::.R (33)
The power m is the smallest integer which satisfies equation
(33) and is obtained by SHIFT operations. Thus, a complete
implementation of the IEM algorithm is achieved by basic
assembler instructions, without the necessity for either a
floating-point unit (FPU) or multiplication operations.
In spite of the attractive features of the IEM, special care is
needed to avoid introducing a new truncation error that can
arise from the right-shifts used in cxX and aY. To avoid the
truncation error, then-bit word is divided into two parts. One
part is q bits long and stores the radius. The other part is p bits
and stores the error. In the IEM, the maximum ER (of one
unit) occurs at 8= 1!'14 or when l/2N iterations have been
completed. Thus, the condition:
1
- N=2P
2
(34)
guarantees that the truncation error is less than one unit. The
relationship between the maximum permitted radius Rm and
the word length n can be determined through the equation:
n =q+p=log
2
Rm + log
2
( N) (35)
Transactions of the ASM E
Fig. 3 Errors in the generated circle
Substituting for N from equation (32) gives:
n = 2log
2
Rm + log
2
1r- 4
and since Rm = 2q
I
q=
2
n+ 1.2
(36)
(37)
Table 1 gives the maximum allowable radius in BLU's for
typical length of computer words (one bit is reserved as a sign
bit):
In practice only the last line in Table 1 would give suf-
ficiently large radii in CNC systems. Therefore, a double-or
even triple-precision word technique will be required in the
control computer. Note that since in the Euler method the
number of iterations are greater than in the IEM and the
maximum error appears at the end of the arc, the equivalent
formula to (37) is:
I
q =
2
n-0.15 (38)
which means that the maximum permitted radius in practice is
even smaller, and the Euler method is useless for CNC ap-
plications.
Taylor Method
For the Taylor method [4] the coefficients A and B in
equation (3) are evaluated by the truncated series:
1
A=l- -a
2
B=a
2 '
(39)
Using these values, the basic equations of the algorithm are
derived. The corresponding segment lengths are obtained
from equation (7):
I
DX(i) = - - a
2
X(i) - aY(i) (40)
2
1
DY(i) = - - a
2
Y(i) + aX(i)
2
and the velocities obtained from equation (9):
Vx (i)=-(Vo!R)[ aX(i)+Y(i)J (41)
Vy (i) =(V
0
/R)[ - aY(i) +X(i)]
The maximum value of the error ER is calculated! according to
equation (14):
ERmax =
2
: (J1 + ! a
4
-1)R=1rRa
3
! !6 (42)
Journal of Engineering for Industry
Table 2 Maximum radius in Taylor
method
n[bits]
15
23
31
q
10
15
21
p
5
8
10
1023
32767
2097151
The error EH is equation (11) is obtained by substitution for
A, which yields:
EH(I) =R -R (i)j1- ! a
2
=R - R (i) + (a2 /8)R (i) (43)
Since the error caused by the truncated series (39) leads to an
increased radius with each step, the maximum value of EH is
obtained at the first iteration, for R ( i) = R, which yields:
(44)
A comparison of two errors in equations (42) and (44) shows
that
'TrCi
ERmax = 2 EHmax (45)
which means that for an angle a which satisfies a<2/7r, the
error EH is the dominant one. In this case the angle a is
determined from equation (44). For EHmax equal to one BLU
a = v8/R (46)
In this case a< 2!1r is satisfied for R > 20 BLUs, which covers
virtually all practical systems (e.g., if lBLU = 10J.Lm,R >0.2
mm). The number of iterations to complete a quarter of a
circle is:
(47)
The algorithm given by equations (40) and (41) requires a
determination of a and a
2
Expressing these values in power
of 2 avoids the use of floating-point unit [5]:
(48)
or
(49)
Again, the power m is determined by SHIFT operations, and
all multiplication instructions in the algorithm are executed in
the same way. Therefore, the Taylor method can be im-
plemented by using basic assembler instructions.
The relationship between the maXimum radius and the
word-length for the Taylor method is:
and since R = 2q:
2
q=
3
n+0.56 (51)
Examples illustrating the dependence of the maximum radius
on the word length are given in Table 2. The last two cases in
Table 2 are practical for CNC systems.
In reference [4] the Taylor method algorithm was applied to
a CNC system controlled by a PDP-8 computer using a triple-
precision word. Successive points along the circular arc were
computed at the rate of 1.4 millisecond per point. The in-
terpolator occupied 600 (12 bit) words of core storage.
Tustin Method
The Tustin method [6] is based on an approximate
NOVEMBER 1982, Vol. 104/403
y

Fig. 4 Circle generation by the Improved Tustin Method
relationship between the derivative operators and the discrete
variablez:
(52)
Using the Tustin approximation yields the following relations
for cosa and sin a, or A and B:
1- (a/2)
2
a
A= I + (a/2)2 ; B = 1 +(a/2)2 (53)
These approximations can be used to generate circular in-
terpolation for CNC systems. Substituting these values of A
and B into equations (7) and (9) gives the basic equations of
the proposed algorithm:
DX(i) = -
1
2
[ a
2
X(i) +a Y(i)] (54)
l+(a/2) 2 .
1 [ 2 ']
DY(i) =
2
- Y(i) +o:X(i)
l + (a/2) 2 .
V (
') V
0
[ a X(') Y(. ]
xI =- R[l+(a/2)2] T I+ 1)
(55)
V ( ') V
0
[ a Y( ') X(')]
Y I = R[l + (a/2)2] - T I + I
The truncation error is found by substituting equation (53)
into (12) which shows that:
ER (i) = 0 (56)
Since the error ER is identically zero, the angle 01 is prescribed
by the error EH given in equation (16). For R (i) =R:
R
EH = R - -== .,..---
.../l+(a/2)2
which can be approximated for small a by:
al
EH= --R
a
2
+8
(57)
(58)
The angle a is derived by equating the error EH toO one unit:
a= .../8/(R - 1) =.../8/ R (59)
which is identical to equation (46).
Improved Tustin Method
The Tustin method yields an ER of zero. From a practical
point of view in manufacturing systems we choose the
maximum error equal to one BLU. Therefore the method can
be improved by increasing the angle a [7). From Fig. 4 the
angle a is evaluated using equation (15):
R+l -
-R-- =.../2/(1 +A)=...Jl + (a/2)2 = 1 +- (61)
- 1 8
404/ Vol. '104, NOVEMBER 1982
Table3 Simulation results for R = 10,000 BLU's
Method N
Simulation Practical
ERmax EHmax ERmax EHmax
Euler 12331 1.008 1.008 1.500 1.500
IEM 3928 1.002 1.002 1.652 1.652
Taylor 56 0.071 0.988 0.633 1.428
Tustin 56 1.016 1.016 0.681 1.681
ITM 40 1.029 0.999 1.662 1.405
Table4 Simulation results for R = 250 BLUs
Method Type N ERmax EHmax
!EM FPU 99 1.557 1.549
IEM ASL 102 1.103 1.103
Taylor FPU 9 0.529 0.999
Taylor ASL 13 0.953 0.953
TableS Features of reference-word interpolators
Method Programming a N
Euler FPU 4hrR .,? R/ 8
IEM FPU or ASL 4/R 1fRI8
Taylor FPU orASL .Jf1R
4
Tustin FPU ../SiR
4
ITM FPU 4/VR
.!!....{R
8
so that the angle a is:
a=.../16/(R - 1)=4/VR (62)
Therefore, the angle a is increased by a factor of Vi
decreasing the number of iterations to:
N= .!!.... VR
8
(63)
Obviously the Tustin methods require considerably more
calculations than the previous methods. Basic assembler
instructions are insufficient to handle the routine and an FPU
is a necessity.
Comparison and Discussion
The interpolation methods requmng an FPU were
simulated for R = 10000 BLU units to produce a quarter of a
circle. The results are presented in Table 3, where N is the
number of iterations and the errors are given in BLUs.
The errors ERmax and EHmax calculated during the
simulation of the methods, are listed in the third and fourth
columns of Table 3. In a real system, however, the DACs and
the position counters can only use integer numbers, so the
reference word must be an integer. This means that DX, DY,
Vx, and Vy must be rounded off to the closest integer value
which increases the error. The maximum errors ERmax and
EHmax for a real system are listed in the fifth and sixth
columns of Table 3. In this example, for real systems, the
interpolation methods generate errors in the range of I to 2
BLUs. In order to limit the error to 1 BLU the angle a should
be decreased. It can be seen that the Improved Tustin Method
will generate an arc with the fewest number of iterations.
In order to compare assembly language (ASL) with the
floating-point unit (FPU) methods, a simulation was per-
formed for producing a 90 degree arc with a radius of 250
units. The results shown in Table 4 indicate that for an
Transactions of the ASME
algorithm programmed with assembly instructions, the
Taylor method is the best. The number of iterations required
to complete! the arc by this method is considerably smaller
than the number needed with IEM, while the are always
within the required tolerance of one BLU.
Finally, the main features of the various algorithms have
been summarized in Table 5.
The interpolators of Table 5 which require an FPU will
require more computer storage and have computer
speed than those implemented with ASL. From Table 5 it can
be seen that if an FPU is available, the Improved Tustin
Method (ITM), which has been introduced in the present
paper, is the favorite. This method produces a circle using the
fewest iterations and guarantees the required accuracy. The
computation of the parameters A and Bin equatiion (53) does
not cause any difficulties since an FPU is used.
If an FPU is not available, or if implementation of
assembly language is preferred, then the Taylor method [4, 5]
should be applied. With this method the maximum allowable
radius is much larger than with the IEM (see Tables 1 and 2),
and the number of iterations required to complete the arc is
smaller by a factor of .J R/2.
Acknowledgment
The authors would like to thank Professors S. Malkin and
G. Ulsoy for their helpful comments.
References
I Koren, Y. , "Design of Computer Control for Manufacturing Systems,"
ASME JOURNAL OF ENGINEERING FOR INDUSTRY, Vol. 101, Ne>. 3, Aug. 1979,
pp. 326-332.
2 Koren, Y ., and Masory, 0., "Reference-Pulse Circular ltnterpolators for
CNC Systems," ASME JOURNAL OF ENGINEERING fOR INDUSTRY, Vol. 103, No.
I, Feb. 1981, pp. 131-136.
3 Koren, Y. , "Interpolator for a CNC System," IEEE Trans. on Com-
puters, Vol. 25, No. I, Jan. 1976, pp. 32-37.
4 Bergren, C., "A Simple Algorithm for Circular Interpolation," Control
Engineering, Sept. 1971, pp. 57-59.
5 Musse, J. , Veron, M., Lepage, F., and Drapier, J., "High Performance
Tool Path Interpolator for CNC System," CIRP Manufacturintg Systems, Vol.
7, No. I , 1978. (Presented in the 9th CIRP Int. Sem. on Manufacturing
Systems.)
6 Cadzow, J . A., and Martens, H. R., Discrete-Time and Computer Control
Systems, Chapter9, Prentice-Hall, New Jersey, 1970.
7 Koren, Y. , Computer Control of Manufacturing Systems, Chapter 5,
McGraw-Hill, New Y6rk, 1983.
APPENDIX 1
Summing the squares of both sides of equation t(6) gives:
Journal of Engineering for Industry
(Al-l)
where
(Al-2)
By successively using equation (Al-I), as the interpolation
proceeds, assuming that the initial point lies on the circle
[R(O) = R], one obtains:
R(i) =C;R (Al-3)
The radial error ER in the ith step is determined from
equation (10) and equation (A1-3):
ER(i) = (C; -l)R (A1-4)
Since the value of C is close to 1, equation (Al-4) is ap-
proximated by:
ER(i) =i(C-l)R (Al-5)
APPENDIX 2
Defining:
M(i) =R(i+ 1) - R(i)
= .JX'-(i +I)+ f2(i + 1) -.JX'- (i) + f2 (i) (A2-1)
The radial error ER at the jth step is derived by summing
the incremental errors M(1), or by integrating along the arcs:
ERU) = t M(i) = iMds
i=O
(A2-2)
The maximum error ERmax is obtained by equating the first
derivative of equation (A2-2) to zero:
dER
- =M(i) =0 (A2-3)
ds
Namely, at R (i + 1) =R (i)
Using the definition of equation (A2-1) we see that the
maximum radial error occurs when
Xl(i + 1) + f2(i +I) =X'- (i) + f2 (i)
Substitution of equation (25) into (A2-4) yields
2[X(i) -X(i+ l)]Y(i)/a=X'-(i+ 1)+ f2 (i)
(A2-4)
(A2-5)
Substituting for X(i + 1) from equation (24) into the left-side
term of equation (A2-5) gives Y(i) =X(i + 1), or
X(i) = Y(i) +aY(i) (A2-6)
NOVEMBER 1982, Vol.104/405