Let's assume the you have two algorithms for computing some single but complicated number (e.g., the Ramsey number $R(5,5)$). Both are provided as high-level, semi-formal textbook descriptions.

  1. The first algorithm, by means of some textual reformatting and filling in the usual gaps in high-level reasoning, can be easily (say, within several days of coding) turned into a (relatively large) computer program in some imperative programming language, including that of Turing machines (provided appropriate tooling support, of course).

  2. The second algorithm can almost be turned into code in some imperative programming language, except there are some unknown constants (e.g., the algorithm contains an assignment "$x := c^2$", where $c$ is nonconstructively defined by Theorem 5.19.38, which says $\exists\,c\in \mathbb{N}\colon\dots$). In the best case, this algorithm has some merits such as creating a connection to some other area of mathematics, and in the worst case, this algorithm is simply "'print $r$;', where $r$ is the constant $R(5,5)$". We know that the second algorithm solves our task, i.e., that a Turing machine corresponding to our high-level description exists, but we don't know some part of this algorithm exactly.

In mathematical writing, how do you call these algorithms when you provide the reader with them? I mean, how do you distinguish between them? Do you say explicit/nonexplicit? Effective/noneffective? Any other pair of terms?


1.   This would be called a constructive proof. Quoting from the linked wikipedia page:

In mathematics, a constructive proof is a method of proof that demonstrates the existence of a mathematical object by creating or providing a method for creating the object. This is in contrast to a non-constructive proof (also known as an existence proof or pure existence theorem) which proves the existence of a particular kind of object without providing an example.

2.   This would be called an existential (or existence) proof:

An existence theorem is purely theoretical if the proof given of it does not also indicate a construction of the kind of object whose existence is asserted. Such a proof is non-constructive, and the point is that the whole approach may not lend itself to construction. In terms of algorithms, purely theoretical existence theorems bypass all algorithms for finding what is asserted to exist. They contrast with "constructive" existence theorems.

Most algorithms are constructive, but not all. From Non-constructive algorithm existence proofs, which also gives a few examples of the latter:

The vast majority of positive results about computational problems are constructive proofs
[...] However, there are several non-constructive results, where an algorithm is proved to exist without showing the algorithm itself.

  • $\begingroup$ @user49915: What you have is a nonconstructive proof of the existence of an algorithm that prints $R(5,5)$. (But yes, such proofs are indeed trivial for algorithms with no input and a finite amount of output). $\endgroup$ – hmakholm left over Monica Jun 27 '18 at 23:07
  • $\begingroup$ @user49915 $R(5,5)$ is a well defined constant. I took your question to be about algorithms for computing $R(5,5)$, and the second part specifically refers to the algorithm contains an assignment "x:=c^2", where c is nonconstructively defined . Some of the answers to Can a problem be simultaneously polynomial time and undecidable? on MO might also be relevant in the context. $\endgroup$ – dxiv Jun 27 '18 at 23:09
  • $\begingroup$ @user49915 You don't have an algorithm unless it's proved to provide a solution. My reading of what the question calls nonconstructively is that (parts of) the algorithm are left out, relying on some (external) proof that they could/might theoretically be filled-in. $\endgroup$ – dxiv Jun 28 '18 at 1:18
  • $\begingroup$ @user49915 For a simpler example, consider an "algorithm" for finding a rational approximation to $\sqrt{2}$ with arbitrary precision $\epsilon$. The constructive way could be "calculate the convergents of the continued fraction of $\sqrt{2}$ and stop once the difference between consecutive terms is $\lt \epsilon$". The existential way, on the other hand, could be "let $p/q$ be any Dirichlet approximation with $q \gt 1/ \sqrt{\epsilon}$". Both are mathematically correct, but the latter relies on a step that's nonconstructive. $\endgroup$ – dxiv Jun 28 '18 at 1:19

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy