Can a program be correct and still not reliable?

Answer 1

It depends on what you mean by the words "correct" and "reliable". A typical notion of correctness means that the program, when executed correctly, does what you want it to do. Similarly, you might want reliability to mean that every run of the program does its job.

This being accepted, it is possible for a "correct" program to prove unreliable, as when you have gamma rays hitting the memory chips and corrupting data. Such cases, while extremely rare (modern digital circuits have extraordinary ECC mechanisms), do occur in practice. An example was when a single bit error in one of Amazon's servers led to a domino crash, and within a few minutes, Amazon's S3 services were down.

Of course, the Amazon example is probably controversial, because fault tolerance is a correctness requirement for their engineers.

It is also for this reason that you'll find building ICs for rockets and satellites is a much harder job than building ICs for everyday use --- the ability to withstand high radiation without corruption being the extra requirement.

At the other end of the spectrum, you have "incorrect" programs which have nevertheless proved extremely reliable. I'm referring to the whole field of randomized algorithms. While these algorithms, strictly speaking are incorrect, they return the correct answer with high probability. Run sufficiently many times, the probability of these algorithms making is a mistake is less than the probability of your being struck by lightning. Unlike correct unreliable programs, these find use literally everywhere in the computer industry.

Answer 2

Yes, a program can definitely be correct but still not reliable. The individual components of the program can be coded properly and be technically correct, but it is quite possible that the program will have bugs and will not work reliably.

Answer 3

Of course. The program was written to be unreliable (the correctness)