each leading aircraft of a group must be time-advanced. This is illustrated for four traffic samples in
figures 5(a) to 5(d). The traffic samples are chosen for light traffic (25 aircraft/hr) and heavy traffic
(40 aircraft/hr), one sample each with relatively low average delay and the other sample with excep-
tionally large delay. Figure 5 shows what causes relatively small and large delays. We have small
average delays when the ETAs are uniformly spread over the time-interval considered and are without
large gaps, and we have large average delays when the opposite is true. We can see that for low-density
traffic or well-spread traffic, TA should not be used, since delay is small already, and the cost in time-
advance for the modest delay reduction is high, 12.16 min in figure 5(a) and 14.49 min in figure 5(c).
There are many short groups, and many aircraft would have to fly faster than their preferred speed
profiles. On the other hand, the cost in time-advance for heavy or bunched traffic is relatively small,
2.55 min in figure 5(b) and 2.22 min in figure 5(d), since only three aircraft needed to be speeded up in
both cases. Figures 5(b) and 5(c) also show the modest improvement that can be achieved when CPS is
added to TA optimization. For figures 5(a) and 5(d), CPS found no position shift that gave reduced
delays. It is difficult to determine a break-even point for TA versus no TA, since both time and fuel are
involved either as savings or as cost for all aircraft whose schedules are affected.
Figures 6(a) and 6(b) show a traffic sample in which CPS is applied with and without permitting
overtakes. In this example, two additional heavy aircraft could be grouped together with overtakes
permitted, resulting in a reduction of the average delay per aircraft from 2.87 rain to 2.72 rain.
It was shown in the Scheduling Algorithm section that a 20-knot headwind upon landing increases
the required time-separations. A traffic sample illustrates this in figure 7, for FCFS only, for both both
no wind and for a 20-knot headwind. In this example, for an identical sequence of ETAs, the average
delay for FCFS scheduling is increased from 2.31 to 4.05 rain. Therefore, winds can play a major role in
causing delays.
Figure 8 shows parts of the traffic-sample diagrams having to do with CPS only. CPS tries to reduce
the length of a group of aircraft, which reduces the average delay of all aircraft. The cost of such delay
reduction is the fuel cost for those aircraft that have to be time-advanced beyond their ETA. Therefore,
CPS shows the most benefit in reduced average delay when the position switching is done early in a
large group, thus reducing the time delay for all following aircraft in that group. Switching at the end of
a group is of little benefit in reducing the average time delay (top example of fig. 8), but controllers pre-
fer to place a heavy aircraft at the end of a group. The remainder of figure 8 shows how CPS groups the
heavy aircraft together by either time-advancing or by delaying the heavy aircraft. In this manner,
groups of two, three, or four heavies are formed. Figure 8 shows only the reduction in delay: the group
becomes shorter. The cost of such position switching depends on data not shown here; namely, whether
the aircraft that are switched toward an earlier arrival time simply have their delay reduced, or if they
have to speed up to arrive earlier than their desired time of arrival.
The optimal CPS algorithm assumes that a tightly scheduled group of aircraft can be reordered such
that it is again tightly scheduled with no more than the minimum required gaps. After calculating the
new order of aircraft, we find that sometimes the delay of some aircraft in the group decreases so that
they have a negative delay of less than 1 min. In such cases, two alternative choices were made to meet
the restriction (see the captions of figs. 9(a) and 9(b)). Either choice satisfies the restrictions at only a
small loss of optimality when many samples are considered. Comparing the total delays for all aircraft in
the sample of figures 9(a) and 9(b) with those of 9(d) and 9(e), we see that there is no clear choice of
11