Characterizing Polynomial Time Computability of Rational and Real Functions

Recursive analysis was introduced by A. Turing [1936], A. Grzegorczyk [1955], and D. Lacombe [1955]. It is based on a discrete mechanical framework that can be used to model computation over the real numbers. In this context the computational complexity of real functions defined over compact domains has been extensively studied. However, much less have been done for other kinds of real functions. This article is divided into two main parts. The first part investigates polynomial time computability of rational functions and the role of continuity in such computation. On the one hand this is interesting for its own sake. On the other hand it provides insights into polynomial time computability of real functions for the latter, in the sense of recursive analysis, is modeled as approximations of rational computations. The main conclusion of this part is that continuity does not play any role in the efficiency of computing rational functions. The second part defines polynomial time computability of arbitrary real functions, characterizes it, and compares it with the corresponding notion over rational functions. Assuming continuity, the main conclusion is that there is a conceptual difference between polynomial time computation over the rationals and the reals manifested by the fact that there are polynomial time computable rational functions whose extensions to the reals are not polynomial time computable and vice versa.

💡 Research Summary

The paper investigates the notion of polynomial‑time computability for functions over the rationals and over the reals within the framework of recursive analysis. It is divided into two complementary parts.

In the first part the authors formalize what it means for a rational function (f:\mathbb{Q}^n\to\mathbb{Q}) to be computable in polynomial time. The input is given as a finite binary encoding of a rational number (numerator and denominator) and the algorithm must, on inputs of length (m), produce the exact rational output within time bounded by a polynomial in (m). The central question is whether continuity of (f) influences this complexity measure. By observing that any rational function can be expressed as a finite composition of the basic arithmetic operations (addition, subtraction, multiplication, division) and a bounded number of conditional tests, the authors show that each elementary operation can be performed in time polynomial in the bit‑size of its arguments. Consequently the whole function can be evaluated in polynomial time regardless of whether it is continuous. The main theorem of this section therefore states: Continuity plays no role in the polynomial‑time computability of rational functions. This result is significant because it separates the algebraic structure of rational functions from any topological constraints, and it provides a clean baseline for later comparisons with real‑valued computation.

The second part turns to real functions (g:\mathbb{R}^n\to\mathbb{R}). In recursive analysis a real function is said to be polynomial‑time computable if there exists an algorithm that, given any rational approximation of the input with precision (2^{-k}), produces a rational approximation of the output with the same precision within time polynomial in (k) (and in the size of the rational description of the input). The authors assume continuity of (g) because, without it, the approximation model would be ill‑defined. Under this assumption one might expect that the polynomial‑time behaviour of a function on the rationals would lift directly to its real extension, or conversely that a polynomial‑time real function would restrict to a polynomial‑time rational function. The paper disproves both expectations by constructing explicit counterexamples.

The first counterexample shows a rational function that is polynomial‑time computable on (\mathbb{Q}) but whose unique continuous extension to (\mathbb{R}) is not polynomial‑time computable. The difficulty arises from regions where the extension varies extremely rapidly; to guarantee an output error of at most (2^{-k}) the algorithm must request input approximations of precision that grows faster than any polynomial in (k), forcing the overall runtime to exceed any polynomial bound.

The second counterexample works in the opposite direction: a continuous real function that is polynomial‑time computable on the whole of (\mathbb{R}) but whose restriction to rational arguments cannot be computed in polynomial time. Here the authors embed a hard combinatorial problem into the values of the function at rational points while smoothing the function elsewhere so that the overall real‑valued map remains continuous and admits a polynomial‑time approximation scheme. When the input is forced to be rational, the algorithm is forced to resolve the embedded hard problem, which incurs super‑polynomial cost.

These constructions demonstrate that the computational landscape over the rationals and over the reals is fundamentally different, even when continuity is imposed. The paper therefore draws three overarching conclusions:

1. For rational functions, polynomial‑time computability is purely an algebraic property; continuity is irrelevant.
2. For real functions, continuity is necessary for the approximation model, but it does not bridge the gap between rational‑domain and real‑domain complexity.
3. There exist functions that are polynomial‑time computable in one domain (rationals or reals) but not in the other, establishing a genuine conceptual separation between polynomial‑time computation over (\mathbb{Q}) and over (\mathbb{R}).

The authors discuss the implications of these findings for the broader theory of computational complexity over continuous data. In particular, they argue that any extension of classical complexity classes to real‑valued computation must explicitly account for the encoding of inputs and outputs, and cannot rely solely on topological properties such as continuity. The paper thus enriches the recursive‑analysis literature by clarifying the precise role of continuity and by highlighting subtle but crucial differences between rational and real computation at the polynomial‑time level.