💬 NLP

Turing or Cantor: That is the Question

This paper argues that Alan Turing's foundational work relies on Georg Cantor's set theory, while proposing new measures of undecidability, extending super-Turing computation models, defining three new complexity classes for undecidable problems (U-complete, D-complete, and H-complete), and presenting a negative resolution to the P vs. NP equivalent for the U-complete class.

Original authors: Eugene Eberbach

Published 2026-04-14

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Eugene Eberbach

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: A Detective Story About "Unsolvable" Problems

Imagine you are a detective trying to solve every crime in the world. You have a giant, perfect rulebook (a computer program) that can solve any crime if you just give it enough time.

Alan Turing (the father of computer science) came along in the 1930s and said, "Wait a minute. Even with my perfect rulebook, there are some crimes I can never solve. No matter how long I try, the rulebook will just spin in circles forever." This is the famous Halting Problem: knowing if a program will eventually stop or run forever.

Georg Cantor (a mathematician from the 1800s) is the paper's "hidden hero." He discovered that there are different sizes of infinity. He proved that the number of Real Numbers (like 3.14159...) is a bigger infinity than the number of Whole Numbers (1, 2, 3...).

The Paper's Main Idea:
The author, Eugene Eberbach, argues that Turing couldn't have discovered these "unsolvable" problems without Cantor's math.

The Analogy: Think of Whole Numbers as the number of possible computer programs (algorithms). Think of Real Numbers as the number of all possible questions or problems.
Cantor proved there are more questions than there are programs to answer them.
Therefore, most questions are "unsolvable" by any computer.

The paper asks: "Turing or Cantor: Who deserves the credit?" The answer is: Both. Cantor built the map showing there are "unsolvable lands," and Turing built the vehicle (the computer) that proved it can't go there.

The New "Difficulty Levels" for Unsolvable Problems

Usually, when we say a problem is "unsolvable," we just stop there. But this paper says, "Not so fast! Let's grade how unsolvable they are."

The author creates three new "difficulty levels" (Complexity Classes) for problems that computers can't solve, inspired by how we grade hard math problems.

1. U-Complete (The "Semi-Solvable" Level)

The Metaphor: Imagine a treasure hunt where you can find the treasure if you look for it, but if the treasure isn't there, you might search forever and never know if it's missing or just hidden deep.
What it means: If the answer is "Yes," the computer can eventually find it and stop. If the answer is "No," the computer runs forever.
Example: The "Halting Problem" itself. If a program stops, we know it stops. If it doesn't, we wait forever.
Status: This is the "easiest" type of unsolvable problem.

2. D-Complete (The "Diagonalization" Level)

The Metaphor: Imagine a library where the librarian tries to write a catalog of every book in the library. But every time they write a list, they have to add a new book that isn't on the list. The list can never be finished or accurate.
What it means: These problems are so tricky that the computer can't even say "Yes" or "No" reliably. It can't even recognize the answer if it sees it.
Why "Diagonalization"? This comes from Cantor's famous trick of crossing out numbers on a diagonal to prove there are more numbers than you can count.
Status: Harder than U-Complete. The computer is completely lost here.

3. H-Complete (The "Hyper-Unsolvable" Level)

The Metaphor: Imagine a problem that requires an infinite amount of time to solve, even if you had a machine that could do infinite things at once. It's like trying to count to a number that is bigger than "infinity."
What it means: These problems are so complex that even if we upgrade our computers to "Super-Computers" (Hypercomputers) that can do infinite steps, they still can't solve them.
Status: The hardest level. The "Mount Everest" of unsolvable problems.

The "Measure of Unsolvable"

The paper suggests a new way to look at problems. Instead of just saying "This is impossible," let's ask: "What percentage of this problem is impossible?"

0% Unsolvable: A normal problem (like sorting a list of names). The computer solves it every time.
50% Unsolvable: A problem where half the time the computer can solve it, and half the time it gets stuck.
100% Unsolvable: A problem where the computer is stuck on every single attempt.

This helps us understand that "unsolvable" isn't just a black-and-white switch; it's a spectrum.

The Infinite Ladder

Finally, the paper proposes a mind-bending idea: There isn't just one level of "impossible."

Just as Cantor showed there are infinite sizes of infinity (Infinity 1, Infinity 2, Infinity 3...), the author suggests there is an infinite ladder of unsolvable problems.

There are problems harder than D-Complete.
There are problems harder than those.
And so on, forever.

The Conclusion: Who Wins?

The paper concludes with a playful verdict on the title question: "Turing or Cantor?"

Turing gave us the tools (the computer) to try and solve problems.
Cantor gave us the map (the math of infinity) that showed us why some problems can never be solved.

The Verdict: You can't have the discovery of the "unsolvable" without the map that showed it existed. So, while Turing is the "Father of Computer Science," Cantor is the "Grandfather" who laid the foundation that made Turing's work possible.

In short: Computers are amazing, but they are limited by the very nature of mathematics. And thanks to Cantor, we know exactly why they are limited.

1. Problem Statement

The paper addresses the foundational limits of computability theory, specifically focusing on the nature of undecidable problems that cannot be solved by Turing Machines (TMs). While Alan Turing is widely recognized as the father of computer science for proving the undecidability of Hilbert's Entscheidungsproblem (Decision Problem) and the Halting Problem, the author argues that Turing's work relies fundamentally on the earlier set-theoretic results of Georg Cantor.

The paper identifies several gaps in current computability theory:

Lack of Classification: Unlike decidable problems (which have complexity classes like P and NP), undecidable problems lack a formal hierarchy of complexity classes.
Binary View of Solvability: The field often treats undecidability as a binary state (solvable vs. unsolvable), ignoring the "degree" of unsolvability or the possibility of solving specific instances.
Underutilization of Cantor: The profound influence of Cantor's diagonalization and cardinality arguments on the structure of undecidable problems is not explicitly formalized in modern complexity theory.

2. Methodology

The author employs a combination of historical analysis, set-theoretic reduction, and the extension of computational models to propose new theoretical frameworks:

Cantor-Based Proofs: The paper utilizes Cantor's diagonal argument and the distinction between countable (natural numbers) and uncountable (real numbers) infinities to provide a shorter, alternative proof of the undecidability of the Entscheidungsproblem.
Measure of Undecidability: A novel metric is proposed based on the probability distribution of input data. Instead of a binary "undecidable" label, the paper suggests measuring the "degree of unsolvability" as the percentage of input instances that are undecidable versus those that are decidable.
Extension of Oracle Machines: The methodology extends Turing's original "Oracle Machines" (o-machines) into a broader class of Super-Turing (Hypercomputational) models. This includes models utilizing interaction, evolution, and infinite time/resources.
Complexity Classification: Inspired by the Cook-Levin theorem for NP-complete problems, the author defines new complexity classes specifically for undecidable problems, categorizing them by their "hardness" relative to different types of infinity (countable vs. uncountable).

3. Key Contributions

A. A Shorter Proof of the Entscheidungsproblem

The paper presents a concise proof that the Entscheidungsproblem is undecidable, relying directly on Cantor's theorem.

Logic: A Universal Turing Machine (UTM) represents a countable set of machines. However, the set of all possible languages (problems) corresponds to the uncountable set of real numbers (via Cantor).
Conclusion: Because the set of problems is uncountable and the set of TMs is countable, there exist problems (languages) that no TM can decide. This establishes the undecidability of the Entscheidungsproblem via cardinality reduction.

B. Measure of Undecidability

The author proposes a quantitative approach to undecidability:

Definition: The degree of unsolvability is defined by the ratio of undecidable input instances to total input instances, based on a probability distribution.
Implication: A problem can be 100% undecidable (all instances are unsolvable) or partially undecidable (some instances are solvable, others are not). This allows for the application of randomized techniques or approximation algorithms to specific instances of "undecidable" problems.

C. Three New Complexity Classes for Undecidable Problems

The paper introduces a hierarchy of complexity classes for problems that TMs cannot solve, analogous to the NP-complete class for intractable problems:

U-complete (Universal Complete):
- Definition: Problems that are semi-decidable (Recursively Enumerable but not Recursive).
- Characteristics: If a string is in the language, the TM halts and accepts in finite steps. If not, it may run forever.
- Examples: The Halting Problem ( $L_u$ ), Post Correspondence Problem (PCP), Busy Beaver Problem.
- Significance: These are the "easiest" undecidable problems; they are solvable if the solution is guessed (similar to NP).
D-complete (Diagonalization Complete):
- Definition: Problems that are not even semi-decidable (Non-Recursively Enumerable).
- Characteristics: No TM can accept strings in the language, nor can it definitively reject strings outside the language in finite time.
- Examples: The Diagonalization Language ( $L_d$ ), the complement of the Halting Problem.
- Significance: These require a higher level of logical power than U-complete problems.
H-complete (Hypercomputation Complete):
- Definition: Problems that are hyper-undecidable, meaning they cannot be decided even with infinite time steps or infinite resources (unless using hypercomputational models).
- Characteristics: These problems lie beyond the reach of any standard or limit-computable algorithm.
- Examples: Hyper-diagonalization languages.
- Significance: These represent the boundary of what is computable even by "Super-Turing" models.

D. The Infinite Hierarchy Hypothesis

The paper hypothesizes an infinite hierarchy of undecidable complexity classes. Drawing on Cantor's hierarchy of infinite cardinalities ( $\aleph_0, \aleph_1, \dots$ ) and Turing's hierarchy of Oracle machines, the author suggests that undecidability is not a single state but an infinite spectrum of increasing difficulty.

4. Results

Proof of Inequality: The paper formally proves that for U-complete problems, the class of recursive languages ($REC$) is not equal to the class of recursively enumerable but non-recursive languages ( $RE \setminus REC$ ).
Negative Answer to a P vs. NP Analog: The paper answers a question analogous to $P \stackrel{?}{=} NP$ for the undecidable realm. It concludes that for U-complete problems, the class of decidable languages is strictly smaller than the class of semi-decidable languages (i.e., $REC \neq RE$ ), effectively answering the question negatively.
Re-evaluation of Turing's Legacy: The results demonstrate that Turing's achievements were contingent on Cantor's set theory. The paper argues that Cantor should be recognized as an "implicit grandfather" of computer science.

5. Significance

Theoretical Expansion: By defining U-complete, D-complete, and H-complete classes, the paper fills a major gap in complexity theory, providing a structured way to analyze and compare undecidable problems.
Practical Implications: The "measure of undecidability" suggests that even for theoretically undecidable problems, practical solutions may exist for specific subsets of inputs. This shifts the focus from "can we solve this?" to "how much of this can we solve?"
Hypercomputation Framework: The extension of Oracle machines to a full class of Super-Turing models provides a theoretical basis for emerging fields like evolutionary computing, neural networks, and interaction machines, positioning them not as anomalies but as necessary extensions of the Turing model.
Historical Context: The paper re-contextualizes the history of computer science, emphasizing that the limits of computation are rooted in the mathematical limits of set theory established by Cantor, rather than solely in Turing's mechanical models.

In summary, Eberbach's paper argues that to fully understand the limits of computation, one must move beyond the binary view of Turing Machines and adopt a hierarchical, Cantor-inspired framework that quantifies the degrees of undecidability and classifies problems into a new spectrum of complexity.