168. Memorandum from Col. Smith to Gen. Taylor, February 181

[Facsimile Page 1]

SUBJECT

  • Probabilities and a Nuclear Test Ban

1. During the discussions of the last week on a nuclear test ban treaty, the statistical experts at ACDA began playing down the importance of developing a thorough detection system by pointing out what was to them an acceptable probability of discovering a violation by random selection of seismic events. Once the probability train had been let out of the station, however, everyone got on board. The result is that it now seems, from a statistical viewpoint, that our proposed overall detection and inspection system is unsound. Before explaining some implications of this conclusion, let me explain briefly how the conclusion was reached.

2. In the discussion Friday with the technicians, Secretary McNamara said that there are four probabilities that must be considered before we can determine our ability to detect tests. They were defined as:

a. P1 is the probability of detection of the event.

b. P2 is the probability that the test will be selected from other events.

c. P3 is the probability that the area selected for the on-site inspection includes the test site.

d. P4 is the probability that the nuclear test will be detected if it falls within the inspected area.

3. Subsequent to the meeting, Mr. Barber of ISA drafted a paper presenting his views on the test ban. He began by setting out the four probabilities and promising to look at them in turn. He did not carry through on his planned approach, primarily because he said P4 depended not only on inspection procedures but also on Russian willingness to help us in our inspections. His overall conclusions were that 6–8 inspections annually were satisfactory, because “we want a few good inspections, not a lot of clumsy ones.”

[Facsimile Page 2]

4. At TAB A, I have used the data in Mr. Barber’s paper to show that for 5 inspections annually under fairly conservative assumptions we would have 1 in 100 chances of discovering a violation, and that for 6 inspections under fairly liberal assumptions, we would have only [Typeset Page 455] 3 in 100. It seems to be that neither case would be very convincing to the US public. Either our data or our inspection criteria must be changed.

5. Implications

The low probabilities of actually proving a violation have in part been foreseen by some. Thus we hear the technical argument that even if the Russians cheated by testing underground, they could not alter the military balance. If this assertion is true (and I am not certain by any means that it is), then it vitiates the need for any inspection system and, in turn, substantiates the Soviet argument that inspections are needed more for political than technical reasons.

6. Our main objective in pressing for an inspection system has been that it would establish the “principle of inspection”. If with this objective we set up a poor system, we will have set a poor precedent. It would be preferable to consider an unpoliced moratorium with a high readiness to test than a poorly inspected test ban and a high level of complacency.

W.Y.S.
  1. “Probabilities and a Nuclear Test Ban.” Secret. 2 pp. National Defense University, Taylor Papers, WYS Chron, January–March, 1963.