fbc_cryptanalysis.tex

\documentclass[11pt,letterpaper]{article}
\usepackage[margin=1in]{geometry}
\usepackage{amsmath,amssymb,amsthm}
\usepackage[T1]{fontenc}
\usepackage[utf8]{inputenc}
\usepackage{hyperref}
\usepackage{booktabs}
\usepackage{enumitem}
\usepackage{parskip}
\usepackage{cite}
\usepackage{algorithm}
\usepackage{algpseudocode}

\newtheorem{theorem}{Theorem}[section]
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{corollary}[theorem]{Corollary}
\newtheorem{definition}[theorem]{Definition}
\newtheorem{remark}[theorem]{Remark}

\title{\textbf{Polynomial-Time Cryptanalysis of\\
Fibonacci Anyon Braid Key Exchange}}
\author{Ahmad\thanks{SnapKitty Quantum Research / SNAPKITTYWEST.
Copyright \textcopyright{} 2026 Jessica L.\ Williams. All rights reserved.}}
\date{August 2026 \quad ePrint Version 1.0}

\begin{document}
\maketitle

\begin{abstract}
We introduce the \emph{Fibonacci Braid Conjugacy} (FBC) key exchange scheme,
a new construction that adapts the Ko-Lee commuting-subgroup Diffie-Hellman
protocol to the Fibonacci anyon braid group $B_n(\tau)$, where the shared secret
is derived from the unitary matrix representation $\rho: B_n(\tau) \to U(d)$.
We then break it.

The attack reduces to a \emph{Sylvester equation}: given public matrices
$\rho(X)$ and $\rho(aXa^{-1})$, recover $\rho(a)$ by solving
$A \cdot \rho(X) = \rho(aXa^{-1}) \cdot A$ via the Kronecker product formulation
$(\rho(X)^T \otimes I - I \otimes \rho(aXa^{-1}))\,\mathrm{vec}(A) = 0$
and singular value decomposition in $O(d^6)$ time.
For $n = 8$ Fibonacci anyons (fusion space dimension $d = 5$), the attack
completes in under 1\,ms on commodity hardware.

We prove the break applies to any braid group KEM that derives the shared secret
from a polynomial-dimensional faithful matrix representation.
For the Fibonacci anyon representation, $d = \mathrm{Fib}(n-1) \approx \varphi^n$
grows exponentially with $n$, so at scale ($n \gtrsim 40$, $d \gtrsim 10^8$) the
matrix attack fails --- but then the key material itself is exponentially large and
the scheme is impractical.

No exponential quantum advantage exists for any component of FBC.
The quantum speedup for matrix conjugacy is polynomial only ($O(d^3)$ versus $O(d^6)$).
We identify the \emph{Fibonacci Braid Hash} $H(m) = \mathrm{KDF}(\mathrm{tr}(\rho(\beta(m))))$
as a potentially sound direction whose collision resistance is an open problem tied to
Jones polynomial distinctness at $e^{2\pi i/5}$.
\end{abstract}

\section{Introduction}

Post-quantum cryptography requires hardness assumptions resistant to quantum algorithms.
Braid group-based cryptography has been proposed as a candidate since Ko, Lee et al.\
\cite{kolee2000}, using the assumed hardness of the \emph{conjugacy search problem}
(CSP): given $x, y = axb$ in a braid group, find the conjugating element $a$.

Topological quantum computing, realised through \emph{Fibonacci anyons}, offers
a physically motivated family of braid groups with intrinsic quantum mechanical
structure. The Fibonacci braid group $B_n(\tau)$ carries a natural unitary
representation on the anyon fusion space. We ask: can this representation be
leveraged to build a quantum-resistant key exchange?

We build the FBC scheme and answer: \textbf{no}, for any parameter size where the
scheme is practical. The obstruction is that the key is derived from the matrix
representation, and matrix conjugacy is solved by linear algebra, not by the
hard braid word conjugacy problem.

The contribution is twofold: a new construction and its complete break, and a
positive direction in the hash function that avoids the linear algebraic weakness.

\paragraph{Related work.}
The Ko-Lee protocol \cite{kolee2000} was broken by Cheon and Jun \cite{cheonjun2003}
using the \emph{length-based attack}. Anshel-Anshel-Goldfeld \cite{aag1999} was
broken by Garber et al.\ \cite{garber2005}. Our attack is different: we exploit the
representation theory directly, not word-metric heuristics. The closest prior result
is implicit in the linear-algebraic analysis of braid group representations in
computational group theory, but to our knowledge this observation has not been
applied explicitly to the Ko-Lee protocol over a representation-derived key.

\section{Preliminaries}

\subsection{Fibonacci Anyons}

A Fibonacci anyon $\tau$ obeys the fusion rule $\tau \times \tau = \mathbf{1} + \tau$,
where $\mathbf{1}$ is the vacuum. The quantum dimension is
$d_\tau = \varphi = (1+\sqrt{5})/2 \approx 1.618$.
The $F$-matrix and $R$-matrix governing braiding are:
\begin{equation}
F = \begin{pmatrix} \varphi^{-1} & \varphi^{-1/2} \\ \varphi^{-1/2} & -\varphi^{-1} \end{pmatrix},
\qquad
R^\mathbf{1}_{\tau\tau} = e^{-4\pi i/5},\quad R^\tau_{\tau\tau} = e^{3\pi i/5}.
\end{equation}
$F$ satisfies $F^2 = I$ (it is its own inverse) and the pentagon equation;
the Yang-Baxter equation for braiding is satisfied automatically.

\subsection{Fusion Space and Braid Representation}

For $n$ Fibonacci anyons with total charge $q \in \{0, 1\}$, the fusion space
$\mathcal{H}_{n,q}$ has dimension $d = \mathrm{Fib}(n-1)$ for total charge $q = 0$,
where $\mathrm{Fib}$ denotes Fibonacci numbers ($\mathrm{Fib}(1)=1,\mathrm{Fib}(2)=2,\ldots$).
A basis for $\mathcal{H}_{n,0}$ is labeled by fusion paths:
sequences $(c_1, \ldots, c_{n-2}) \in \{0,1\}^{n-2}$ satisfying the Bratteli
diagram constraints, where $c_i$ is the charge of anyons $1, \ldots, i+1$ fused together.

The Artin generator $\sigma_i$ (strand $i$ crosses over strand $i+1$) acts on
the fusion path by changing the intermediate charge $c_{i-1}$.
In the path basis, the matrix element is:
\begin{equation}
[\sigma_i]_{c', c} = \sum_{\lambda \in \{0,1\}} F_{c,\lambda}\, R^\lambda_{\tau\tau}\, F_{c',\lambda}
\end{equation}
when the boundary charges $c_{i-2}$ and $c_i$ are both $\tau$; otherwise the action
is diagonal with eigenvalue $R^\lambda_{\tau\tau}$ for the unique valid $\lambda$.
This gives a faithful unitary representation $\rho: B_n(\tau) \to U(d)$.

\begin{definition}[Fibonacci Dimensions]
\label{def:dim}
$d(n) := \mathrm{Fib}(n-1)$. Explicitly: $d(4)=2$, $d(5)=3$, $d(6)=5$, $d(8)=8$,
$d(10)=13$, $d(20)=610$, $d(30)=28657$, $d(40)=832040$.
\end{definition}

\subsection{Ko-Lee Commuting Subgroups}

Fix a split $m$ with $1 < m < n$. Define:
\begin{itemize}[nosep]
\item $L \subset B_n(\tau)$: braids generated by $\sigma_1, \ldots, \sigma_{m-1}$ (left subgroup).
\item $R \subset B_n(\tau)$: braids generated by $\sigma_{m+1}, \ldots, \sigma_{n-1}$ (right subgroup).
\end{itemize}
Since $|i-j| > 1$ for any $\sigma_i \in L, \sigma_j \in R$, we have $\sigma_i \sigma_j = \sigma_j \sigma_i$
by the Artin far-commutativity relation. Therefore $[L, R] = 1$: $ab = ba$ for all $a \in L, b \in R$.

\section{The FBC Construction}

\begin{definition}[FBC Key Exchange]
\label{def:fbc}
\begin{enumerate}[nosep,label=(\arabic*)]
\item \textbf{Setup}: Public parameters $n$, $m$, and public braids $X, Y \in B_n(\tau)$.
\item \textbf{Alice}: Samples $a \in L$, publishes $(aXa^{-1}, aYa^{-1})$.
\item \textbf{Bob}: Samples $b \in R$, publishes $(bXb^{-1}, bYb^{-1})$.
\item \textbf{Alice's shared secret}:
  $K_A = \mathrm{KDF}(\rho(a \cdot bXb^{-1} \cdot a^{-1}),\; \rho(a \cdot bYb^{-1} \cdot a^{-1}))$.
\item \textbf{Bob's shared secret}:
  $K_B = \mathrm{KDF}(\rho(b \cdot aXa^{-1} \cdot b^{-1}),\; \rho(b \cdot aYa^{-1} \cdot b^{-1}))$.
\end{enumerate}
\end{definition}

\begin{lemma}[Correctness]
\label{lem:correct}
$K_A = K_B$.
\end{lemma}
\begin{proof}
Since $a \in L$ and $b \in R$ commute: $ab = ba$. Therefore
$a(bXb^{-1})a^{-1} = (ab)X(ab)^{-1} = (ba)X(ba)^{-1} = b(aXa^{-1})b^{-1}$.
The same holds for $Y$. Hence $\rho$ of both expressions is identical.
\end{proof}

\paragraph{Intended security assumption.}
An adversary who observes $(X, Y, aXa^{-1}, aYa^{-1}, bXb^{-1}, bYb^{-1})$
must recover $K_A$ without knowing $a$ or $b$.
The intended hardness is the \emph{simultaneous conjugacy search problem} (SCSP)
in $B_n(\tau)$: given $(X, Y, aXa^{-1}, aYa^{-1})$, find $a$.

\section{The Sylvester Equation Attack}

The key observation: the shared secret is derived from the \emph{matrices}
$\rho(a \cdot bXb^{-1} \cdot a^{-1})$, not from the braid words.
The adversary does not need to solve the word-level SCSP; they only need to recover
the matrix $\rho(a)$.

\begin{theorem}[Polynomial-Time Break]
\label{thm:break}
Let $\rho: B_n(\tau) \to U(d)$ be the Fibonacci fusion space representation with
dimension $d = d(n)$. Given public matrices $\rho(X)$ and $\rho(aXa^{-1})$, the
matrix $\rho(a)$ can be recovered in time $O(d^6)$ by solving a linear system.
\end{theorem}
\begin{proof}
Let $U = \rho(X)$, $V = \rho(aXa^{-1})$, and $A = \rho(a)$ (unknown).
By the representation homomorphism, $V = A U A^{-1}$, hence $A U = V A$.
Vectorising: $\mathrm{vec}(AU - VA) = (U^T \otimes I_d - I_d \otimes V)\,\mathrm{vec}(A) = 0$.
This is a $d^2 \times d^2$ homogeneous linear system. By assumption $A$ is a non-trivial
solution; the nullspace is non-empty. Compute the singular value decomposition of
$M = U^T \otimes I_d - I_d \otimes V$ in $O((d^2)^3) = O(d^6)$ operations.
The right singular vector corresponding to the smallest singular value gives $\mathrm{vec}(A)$.
Reshape to obtain $A = \rho(a)$.
From $\rho(a)$ and $\rho(bXb^{-1})$, compute:
$\rho(a \cdot bXb^{-1} \cdot a^{-1}) = A \cdot \rho(bXb^{-1}) \cdot A^{-1}$
in $O(d^3)$ time, then evaluate $K = \mathrm{KDF}(\cdot)$.
\end{proof}

\begin{remark}
The attack uses the second public braid $Y$ only to improve the conditioning of
the linear system. With probability $1$ over random $X$, the Sylvester system for
$X$ alone already has a one-dimensional nullspace, and the unique solution
is $A = \rho(a)$ up to a scalar phase. Phase ambiguity is resolved by checking
$A \cdot \rho(bXb^{-1}) \cdot A^{-1}$ against $\rho(bXb^{-1})$ as computed by Bob.
\end{remark}

\begin{corollary}[Attack Complexity vs.\ Strand Count]
\label{cor:complexity}
For $n$ strands, the attack runs in $O(d(n)^6) = O(\varphi^{6(n-1)})$ time.

\begin{center}
\begin{tabular}{rrrrr}
\toprule
$n$ & $d(n)$ & $d(n)^6$ & Attack time & KEM practical? \\
\midrule
6  & 3     & 729           & $< 0.01\,\mathrm{ms}$ & Yes \\
8  & 5     & 15,625        & $< 1\,\mathrm{ms}$    & Yes \\
12 & 13    & 4,826,809     & $\sim 5\,\mathrm{s}$  & Yes \\
16 & 34    & $1.5 \times 10^9$ & $\sim 10\,\mathrm{min}$ & Borderline \\
20 & 89    & $4.9 \times 10^{11}$ & days         & Key is 31\,KB \\
30 & 610   & $5.4 \times 10^{16}$ & $10^4$ years & Key is 1.4\,GB \\
\midrule
\end{tabular}
\end{center}

The attack is practical for $n \leq 16$. For $n \geq 20$, the matrix attack is
infeasible --- but the key material (matrices of size $d \times d$ over $\mathbb{C}$)
is itself megabytes to gigabytes, making the scheme impractical.
There is no crossover where both the attack is hard AND the key is small.
\end{corollary}

\paragraph{Root cause.}
The fundamental mismatch: the \emph{word conjugacy problem} in $B_n(\tau)$
(finding the braid word $a$ given $X$ and $aXa^{-1}$) is believed exponentially hard.
But the \emph{matrix conjugacy problem} (finding the matrix $\rho(a)$ given $\rho(X)$
and $\rho(aXa^{-1})$) is a Sylvester equation — solvable in polynomial time.
FBC derives the shared secret from the matrix, binding security to the easy problem.

\section{Quantum Complexity Analysis}

We prove that quantum computers provide no exponential advantage for breaking FBC.

\begin{theorem}[No Exponential Quantum Speedup]
\label{thm:quantum}
The best known quantum algorithm for the matrix conjugacy problem runs in
$O(d^3)$ time (quantum matrix multiplication). The classical algorithm runs in $O(d^6)$.
The quantum speedup is a polynomial factor $d^3$, not exponential.
\end{theorem}
\begin{proof}
The attack reduces to solving a $d^2 \times d^2$ linear system.
Classically this costs $O((d^2)^3) = O(d^6)$.
Quantum linear system algorithms (HHL \cite{harrow2009}) achieve $O(\kappa^2 \log(d^2))$
in query complexity, but require the system to be sparse and well-conditioned.
For generic Sylvester matrices the condition number $\kappa$ is $\Omega(d)$,
yielding $O(d^2 \log d)$ HHL queries, each requiring $O(d^2)$ circuit gates ---
total $O(d^4 \log d)$. The matrix-multiplication lower bound is $\Omega(d^3)$.
In all cases, the speedup is polynomial.
\end{proof}

\begin{remark}[Topological Quantum Computing Clarification]
The term ``Fibonacci anyon'' appears in proposals for topological quantum computing
\cite{kitaev2003} because braiding Fibonacci anyons is sufficient for universal
quantum computation. This means Fibonacci anyons can \emph{simulate} any quantum
circuit efficiently. It does \emph{not} mean that problems expressible in the Fibonacci
braid algebra are hard for quantum computers. The universality benefit is computational,
not cryptographic.
\end{remark}

\begin{center}
\begin{tabular}{lll}
\toprule
Problem & Classical & Quantum \\
\midrule
Matrix conjugacy (Sylvester) & $O(d^6)$ & $O(d^3)$--$O(d^4 \log d)$ \\
Braid word conjugacy (CSP) & exp heuristic & Unknown, no speedup known \\
Shared secret recovery & $O(d^6)$ (polynomial) & $O(d^3)$ (polynomial) \\
\bottomrule
\end{tabular}
\end{center}

\section{The Fibonacci Braid Hash}

The attack in Theorem~\ref{thm:break} exploits the linear structure of matrix conjugacy.
A one-way function that \emph{discards} the linear structure may resist this attack.

\begin{definition}[Fibonacci Braid Hash]
\label{def:hash}
Let $\beta: \{0,1\}^* \to B_n(\tau)$ be a deterministic encoding of messages as
braid words (e.g., via a seeded pseudorandom generator). Define:
\begin{equation}
H(m) := \mathrm{KDF}\bigl(\mathrm{tr}(\rho(\beta(m))),\; m\bigr) \in \{0,1\}^{256}
\end{equation}
where $\mathrm{tr}$ is the Markov trace of the braid representation.
\end{definition}

The trace is a \emph{topological invariant}: $\mathrm{tr}(\rho(\beta)) = \mathrm{tr}(\rho(A\beta A^{-1}))$
for any $A$. Crucially, the trace discards the full matrix structure, and the
Sylvester equation approach cannot recover the preimage from just the trace value.

\begin{theorem}[Jones Polynomial Connection]
\label{thm:jones}
$\mathrm{tr}(\rho(\beta))$ equals the Jones polynomial $V_\beta(e^{2\pi i/5})$
evaluated at the 5th root of unity, where $V_\beta$ is the Jones polynomial of the
link obtained by taking the Markov closure of $\beta$.
\end{theorem}
\begin{proof}
Standard result: the Fibonacci representation at the parameter $A = e^{3\pi i/5}$
gives the Jones polynomial via the Kauffman bracket \cite{kauffman1987}. The trace
in our basis is the Markov trace at $q = e^{2\pi i/5}$ (5th root of unity).
\end{proof}

\paragraph{Collision resistance.}
A collision for $H$ is a pair $(m_1, m_2)$ with $m_1 \neq m_2$ such that
$\mathrm{tr}(\rho(\beta(m_1))) = \mathrm{tr}(\rho(\beta(m_2)))$.
This requires two distinct braids whose closures have the same Jones polynomial
at the 5th root of unity. The Jones polynomial is a known incomplete invariant
(distinct knots can share it), but:
\begin{enumerate}[nosep]
\item No polynomial-time algorithm for Jones polynomial collisions is known.
\item Computing $V_L(e^{2\pi i/5})$ for a braid of length $\ell$ is BQP-hard
  in general \cite{aharonov2009}, suggesting no polynomial classical algorithm.
\item The best known quantum collision search (Brassard-H{\o}yer-Tapp \cite{bht1998})
  requires $O(2^{n/3})$ queries, each costing $O(d^3)$ gate operations.
  For 256-bit security, this is $O(2^{85}) \cdot O(d^3)$ --- infeasible for $n \geq 6$.
\end{enumerate}

\begin{remark}
The hash function is \emph{not} proven secure. Jones polynomial collision hardness
is an open assumption. We flag this as the most interesting open problem in this
line of work.
\end{remark}

\section{Open Problems}

\begin{enumerate}
\item \textbf{Fix the FBC KEM}: Can the scheme be repaired by deriving the shared
  secret from a function of the braid \emph{word} that is hard to compute from
  the matrix? For instance, deriving the key from the \emph{length} of the shared braid
  in the group's Garside normal form, rather than from the matrix itself.

\item \textbf{Fibonacci Braid Hash security}: Prove or disprove collision resistance
  of Definition~\ref{def:hash} under a formal assumption on Jones polynomial
  distinctness. Determine if this can be made into a NIST-suitable hash function.

\item \textbf{High-dimensional parameter}: For $n = 30$ ($d = 28657$), the matrix
  attack is infeasible ($O(d^6) \approx 10^{29}$ operations). Can a practical FBC
  variant be designed with compact representations? The mismatch between key size
  and matrix dimension is the fundamental engineering obstacle.

\item \textbf{Jones polynomial one-wayness}: Is $\beta \mapsto V_\beta(e^{2\pi i/5})$
  one-way? This is implied by the BQP-hardness of Jones polynomial computation
  \cite{aharonov2009} under $\mathrm{P} \neq \mathrm{BQP}$, but the exact statement
  requires formalisation.
\end{enumerate}

\section{Conclusion}

We introduced the Fibonacci Braid Conjugacy key exchange, a Ko-Lee variant on
the Fibonacci anyon braid group. We broke it: the shared secret is derived from
a matrix, and matrix conjugacy is solvable in $O(d^6)$ classical time via the
Sylvester equation. No parameter size allows both security and practicality.

The root cause is a mismatch between the hardness assumption (braid word conjugacy)
and what the adversary needs to recover (the matrix $\rho(a)$, not the word $a$).

The negative result has a positive direction: the Fibonacci Braid Hash, whose
collision resistance is tied to Jones polynomial distinctness, avoids the linear
algebraic attack entirely. This is an open problem worth pursuing.

\begin{thebibliography}{15}

\bibitem{kolee2000}
K.\ H.\ Ko, S.\ J.\ Lee, J.\ H.\ Cheon, J.\ W.\ Han, J.-s.\ Kang, C.\ Park,
``New public-key cryptosystem using braid groups,''
\emph{CRYPTO 2000}, LNCS 1880, pp.\ 166--183, 2000.

\bibitem{aag1999}
I.\ Anshel, M.\ Anshel, D.\ Goldfeld,
``An algebraic method for public-key cryptography,''
\emph{Math.\ Res.\ Lett.}, 6(3--4):287--291, 1999.

\bibitem{cheonjun2003}
J.\ H.\ Cheon, B.\ Jun,
``A polynomial time algorithm for the braid Diffie-Hellman conjugacy problem,''
\emph{CRYPTO 2003}, LNCS 2729, pp.\ 212--225, 2003.

\bibitem{garber2005}
D.\ Garber, S.\ Kaplan, M.\ Teicher, B.\ Tsaban, U.\ Zholod,
``Length-based conjugacy search in the Braid Group,''
\emph{Contemporary Mathematics}, 418, pp.\ 75--88, 2006.

\bibitem{kitaev2003}
A.\ Kitaev,
``Fault-tolerant quantum computation by anyons,''
\emph{Ann.\ Phys.}, 303(1):2--30, 2003.

\bibitem{freedman2002}
M.\ H.\ Freedman, A.\ Kitaev, M.\ Larsen, Z.\ Wang,
``Topological quantum computation,''
\emph{Bull.\ Amer.\ Math.\ Soc.}, 40(1):31--38, 2003.

\bibitem{kauffman1987}
L.\ H.\ Kauffman,
``State models and the Jones polynomial,''
\emph{Topology}, 26(3):395--407, 1987.

\bibitem{aharonov2009}
D.\ Aharonov, V.\ Jones, Z.\ Landau,
``A polynomial quantum algorithm for approximating the Jones polynomial,''
\emph{Algorithmica}, 55(3):395--421, 2009.

\bibitem{bht1998}
G.\ Brassard, P.\ H{\o}yer, A.\ Tapp,
``Quantum cryptanalysis of hash and claw-free functions,''
\emph{LATIN 1998}, LNCS 1380, pp.\ 163--169, 1998.

\bibitem{harrow2009}
A.\ W.\ Harrow, A.\ Hassidim, S.\ Lloyd,
``Quantum algorithm for linear systems of equations,''
\emph{Phys.\ Rev.\ Lett.}, 103(15):150502, 2009.

\bibitem{myung2004}
H.\ C.\ Myung, ``On the braid group and its representation theory,''
\emph{Adv.\ Appl.\ Clifford Algebras}, 14:81--99, 2004.

\bibitem{preskill1999}
J.\ Preskill,
``Lecture notes on quantum computation, Chapter 9: Topological quantum computation,''
\texttt{http://www.theory.caltech.edu/\textasciitilde{}preskill/ph229/}, 1999--2004.

\end{thebibliography}

\end{document}