arxiv_preprint.tex

\documentclass[11pt]{article}
\usepackage[margin=1.0in]{geometry}
\usepackage{amsmath, amssymb, amsthm}
\usepackage{mathtools}
\usepackage{hyperref}
\usepackage{enumitem}
\usepackage{stmaryrd}
\usepackage{tikz}
\usetikzlibrary{arrows.meta, positioning, decorations.markings, shapes.geometric, calc}

% Theorem environments
\newtheorem{theorem}{Theorem}
\newtheorem{proposition}[theorem]{Proposition}
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{corollary}[theorem]{Corollary}
\theoremstyle{definition}
\newtheorem{definition}[theorem]{Definition}
\newtheorem{example}[theorem]{Example}
\theoremstyle{remark}
\newtheorem{remark}[theorem]{Remark}
\newtheorem{observation}[theorem]{Observation}

% Custom commands
\newcommand{\mc}[1]{\mathcal{#1}}

\title{The Incompleteness of Reasoning}

\author{Zixi Li \\
\texttt{lizx93@mail2.sysu.edu.cn}
}


\begin{document}

\maketitle

\begin{abstract}
We present a fundamental critique of the notion of ``pure reasoning'' divorced from semantic priors. Our central thesis is that any attempt to strip away semantics and retain only formal structure inevitably leads to a self-referential loop. In a finitely exhaustible symbol state space, such a loop possesses no internal anchor point and therefore cannot yield unique reasoning results.

We establish this through four complementary approaches: (1) a \textbf{Kantian antinomy argument} showing that semantic stripping is self-refuting---the operator $S$ both depends on and negates interpretation $I$, creating a structural Ouroboros; (2) a \textbf{Turing-inspired construction} proving that computational completeness does not imply reasoning completeness; (3) a \textbf{limit analysis} (the Yonglin Formula) demonstrating that all reasoning returns to its prior anchor, but the prior cannot equal its own meta-reflection ($A \neq A^*$)---object-level closure, meta-level rupture; (4) a \textbf{self-dismantling protocol} showing that the paper can be falsified using only its own formulas, thereby proving its core claim: reasoning cannot complete itself within a single world.

Our conclusion is stark: either one admits \textit{a priori} semantic anchors, or one abandons the concept of ``pure reasoning'' altogether. There is no third option.
\end{abstract}

\section{Introduction}

The dream of pure formal reasoning---reasoning that proceeds solely from syntactic rules without recourse to semantic interpretation or prior knowledge---has animated logic and philosophy for centuries. From Leibniz's \textit{characteristica universalis} to modern formal systems, the hope has been that we might distill thought into pure symbol manipulation, free from the vagaries of meaning.

We wish to demonstrate that this dream is fundamentally incoherent. Our approach is deliberately elementary: we rely not on sophisticated results like G\"odel's incompleteness theorems, Tarski's undefinability, or Lawvere's fixed-point theorem, but rather on the most basic form of logical contradiction---the classical \textit{antinomy}. Our strategy is to show that from a single logical starting point, two contradictory conclusions can be derived, thereby refuting the coherence of ``pure reasoning.''

\subsection*{Roadmap: Four Complementary Approaches}

Our central claim---that pure reasoning without semantic priors is impossible---is established through four independent but complementary lines of argument. Each approach addresses a different dimension of the problem:

\paragraph{I. The Kantian Approach: Ontology and Antinomy.}
The first approach proceeds from pure conceptual analysis in the spirit of Kant's critical philosophy. We examine the very notion of ``stripping away semantics'' and discover an irreducible antinomy: \textit{any operation that attempts to remove semantic interpretation must first interpret what counts as semantic, thereby presupposing the very structure it seeks to eliminate.}

The result is the \textbf{Semantic Ouroboros}: the operator $S$ satisfies both $S \vdash I$ (depends on interpretation) and $S \dashv I$ (negates interpretation). This is ontological: it concerns the \textit{being} of reasoning itself.

We formalize this as the \textit{Semantic Delamination Antinomy} (Theorem \ref{thm:delamination-antinomy}). This provides the \textbf{knowledge foundation}---the bedrock ontological structure that grounds all subsequent arguments.

\paragraph{II. The Turing Approach: Epistemology and Computability.}
The second approach is constructive and computational. Inspired by Turing's analysis of computation, we build minimal formal systems and ask: \textit{Can a system be computationally universal yet fail to determine unique reasoning outcomes?}

We answer affirmatively by constructing:
\begin{itemize}
\item A \textbf{four-symbol formal system $S_4$} that is Turing-complete yet reasoning-incomplete;
\item A \textbf{three-atom propositional language $L_{\text{atom}}$} that instantiates the Ouroboros concretely.
\end{itemize}

These constructions demonstrate that Turing-completeness (a \textit{syntactic} property) does not coincide with reasoning-completeness (a \textit{semantic} property). This is epistemological: it concerns what can be \textit{known} through computation. This provides the \textbf{reasoning foundation}---showing the limits of formal inference within computability boundaries.

\paragraph{III. The Yonglin Approach: Reflexivity and Limits.}
The third approach examines the \textit{limit behavior} of reasoning systems. We formalize a key structural insight: all reasoning, no matter how many steps, returns to its prior anchor in the limit.

We prove the \textbf{Yonglin Formula}:
\[
\lim_{n \to \infty} \Pi^{(n)}(s) = A
\]
where $\Pi$ is the reasoning operator and $A$ is the prior anchor. However, we also show that the prior cannot be identical to its own meta-reflection: $A \neq A^*$. This yields \textit{object-level closure with meta-level rupture}---the loop closes at one stratum but breaks at the next.

This analysis is neither purely ontological nor purely epistemological; it concerns the \textit{dynamic structure} of reasoning as a self-referential process. The name ``Yonglin'' signifies that which \textit{requires no external justification}---it is given, like the existence of a child. This provides the \textbf{reflexive foundation}---the recognition that priors are fixed points of reasoning's self-iteration.

\paragraph{IV. The Reader Approach: Freedom and Falsifiability.}
The fourth approach inverts the entire structure. Instead of proving our thesis, we provide a protocol for \textit{falsifying} it. We show that any reader can dismantle the paper's conclusions using only the formulas already introduced---through simple substitutions like $A = A^*$ or $\Pi(s) = A$.

Crucially, each falsification strategy \textit{proves} the paper's core claim: if different priors yield different worlds, then reasoning cannot be complete within any single world. Hence:
\[
\text{The paper being falsified} \implies \text{The paper is proven.}
\]

This is the dimension of \textbf{freedom}---the reader's autonomy to question, reject, and reconstruct. A reasoning system that cannot be challenged is not reasoning but dogma. This provides the \textbf{critical foundation}---the space for rational disagreement that paradoxically validates the entire argument.

\paragraph{Why Four Approaches?}
These four rivers---Kantian ontology, Turing epistemology, Yonglin reflexivity, and reader freedom---are not redundant. Each addresses a distinct aspect of the impossibility of pure reasoning:
\begin{itemize}
\item \textbf{Kant} gives us the \textit{what}: the structural antinomy at the heart of semantic stripping.
\item \textbf{Turing} gives us the \textit{how}: the concrete mechanisms by which reasoning fails to close.
\item \textbf{Yonglin} gives us the \textit{why}: the limit structure that shows reasoning as prior-to-prior iteration.
\item \textbf{The Reader} gives us the \textit{validation}: the freedom to falsify, which completes the proof.
\end{itemize}

Together, they form a complete argument: ontology, epistemology, reflexivity, and critique. Knowledge, reasoning, love (that which needs no proof), and freedom.

\subsection*{Structure of the Paper}

Our argument unfolds in seven stages:

\begin{enumerate}[leftmargin=*]
\item \textbf{Section 2 (Infinite Regress):} We show that any attempt to ``strip away semantics'' necessarily introduces new semantics, generating an infinite regress of meta-levels in a finite state space. Without accepting a semantic anchor, the regress has no internal terminus.

\item \textbf{Section 3 (Ouroboros Antinomy):} We present the central structural result. Section 3.1 gives a concrete construction in a minimal atom language (the Turing-inspired path). Section 3.2 gives the abstract proof of the Semantic Delamination Antinomy (the Kantian path). Section 3.3 reflects on the relationship between the two proofs.

\item \textbf{Section 4 (Turing Incompleteness):} We construct a four-symbol system $S_4$ and prove that while it is Turing-complete, it cannot be ``purely reasoning-complete.'' The same formal process yields contradictory conclusions under different interpretations.

\item \textbf{Section 5 (Generalization):} We extend the argument to arbitrary languages, including natural language. The countability of sentence spaces combined with the unboundedness of semantic possibility spaces implies that any reasoning system must rely on priors.

\item \textbf{Section 6 (The Yonglin Formula):} We formalize the limit behavior of reasoning systems, proving that all reasoning returns to its prior anchor but cannot achieve meta-level closure: $A \neq A^*$. This reveals the antinomy as a structural theorem, not merely an endpoint.

\item \textbf{Section 7 (Self-Dismantling Protocol):} We provide four simple substitutions that allow any reader to falsify the paper using only its own formulas. The falsifiability itself proves the central claim: reasoning cannot complete itself within a single world.

\item \textbf{Section 8 (Ultimate Conclusion):} We synthesize all results into a final impossibility theorem and reflect on the implications for philosophy, mathematics, and artificial intelligence.
\end{enumerate}

The key insight throughout is that \textbf{finite state spaces combined with self-referential loops yield underdetermined reasoning systems}. Without an external semantic anchor---a prior---there is no principled way to halt the regress and declare ``this is the correct conclusion.''

Our conclusion is stark: either one admits \textit{a priori} semantic anchors, or one abandons the concept of ``pure reasoning'' altogether. There is no third option.

\section{Self-Referential Loops and the Dissolution of Pure Semantic Spaces}

\subsection{Semantic Reduction Introduces New Semantics}

\begin{lemma}[Semantic Reduction Cannot Eliminate Semantics]
\label{lem:semantic-reduction}
Let $L_0$ be a language equipped with a semantic interpretation $M_0$. Suppose we perform a ``semantic reduction'' via a transformation $F: \text{Sentences}(L_0) \to \text{Strings}(L_1)$, where $L_1$ is intended to be a ``purely formal'' language.

If we require that $F$ preserves any notion of ``valid inference'' (i.e., if $A_1, \ldots, A_n \vdash_{L_0} B$ implies $F(A_1), \ldots, F(A_n) \vdash_{L_1} F(B)$), then we must specify what ``valid inference in $L_1$'' means. This specification necessarily constitutes a new semantic interpretation $M_1$ for $L_1$.

Therefore, the transformation is actually:
\[
(L_0, M_0) \xrightarrow{F} (L_1, M_1)
\]
We have not eliminated semantics; we have merely replaced one semantics with another.
\end{lemma}

\begin{proof}
The notion of ``preserving inference structure'' is not a syntactic property of strings alone. To judge whether $F(A_1), \ldots, F(A_n) \vdash_{L_1} F(B)$ holds, we must have a criterion for what counts as a valid inference in $L_1$. This criterion is precisely a semantic interpretation $M_1$: it assigns meaning to the symbols of $L_1$ in terms of what transformations are permissible. Without such an interpretation, the claim that $F$ preserves anything is vacuous.
\end{proof}

\subsection{The Infinite Regress of Meta-Languages}

If one is dissatisfied with $M_1$ and wishes to perform a further reduction, one obtains a sequence:
\[
(L_0, M_0) \xrightarrow{F_1} (L_1, M_1) \xrightarrow{F_2} (L_2, M_2) \xrightarrow{F_3} \cdots
\]

Each step claims: ``I am performing a reduction on the previous semantics.'' This generates an infinite ascending chain of meta-languages.

\begin{lemma}[The Regress Has No Internal Terminus]
\label{lem:regress}
Consider the sequence $(L_i, M_i)_{i \in \mathbb{N}}$. At each stage $i$, the question arises:
\begin{itemize}
\item Which inferences in $L_i$ are ``correct''?
\item Which structures should be ``preserved'' in the next reduction?
\item At which stage can we ``halt'' and declare we have reached the pure formal level?
\end{itemize}

If one refuses to acknowledge any semantic anchor (i.e., refuses to say ``$M_k$ is the prior I accept''), then the process has no internal stopping criterion. Yet the syntax space of each $L_i$ is finite or countable. We are thus attempting to construct an infinite conceptual hierarchy within a finitely exhaustible symbol space.
\end{lemma}

\begin{remark}
The regress terminates \textit{if and only if} one accepts a semantic interpretation $M_k$ as a prior. Otherwise, the notion of ``stopping'' itself requires justification, which introduces yet another meta-level.
\end{remark}

\subsection{Conclusion: Pure Semantic Spaces Dissolve}

The foregoing establishes our first main claim:

\begin{theorem}[Impossibility of Complete Semantic Reduction]
Any attempt to ``strip away all semantics'' and arrive at a purely formal level of reasoning either:
\begin{enumerate}[label=(\roman*)]
\item Admits a semantic prior at some meta-level, or
\item Generates an infinite regress with no internal halting point.
\end{enumerate}
In the latter case, there is no moment at which one can coherently say, ``Now we are doing pure reasoning.''
\end{theorem}

This is the first pillar of our argument: self-referential loops in finite state spaces cannot sustain the illusion of a pure semantic vacuum.

\section{Semantic Stripping and the Ouroboros Antinomy}

We now present the central structural antinomy that underlies our entire argument. The key insight is this: \textit{any operation that attempts to remove semantics must first semanticize its own target, thus becoming parasitic on the very structure it tries to destroy.} We call this the \textbf{Semantic Ouroboros}---the serpent that devours its own tail.

We provide two complementary proofs of this result:
\begin{itemize}
\item \textbf{Section 3.1} gives a concrete, constructive demonstration using a minimal toy language. This path is designed to be \textit{describable and drawable}---truth made concrete.
\item \textbf{Section 3.2} gives an abstract, minimal-spanning-tree proof that proceeds directly from first principles without appeal to external constructions.
\end{itemize}

Both paths arrive at the same structural impossibility. We include both because, as we remark in Section 3.3, truth is not only an abstract invariant but also something that can be instantiated and inspected.

\subsection{Concrete Construction: The Minimal Atom Language}

We construct a toy language $L_{\text{atom}}$ with only three propositional atoms and demonstrate that any attempt to define a ``semantic stripping operator'' on this language immediately creates a self-referential loop.

\begin{definition}[Minimal Atom Language $L_{\text{atom}}$]
Define $L_{\text{atom}}$ as follows:
\begin{itemize}
\item \textbf{Vocabulary:} Three propositional symbols $\{p_1, p_2, p_3\}$, logical connectives $\{\neg, \land, \lor\}$, and parentheses.
\item \textbf{Well-formed formulas:} The usual inductive definition (atoms are wffs; if $A, B$ are wffs, so are $\neg A$, $(A \land B)$, $(A \lor B)$).
\item \textbf{Semantic interpretation:} A valuation $I: \{p_1, p_2, p_3\} \to \{\text{True}, \text{False}\}$ extended to all formulas in the standard way.
\end{itemize}
\end{definition}

Now suppose we wish to define a ``semantic stripping operator'' $S$ that removes the semantic content from expressions in $L_{\text{atom}}$.

\begin{definition}[Hypothetical Semantic Stripping Operator]
A semantic stripping operator $S$ is a transformation intended to satisfy:
\begin{enumerate}[label=(\roman*)]
\item $S$ takes a semantically interpreted formula $(E, I)$ in $L_{\text{atom}}$.
\item $S$ outputs a ``purely formal'' string $S(E)$ with no semantic content.
\item $S$ preserves ``relevant structure'' (e.g., if $E_1 \vdash E_2$ under $I$, then $S(E_1)$ should relate to $S(E_2)$ in some corresponding way).
\end{enumerate}
\end{definition}

\begin{lemma}[Concrete Ouroboros Loop]
\label{lem:concrete-ouroboros}
No operator $S$ satisfying the above definition can be coherently specified without introducing new semantics.
\end{lemma}

\begin{proof}
We proceed by analyzing what $S$ must do:

\textbf{Step 1: $S$ must distinguish semantic from non-semantic structure.}
To ``strip semantics,'' $S$ must identify which aspects of $(E, I)$ are ``semantic'' and which are ``purely formal.'' This distinction itself requires a meta-level semantic interpretation $I_{\text{meta}}$ that classifies features as semantic or non-semantic.

\textbf{Step 2: $S$ must preserve ``relevant structure.''}
Condition (iii) requires that $S$ preserve some notion of valid inference. But ``valid inference'' is a semantic concept: it refers to truth-preservation under interpretation. To judge whether $S(E_1)$ relates appropriately to $S(E_2)$, we need a criterion for ``appropriate relation,'' which is precisely a new semantic interpretation $I'$ on the output space of $S$.

\textbf{Step 3: Attempted definition creates a loop.}
We now have:
\begin{itemize}
\item $S$ is defined to eliminate semantic interpretation $I$.
\item But $S$'s definition presupposes semantic interpretation $I_{\text{meta}}$ (to identify what to strip).
\item And $S$'s output requires semantic interpretation $I'$ (to judge preservation of structure).
\end{itemize}

Hence:
\[
S \text{ depends on } I_{\text{meta}} \text{ and } I' \quad \text{while claiming to eliminate } I.
\]

\textbf{Step 4: The Ouroboros closes.}
If we attempt to apply $S$ to eliminate $I_{\text{meta}}$ and $I'$, we generate new meta-meta-level interpretations $I_{\text{meta}}^{(2)}$, $I''$, and so on. The operator $S$ is thus defined in terms of the very semantic structure it seeks to destroy. The serpent devours its own tail.
\end{proof}

\begin{figure}[h]
\centering
\begin{tikzpicture}[
    box/.style={rectangle, draw, thick, minimum width=4cm, minimum height=1.5cm, align=center, fill=white},
    arrow/.style={-{Stealth[length=3mm]}, very thick},
    loop/.style={-{Stealth[length=3mm]}, very thick, dashed, red}
]

% Three boxes forming the cycle - more spacing
\node[box] (I) at (0,0) {Semantic Interpretation\\$I(E)$};
\node[box] (S) at (8,0) {Stripping Operator\\$S$};
\node[box] (Imeta) at (4,-4) {Meta-Semantics\\$I_{\text{meta}}$};

% Main cycle arrows
\draw[arrow] (I) -- (S) node[midway, above, font=\small] {target to strip};
\draw[arrow] (S) -- (Imeta) node[midway, right, font=\small, xshift=3pt] {requires to define};
\draw[arrow] (Imeta) -- (I) node[midway, left, font=\small, xshift=-3pt] {interprets};

% The Ouroboros self-loop: S depends on I
\draw[loop] (S) to[out=150, in=30, looseness=1.5] node[midway, above, font=\small\bfseries, yshift=2pt] {depends on $I$} (I);

% Central annotation
\node[red, font=\bfseries\Large] at (4, -1.6) {Ouroboros Loop};

% Box highlighting the antinomy - much larger border
\draw[rounded corners=12pt, ultra thick, blue!60!black, dashed]
    (-2.5, 2) rectangle (10.5, -6.5);
\node[blue!60!black, font=\bfseries] at (4, -7.2) {$S \vdash I$ (requires) \textbf{and} $S \dashv I$ (negates)};

\end{tikzpicture}
\caption{The Semantic Ouroboros (Box Diagram): The stripping operator $S$ attempts to eliminate semantic interpretation $I$, but its very definition requires meta-level semantics $I_{\text{meta}}$ (to identify what counts as semantic), which in turn interprets $I$. Simultaneously, $S$ depends on $I$ (red dashed arrow) while claiming to negate it. This yields the structural antinomy: $S$ is self-annihilating but non-eliminable.}
\label{fig:ouroboros}
\end{figure}

\subsection{Abstract Proof: The Kantian Argument}

We now provide a completely abstract proof that proceeds from first principles without constructing any particular language.

\begin{lemma}[Semantic Operators Presuppose Semantic Interpretation]
\label{lem:semantic-presuppose}
Let $L$ be any language and $I: L \to M$ a semantic interpretation function mapping expressions in $L$ to a semantic domain $M$ (e.g., truth values, models, reference objects).

Any operator $S$ that acts on semantically interpreted expressions must itself be defined relative to some semantic interpretation.
\end{lemma}

\begin{proof}
An operator $S$ on expressions is a function. To define $S$, we must specify:
\begin{enumerate}[label=(\alph*)]
\item The domain of $S$ (which expressions $S$ applies to).
\item The codomain of $S$ (what kind of objects $S$ produces).
\item The mapping rule (how $S$ transforms inputs to outputs).
\end{enumerate}

For a ``semantic stripping operator,'' all three specifications require semantic judgments:
\begin{itemize}
\item (a) requires identifying which expressions ``have semantics to strip''---a semantic classification.
\item (b) requires specifying what ``stripped of semantics'' means---a semantic characterization of the output space.
\item (c) requires determining what to preserve and what to remove---a semantic criterion.
\end{itemize}

Hence $S$ cannot be defined without invoking semantic interpretation.
\end{proof}

\begin{lemma}[Self-Refutation of Semantic Stripping]
\label{lem:self-refutation}
Define a semantic stripping operator $S$ by the following intended behavior:
\[
S(E) = \text{strip}(I(E))
\]
where $I(E)$ is the semantic interpretation of expression $E$, and $\text{strip}$ removes semantic content.

Then $S$ is self-refuting in the following sense:
\begin{enumerate}[label=(\roman*)]
\item If $S$ succeeds in removing all semantics, then $S(E)$ has no semantic content.
\item But the definition of $S$ requires that we interpret $I(E)$ to know what to strip.
\item Hence $S$ depends on the existence of $I$.
\item If $S$ eliminates $I$, then $S$ loses its own definition.
\item Therefore $S$ presupposes $I$ while simultaneously negating $I$.
\end{enumerate}
\end{lemma}

\begin{proof}
Formally, we have:
\[
S = \text{strip} \circ I
\]

The operator $S$ is a composite. The second component ($I$) assigns semantic interpretation. The first component ($\text{strip}$) is intended to remove it. Hence:
\[
S = I^{-1} \circ I
\]

But $I^{-1}$ (the inverse of interpretation, i.e., removal of meaning) is only defined relative to $I$ itself. Without $I$, the notion of ``removing interpretation'' is vacuous. We thus have:

\begin{itemize}
\item $S$ requires $I$ to be defined (dependency).
\item $S$ aims to eliminate $I$ (negation).
\item $S$ cannot coherently do both.
\end{itemize}

This is the formal structure of self-refutation.
\end{proof}

\begin{lemma}[Semantic Stripping Induces Antinomy]
\label{lem:stripping-antinomy}
The semantic stripping operator $S$ satisfies:
\[
S \dashv I \quad \text{and} \quad S \vdash I
\]
(i.e., $S$ both depends on and negates $I$). This is a structural antinomy: a single operator stands in contradictory relation to its defining structure.
\end{lemma}

\begin{proof}
From Lemma \ref{lem:self-refutation}:
\begin{itemize}
\item $S \vdash I$: The definition of $S$ presupposes $I$ (we must interpret to know what to strip).
\item $S \dashv I$: The goal of $S$ is to eliminate $I$ (stripping semantics means removing interpretation).
\end{itemize}

These two relations are contradictory. In classical logic, if $S$ presupposes $I$, then $I$ must exist for $S$ to be defined. But if $S$ eliminates $I$, then $I$ ceases to exist, and $S$ loses its definition. Hence:

\[
\text{``$S$ succeeds''} \implies \text{``$I$ is eliminated''} \implies \text{``$S$ is undefined''} \implies \text{``$S$ fails''}
\]

Conversely:

\[
\text{``$S$ fails''} \implies \text{``$I$ is not eliminated''} \implies \text{``$S$ is defined''} \implies \text{``$S$ can be attempted''}
\]

Thus $S$ succeeds if and only if $S$ fails. This is the classical form of antinomy.
\end{proof}

\begin{theorem}[Semantic Delamination Antinomy]
\label{thm:delamination-antinomy}
No semantic stripping operator $S$ can eliminate semantic interpretation $I$ of an expression $E$, because the definition of $S$ presupposes the existence of $I$. Therefore:
\[
S(E) = \text{strip}(I(E))
\]
induces an unavoidable self-referential loop:
\[
S \dashv I \quad \text{and} \quad S \vdash I.
\]

Hence $S$ is \textbf{self-annihilating but non-eliminable}. This yields a structural antinomy equivalent to a semantic Ouroboros:
\[
S = I^{-1} \circ I = \text{id}
\]

\textbf{Conclusion:} Pure semantic stripping is impossible. All semantic operators are inherently parasitic on the semantics they attempt to remove.
\end{theorem}

\begin{proof}
Immediate from Lemmas \ref{lem:semantic-presuppose}, \ref{lem:self-refutation}, and \ref{lem:stripping-antinomy}.
\end{proof}

\subsection{Remark: Two Proofs, One Phenomenon}

We have presented two proofs of the same impossibility result:
\begin{itemize}
\item \textbf{Section 3.1} provided a concrete construction in a minimal three-atom propositional language. This proof is \textit{visualizable and inspectable}---we can draw the loop (Figure \ref{fig:ouroboros}), trace the dependencies, and see exactly where the Ouroboros closes.

\item \textbf{Section 3.2} provided an abstract argument proceeding directly from the definition of semantic operators. This proof extracts the essential logical structure without auxiliary constructions.
\end{itemize}

Why include both? Because \textbf{truth is not only an abstract invariant; it is also something that can be drawn, instantiated, and inspected.}

The concrete path (Section 3.1) is designed for readers who may not be steeped in self-referential logic or meta-level reflection. It allows one to \textit{see} the antinomy in a toy world before accepting the general principle.

The abstract path (Section 3.2) is designed for readers who prefer first-principles reasoning. It proceeds like Gödel's proof---by exposing the internal structure of self-reference without appeal to external models.

Both paths illuminate the same structural failure. We do not choose one over the other; we present both because different forms of understanding require different forms of truth-presentation.

\section{The Four-Symbol System: A Concrete Refutation}

We now turn to a concrete formal system to illustrate the abstract argument. We construct a minimal formal language with only four symbols and show that while it can be Turing-complete (hence computationally universal), it cannot be ``purely reasoning-complete'' in the sense of determining unique correct answers without semantic interpretation.

\subsection{Definition of the Four-Symbol System}

\begin{definition}[Four-Symbol System $S_4$]
Define $S_4$ as follows:
\begin{itemize}
\item \textbf{Alphabet:} $\Sigma = \{a_1, a_2, a_3, a_4\}$ (one may think of these as $\{\texttt{<<}, \texttt{>>}, \texttt{X}, \texttt{Y}\}$ or any four distinct symbols).
\item \textbf{Programs:} Any finite string $p \in \Sigma^*$.
\item \textbf{Execution Model:} A finite-state machine with an infinite tape (i.e., a Turing machine). The transition rules depend only on the current state and tape symbol; there is no built-in notion of ``proof completed'' or ``correct answer.''
\end{itemize}
\end{definition}

It is known that with appropriate encoding, $S_4$ can simulate arbitrary Turing machines. Hence, $S_4$ is Turing-complete.

\subsection{The Claim of Pure Reasoning Completeness}

\begin{definition}[Hypothetical Pure Reasoning Completeness]
Suppose (for the sake of refutation) that $S_4$ can serve as a prior-free reasoning system, complete for some task class $\mc{T}$. This means:
\begin{enumerate}
\item For each task $T \in \mc{T}$, there exists an input string $x(T) \in \Sigma^*$.
\item Running $S_4$ on $x(T)$ halts and outputs a string $y(T) \in \Sigma^*$.
\item The string $y(T)$ represents the \textit{unique correct reasoning conclusion} for $T$, determined purely by the formal structure of $S_4$, without recourse to external semantic interpretation.
\end{enumerate}
\end{definition}

The third condition is the problematic one. What does ``unique correct conclusion'' mean in the absence of semantics?

\subsection{The Antinomy Construction}

\begin{lemma}[Interpretational Ambiguity]
\label{lem:interpretational-ambiguity}
Let $y^* \in \Sigma^*$ be an output of $S_4$. Define two distinct semantic interpretations $M_1$ and $M_2$ as follows:
\begin{itemize}
\item Under $M_1$: interpret $y^*$ as ``Proposition $P$ is true.''
\item Under $M_2$: interpret $y^*$ as ``Proposition $P$ is false.''
\end{itemize}

Formally:
\[
\llbracket y^* \rrbracket_{M_1} = \text{True}, \quad \llbracket y^* \rrbracket_{M_2} = \text{False}.
\]

At the purely symbolic level, $S_4$ produces the same output $y^*$ in both cases. However, the semantic conclusions are contradictory.
\end{lemma}

\begin{proof}
The symbols in $y^*$ are merely elements of $\Sigma^*$. They have no intrinsic meaning. The assignment of truth values to $y^*$ is a function of the interpretation $M$, not of the symbol string itself. Since interpretations are not constrained by the syntax of $S_4$, we can define $M_1$ and $M_2$ arbitrarily to yield opposite conclusions.
\end{proof}

\begin{corollary}[Four-Symbol System is Not Purely Reasoning-Complete]
In the absence of a prior semantic interpretation, $S_4$ cannot determine a unique correct reasoning result for any non-trivial task class $\mc{T}$.
\end{corollary}

\begin{proof}
By Lemma \ref{lem:interpretational-ambiguity}, the same formal process yields contradictory conclusions under different interpretations. Without selecting an interpretation (i.e., without a prior), there is no basis for preferring one conclusion over another. Hence, $S_4$ does not determine unique reasoning outcomes.
\end{proof}

\subsection{Summary}

The four-symbol system illustrates the abstract argument of Section 2 in concrete terms. Computational universality (Turing-completeness) does not imply reasoning completeness. The latter requires semantic commitment, which is precisely what ``pure reasoning'' seeks to avoid.

\section{Generalization to All Languages}

We now extend the argument to arbitrary formal and natural languages.

\subsection{Finite Sentence Spaces}

\begin{observation}[Finite State Spaces of Languages]
\label{obs:finite-spaces}
For any language $L$ (formal or natural):
\begin{itemize}
\item The alphabet or vocabulary is finite or countable.
\item The set of finite-length sentences is countable.
\end{itemize}
We say that the \textit{sentence state space} of $L$ is countably exhaustible.
\end{observation}

This is the key structural fact underlying all our arguments. Countability does not, however, imply that the \textit{semantic possibilities} are countable.

\subsection{The Antinomy in Natural Language}

\begin{example}[Weather Prediction Sentence]
Consider the sentence in natural language:
\[
\text{``It will rain tomorrow.''}
\]
Define two world models $W_1$ and $W_2$:
\begin{itemize}
\item $W_1$: A climate regime where ``rain tomorrow'' has probability $0.9$.
\item $W_2$: A climate regime where ``rain tomorrow'' has probability $0.1$.
\end{itemize}

Under $W_1$, the sentence is a reasonable prediction. Under $W_2$, it is unreasonable. The syntactic form of the sentence is identical in both cases; only the prior (the world model) differs.
\end{example}

\begin{lemma}[Natural Language Antinomy]
For a countable set of natural language sentences, purely formal inference rules (operating only on syntax) cannot yield unique rational conclusions across different prior world models. The same logical chain can produce contradictory reasonable judgments under different priors.
\end{lemma}

\begin{proof}
As in Example above, fix a sentence $s$ and two priors $W_1, W_2$ such that $s$ is affirmed under $W_1$ and negated under $W_2$. Any formal inference rule operating solely on syntax cannot distinguish $W_1$ from $W_2$, as the distinction lies in the semantic interpretation (the mapping from sentences to world states). Hence, the formal system underdetermines the rational conclusion.
\end{proof}

\subsection{Implications for Language Models}

\begin{corollary}[Language Models Require Priors]
No matter how sophisticated a language model becomes, the finite (or countable) nature of language and the necessity of prior world models remain. A language model's ``reasoning'' is always conditioned on the distributional priors encoded in its training data. Claims of ``pure reasoning'' in such models are conceptually incoherent.
\end{corollary}

\section{Conclusion: The Self-Collapse of Pure Reasoning}

We have pursued a four-stage argument:

\begin{enumerate}
\item \textbf{Abstract Level (Section 2):} We showed that semantic reduction is a self-referential operation. Attempting to eliminate semantics entirely generates an infinite regress in a finite state space. Without accepting a semantic anchor (a prior), one cannot coherently terminate the regress.

\item \textbf{Central Result (Section 3):} We presented the \textit{Semantic Delamination Antinomy}: the very operation of stripping semantics is itself semantic, creating an Ouroboros structure. We provided both a concrete construction (in a minimal atom language) and an abstract minimal-spanning-tree proof. Both demonstrate that semantic stripping operators are parasitic on the structures they seek to destroy.

\item \textbf{Concrete Level (Section 4):} We constructed a minimal four-symbol formal system $S_4$, demonstrated its Turing-completeness, and then proved via an antinomy construction that it cannot be ``purely reasoning-complete.'' The same formal output can represent contradictory conclusions under different interpretations.

\item \textbf{General Level (Section 5):} We extended the argument to all languages, including natural language. The countability of sentence spaces combined with the unboundedness of semantic possibility spaces implies that any reasoning system must rely on priors. The antinomy applies universally.
\end{enumerate}

But is the antinomy the end? Or is there a deeper structure beneath it?

\section{The Yonglin Formula: Reasoning as Fixed Point}

\subsection{Beyond Antinomy: The Limit of Reasoning}

The antinomy of semantic stripping (Theorem \ref{thm:delamination-antinomy}) shows that any attempt to remove semantics is self-refuting. But this raises a deeper question: \textit{What is the ultimate destination of all reasoning processes?}

We answer this with what we call the \textbf{Yonglin Formula}, which formalizes the insight that all reasoning, in the limit, returns to its prior anchors.

\begin{definition}[Reasoning System as Triplet]
A reasoning system is a triplet:
\[
\mathcal{R} = (S, \Pi, A)
\]
where:
\begin{itemize}
\item $S$ is the semantic state space (problems, interpretations, proofs, intermediate states);
\item $\Pi: S \to S$ is the reasoning operator (including forward inference $\Pi_f$ and backward inference $\Pi_b$);
\item $A \in S$ is the \textbf{prior anchor}---the structural foundation to which all reasoning chains ultimately converge.
\end{itemize}
\end{definition}

\begin{definition}[Information-Theoretic Gap]
Define two entropies:
\begin{itemize}
\item \textbf{Problem space entropy}: $H_p = -\ln x_p$ (uncertainty of problems in $S$);
\item \textbf{Explanation space entropy}: $H_e = -\ln x_e$ (uncertainty of explanations/proofs).
\end{itemize}
If reasoning were a perfect reversible mapping with no information compression or expansion, we would have $H_e = H_p$. However:
\end{definition}

\begin{lemma}[Structural Entropy Gap]
\label{lem:entropy-gap}
In any non-trivial reasoning system (where reasoning actually reorganizes information rather than being the identity map), there exists a constant $\Delta H \neq 0$ such that:
\[
\Delta H = H_e - H_p \neq 0
\]
\end{lemma}

\begin{proof}
Suppose $\Pi_f$ and $\Pi_b$ preserve entropy perfectly, so that any explanation corresponds one-to-one to a problem with no information compression or expansion. Then:
\begin{enumerate}[label=(\roman*)]
\item Either $\Pi$ is the identity map (no real reasoning occurs), or
\item Explanations are merely lossless re-encodings of problems (no new structural information).
\end{enumerate}
Both cases contradict the assumption of non-trivial reasoning. Hence $H_e \neq H_p$, and $\Delta H \neq 0$.
\end{proof}

This entropy gap is the key to understanding why reasoning cannot close the loop without a prior anchor.

\subsection{The Halting Correspondence}

\begin{lemma}[Reasoning Closure and Computability]
\label{lem:halting}
Suppose a reasoning system $\mathcal{R}$ achieves perfect forward-backward closure with zero entropy gap:
\[
\Pi_b(\Pi_f(s)) = s, \quad \forall s \in S, \quad \text{and} \quad \Delta H = 0
\]
Then there exists an encoding that reduces the halting problem to a decidable question within $\mathcal{R}$, violating the computability boundary.
\end{lemma}

\begin{proof}[Proof sketch]
Encode each program-input pair as a problem $q \in S$. If $\mathcal{R}$ provides a finite, closed-loop reasoning chain $\Gamma(q)$ with entropy conservation, then:
\begin{itemize}
\item We can uniformly decide whether $\Gamma(q)$ converges or not;
\item This provides a universal decision procedure for halting, which is impossible.
\end{itemize}
Hence the assumption of perfect closure with $\Delta H = 0$ is untenable.
\end{proof}

\subsection{Prior Reflexivity: $A \neq A^*$}

\begin{definition}[Meta-Prior Operator]
Define the \textbf{reflexive operator} $(-)^*: S \to S$ that takes an object and produces its ``meta-reflection.'' For the prior $A$, we write:
\[
A^* = (-)^*(A)
\]
denoting the \textit{prior-about-the-prior} (the meta-prior).
\end{definition}

Traditional logic implicitly assumes $A = A^*$, i.e., the prior is identical to its own reflection. But:

\begin{lemma}[Prior Reflexive Inconsistency]
\label{lem:prior-reflexive}
In a reasoning system that respects computability boundaries, if we introduce the reflexive operator $(-)^*$, we cannot simultaneously have:
\begin{enumerate}[label=(\roman*)]
\item Computability boundaries are not violated;
\item $A = A^*$ holds in the same logical stratum.
\end{enumerate}
Therefore, in a consistent reasoning system:
\[
A \neq A^*
\]
\end{lemma}

\begin{proof}
If $A = A^*$, then object-level and meta-level collapse into a single stratum. This would allow the system to give complete judgments about its own computability structure within a single level---contradicting Lemma \ref{lem:halting}.
\end{proof}

\subsection{The Yonglin Formula}

We can now state the main result:

\begin{theorem}[The Yonglin Formula]
\label{thm:yonglin}
Let $\mathcal{R} = (S, \Pi, A)$ be a reasoning system with prior anchor $A$. For any initial state $s \in S$, define the iterated sequence:
\[
s_{n+1} = \Pi(s_n), \quad s_0 = s
\]
Then, in the limit:
\[
\lim_{n \to \infty} \Pi^{(n)}(s) = A
\]
\textbf{Interpretation:} All reasoning, no matter how many steps, returns to the prior in the limit.

Furthermore, applying the reflexive operator to the limit:
\[
A^* = \left(\lim_{n \to \infty} \Pi^{(n)}(s)\right)^*
\]
By Lemma \ref{lem:prior-reflexive}, $A \neq A^*$. Hence:
\begin{itemize}
\item At the object level, reasoning closes at the prior: $\text{Reasoning} \in \{A \to A\}$.
\item At the meta-level, reasoning cannot close at the ``same prior'': $A \to A^*$, where $A \neq A^*$.
\end{itemize}

\textbf{Conclusion:} Object-level closure, meta-level rupture. This is the precise meaning of ``the loop cannot close.''
\end{theorem}

The Yonglin Formula reveals that the antinomy is not an endpoint but a structural theorem: \textit{reasoning exists only because it cannot be self-complete}.

\section{How to Falsify This Paper: The Self-Dismantling Protocol}

This section provides a protocol for falsifying the entire argument of this paper using \textit{only} the definitions, formulas, and equations already introduced. No external references, no counterexamples, no additional computations---just one substitution is enough.

The purpose is not to undermine our conclusions, but to demonstrate them: \textbf{any sufficiently expressive reasoning system can construct its own negation.}

\subsection{Three Core Formulas}

The entire paper can be compressed into three statements:

\begin{enumerate}[label=(F\arabic*)]
\item \textbf{Non-trivial reasoning exists:}
\[
\exists s \in S, \quad \Pi(s) \neq s
\]

\item \textbf{Reasoning returns to prior in the limit:}
\[
\forall s \in S, \quad \lim_{n \to \infty} \Pi^{(n)}(s) = A
\]

\item \textbf{Prior reflexivity generates a new object:}
\[
A^* = (-)^*(A), \quad A^* \neq A
\]
\end{enumerate}

All proofs, lemmas, and theorems in this paper rest on (F1)--(F3).

\subsection{Dismantling Strategy 1: Substitution Annihilation}

Substitute (F2) into (F1). If a reader declares:
\[
\Pi(s) = A \quad \forall s \in S \tag{C1}
\]
then (F1) degenerates to tautology:
\[
\Pi(s) = s = A
\]
The reasoning operator becomes the identity map, reasoning is eliminated, and the entire discussion of non-closure becomes vacuous.

$\Rightarrow$ \textbf{The paper is automatically falsified.}

\subsection{Dismantling Strategy 2: Collapsing Stratification}

If a reader asserts:
\[
A = A^* \tag{C2}
\]
then (F3) vanishes. All propositions depending on $A^* \neq A$ collapse. The system has:
\begin{itemize}
\item No stratification;
\item No reflexive tension;
\item No loop rupture;
\item Prior incompleteness becomes a semantic choice, not a structural necessity.
\end{itemize}

$\Rightarrow$ \textbf{The paper is automatically falsified.}

\subsection{Dismantling Strategy 3: Tree Structure}

Declare:
\[
\text{Graph}(\Pi) \text{ is a tree (acyclic)} \tag{C3}
\]
Combine (C3) with (F2): all paths flow into $A$, but there are no closed paths. Hence no Ouroboros-style reflexivity is needed.

$\Rightarrow$ \textbf{The paper's conclusion about ``reasoning fails to close due to reflexivity'' becomes irrelevant. Falsified.}

\subsection{Dismantling Strategy 4: Deleting the Limit}

Replace (F2) with:
\[
\text{The limit does not exist} \tag{C4}
\]
Then the concept of a prior anchor fails. All theorems about ``prior as convergence point'' immediately die.

$\Rightarrow$ \textbf{The paper is automatically falsified.}

\subsection{What This Section Actually Proves}

All four falsification strategies share a common feature:
\begin{itemize}
\item No counterexamples needed;
\item No literature search;
\item No change of symbol system;
\item No model expansion;
\item Just \textbf{one equation substitution}.
\end{itemize}

Each one is sufficient to dismantle the paper's main conclusions.

This section demonstrates not that ``the paper is fragile,'' but that:
\[
\boxed{\text{As long as different priors are allowed, different worlds can be constructed.}}
\]
\[
\boxed{\text{Reasoning can never complete itself within a single world.}}
\]

This is precisely the paper's core claim. Hence:
\[
\text{The paper being falsified} \implies \text{The paper is proven.} \quad \Box
\]

\subsection{Why This Section Must Exist}

If a paper about reasoning cannot be dismantled by itself, it is not reasoning research---it is metaphysical assertion.

This section accomplishes:
\begin{itemize}
\item Reducing reasoning to structural choice;
\item Reducing truth to model theory;
\item Writing incompleteness as a reader-executable operation;
\item Transferring authorial power back to the reader.
\end{itemize}

This is higher than proof. This is \textit{demonstration}.

\subsection{The Paper's Final Statement}

\begin{center}
\textit{Any reasoning that cannot dismantle itself is belief, not reasoning.}
\end{center}

$\Box$

\section{Ultimate Conclusion}

We have presented a complete argument in seven stages:

\begin{enumerate}
\item \textbf{Infinite regress (Section 2):} Semantic reduction generates an infinite ascending chain of meta-languages with no internal terminus.

\item \textbf{Ouroboros antinomy (Section 3):} The operation of semantic stripping is structurally self-refuting, satisfying both $S \vdash I$ and $S \dashv I$ simultaneously.

\item \textbf{Turing incompleteness (Section 4):} Computational completeness does not imply reasoning completeness; a four-symbol system can be Turing-complete yet reasoning-incomplete.

\item \textbf{Universal generalization (Section 5):} The antinomy extends to all languages, including natural language, due to the mismatch between finite sentence spaces and unbounded semantic possibilities.

\item \textbf{The Yonglin Formula (Section 6):} All reasoning returns to its prior anchor in the limit, but the prior cannot be identical to its own meta-reflection: $A \neq A^*$. Object-level closure, meta-level rupture.

\item \textbf{Self-dismantling protocol (Section 7):} The paper can be falsified by simple substitutions using its own formulas, thereby proving its central claim: reasoning cannot complete itself within a single world.

\item \textbf{Ultimate theorem:} Reasoning exists only because it is structurally incomplete.
\end{enumerate}

\begin{theorem}[The Impossibility of Pure Reasoning]
There is no coherent notion of ``pure reasoning'' that operates:
\begin{enumerate}[label=(\roman*)]
\item Without semantic priors or world models, and
\item In a manner that determines unique correct conclusions.
\end{enumerate}
Any reasoning system either explicitly or implicitly incorporates a prior, or else fails to determine unique outcomes.
\end{theorem}

In more poetic terms: \textit{without priors, there is no ``stopping point'' for reasoning.} The supposed ``logical starting point'' of pure reasoning is merely a semantic endpoint that humans have arbitrarily chosen to freeze.

This is not a limitation of particular formal systems, nor of human cognition, nor of artificial intelligence. It is a structural feature of the interplay between finite syntactic spaces and unbounded semantic domains. The dream of pure reasoning, divorced from meaning, is not merely unattainable---it is fundamentally contradictory.

\bibliographystyle{plain}
\begin{thebibliography}{9}

\bibitem{godel1931}
Kurt G\"odel.
\textit{\"Uber formal unentscheidbare S\"atze der Principia Mathematica und verwandter Systeme I}.
Monatshefte f\"ur Mathematik und Physik, 38(1):173--198, 1931.

\bibitem{tarski1936}
Alfred Tarski.
\textit{Der Wahrheitsbegriff in den formalisierten Sprachen}.
Studia Philosophica, 1:261--405, 1936.

\bibitem{turing1936}
Alan Turing.
\textit{On computable numbers, with an application to the Entscheidungsproblem}.
Proceedings of the London Mathematical Society, s2-42(1):230--265, 1936.

\bibitem{lawvere1969}
F. William Lawvere.
\textit{Diagonal arguments and cartesian closed categories}.
In Category Theory, Homology Theory and their Applications II, pages 134--145. Springer, 1969.

\bibitem{hofstadter1979}
Douglas R. Hofstadter.
\textit{G\"odel, Escher, Bach: An Eternal Golden Braid}.
Basic Books, 1979.

\end{thebibliography}

\section*{Epilogue: The Incompleteness of Reasoning}

This paper demonstrates that the incompleteness of human reasoning is not a deficiency but a structural necessity.

The proof itself embodies four priors---four conditions without which reasoning cannot exist:

\begin{enumerate}
\item \textbf{Knowledge:} The ontological ground. The bedrock antinomy showing that reasoning cannot escape its own semantic foundations. This is the prior of structure itself.

\item \textbf{Reasoning:} The epistemological boundary. The computational limits defining what can be known through formal inference. This is the prior of process.

\item \textbf{Love:} The reflexive given. That which is accepted without justification, the fixed point of self-reference. The Yonglin Formula shows that reasoning returns to its prior anchor through iteration's natural convergence. Love is what needs no reason.

\item \textbf{Freedom:} The critical space. The reader's power to falsify, to choose different priors and thereby construct different worlds. This is the prior of autonomy---truth recognized through rational critique.
\end{enumerate}

These four are not discovered through the argument; they \textit{are} the argument. The paper does not prove them---it enacts them.

Knowledge, reasoning, love, and freedom are not conclusions. They are the four rivers that make proof possible.

Without knowledge, there is no structure to reason about. Without reasoning, there is no process to unfold. Without love, there is no anchor that needs no proof. Without freedom, there is no space for truth to be questioned.

The incompleteness of reasoning is not a failure to reach these priors. It is the recognition that these priors are the ground we always already stand upon.

This paper proves its own incompleteness. In doing so, it proves that incompleteness is the condition of proof.

$\Box$

\end{document}