Most general order-preserving function

  • Thread starter Thread starter Dragonfall
  • Start date Start date
  • Tags Tags
    Function General
AI Thread Summary
The discussion centers on the characteristics of order-preserving functions, particularly those that are strictly increasing and efficiently computable. Participants clarify that strictly increasing functions satisfy the condition f(a) < f(b) for a < b, but there is confusion regarding the uniform distribution of outputs. The need for a decomposable function structure is emphasized, where a function can be expressed as a composition of two parts, making it difficult to retrieve input values from limited information. The conversation hints at developing a new protocol related to secure computation, specifically for Yao's Millionaires' Problem, leveraging these function properties. Overall, the thread explores the complexities of defining and utilizing order-preserving functions in computational contexts.
Dragonfall
Messages
1,023
Reaction score
5
Is there an easy description for all the real to real functions that are order-preserving? I can only think of linear functions
 
Mathematics news on Phys.org
Dragonfall said:
Is there an easy description for all the real to real functions that are order-preserving? I can only think of linear functions
What do you mean by "order-preserving?" Is it the idea that if a < b, then f(a) < f(b)? If so any function that is strictly increasing satisfies the latter inequality. If you mean something different, you need to clarify your question.
 
What I need is a class of functions that is strictly increasing, very efficient to compute, and when one is selected uniformly and applied to x < y, the pair (f(x), f(y)) is uniformly distributed.
 
For every x,y? Then your class of functions has to be uncountable. What do you mean with "selected uniformly"? There is no clear definition of this on an arbitrary uncountable set.
Also, uniform distributions over the full real line do not exist.

f(x)=mx+c with suitable distributions of m and c can give a uniform distribution of (f(x),f(y)) in some finite range, or give nice non-uniform distributions as well.
 
OK the domain is integers [1, ..., n], forget about what the range is for now, we can decide that after. I have a pair of numbers (a, b). I want to apply a function at random from this class so that (f(a), f(b)) is uniformly random and f(a) < f(b).

f has to be efficiently computable. It also has to be "decomposable" into two parts f_1, f_2 such that f(x) = f_2 ( f_1 (x) ). Knowing ONLY ONE of f_1 or f_2, and given f(x), it should be infeasible to compute x.

I'm thinking Ax + B for A > 0 and random B. Knowing Ax+B, and only one of A or B doesn't tell you anything about x.
 
Uniformly random where?

Every f is "decomposable" that way, simply set f_1(x)=f(x), f_2(x)=x (or vice versa).
Knowing only one of an arbitrary and unknown "decomposition" makes the task impossible - see my example, one of the functions could be just the identify function.

I think it would really help if you give more context. What do you actually want to do?
 
mfb said:
Uniformly random where?

Every f is "decomposable" that way, simply set f_1(x)=f(x), f_2(x)=x (or vice versa).
Knowing only one of an arbitrary and unknown "decomposition" makes the task impossible - see my example, one of the functions could be just the identify function.

I think it would really help if you give more context. What do you actually want to do?

No. If f_1 = f then knowing f_1 and knowing f(x) gives you x, assuming it's not one-way and is one to one.

You know that zero-knowledge proof of graph isomorphism? You have two permutations. Composition gives you the permutation. Knowing just one gives you nothing. I'm trying to come up with something similar for two ordered committed numbers.
 
Dragonfall said:
I'm trying to come up with something similar for two ordered committed numbers.
So who is supposed to know what? What is the protocol you want to use, and what is the result?
 
You're not supposed to know that. I'm supposed to come up with it. And I have! Thanks for the inputs.
 
Back
Top