Proof of the ADE Classification ================================ We prove that a connected simple graph has a finite root system under the mutation game if and only if it is one of the Dynkin diagrams :math:`A_n`, :math:`D_n`, :math:`E_6`, :math:`E_7`, or :math:`E_8`. The spectral criterion ---------------------- Let :math:`G` be a connected simple graph on :math:`n` vertices with adjacency matrix :math:`A` and Cartan matrix :math:`C = 2I - A`. As established in :doc:`background`, the root system is finite if and only if :math:`C` is positive definite. Since the eigenvalues of :math:`C` are :math:`2 - \lambda_i` where :math:`\lambda_i` are the eigenvalues of :math:`A`, positive definiteness of :math:`C` is equivalent to: .. math:: \rho(A) < 2 where :math:`\rho(A)` denotes the spectral radius (largest eigenvalue) of :math:`A`. We now classify all connected simple graphs satisfying this condition. Step 1: The graph must be a tree --------------------------------- **Claim.** If :math:`G` contains a cycle, then :math:`\rho(A) \geq 2`. *Proof.* The cycle graph :math:`C_m` on :math:`m` vertices has eigenvalues :math:`2\cos(2\pi k/m)` for :math:`k = 0, 1, \ldots, m-1`. The largest eigenvalue is :math:`2\cos(0) = 2`. If :math:`G` contains :math:`C_m` as a subgraph, then by the interlacing theorem for graph spectra, :math:`\rho(A_G) \geq \rho(A_{C_m}) = 2`. Therefore :math:`G` must be acyclic. Since :math:`G` is connected, it is a tree. :math:`\square` Step 2: Maximum degree at most 3 ---------------------------------- **Claim.** Every vertex of :math:`G` has degree at most 3. *Proof.* The star graph :math:`K_{1,m}` (one center connected to :math:`m` leaves) has adjacency eigenvalues :math:`\pm\sqrt{m}` and :math:`0` with multiplicity :math:`m-1`. Thus :math:`\rho(K_{1,m}) = \sqrt{m}`. For :math:`m \geq 4`, we have :math:`\sqrt{m} \geq 2`. If :math:`G` contains a vertex of degree :math:`\geq 4`, then :math:`G` contains :math:`K_{1,4}` as a subgraph, and by interlacing :math:`\rho(A_G) \geq 2`. :math:`\square` Step 3: At most one branch point ---------------------------------- **Claim.** :math:`G` has at most one vertex of degree 3 (branch point). *Proof.* Suppose :math:`G` has two vertices :math:`u, v` of degree 3. Since :math:`G` is a tree, there is a unique path from :math:`u` to :math:`v`. Each of :math:`u` and :math:`v` has two additional arms branching off this path. This means :math:`G` contains the graph :math:`T` formed by two degree-3 vertices connected by a path with an extra leaf at each end: .. math:: \bullet - u - \cdots - v - \bullet with at least one additional edge at both :math:`u` and :math:`v`. The smallest such graph is: .. math:: \bullet - \bullet - \bullet - \bullet - \bullet - \bullet (a path of length 5, i.e. :math:`D_4` with an extra arm). One can verify directly (or by interlacing with the known :math:`\tilde{D}_4` extended Dynkin diagram) that two branch points force :math:`\rho(A) \geq 2`. :math:`\square` Step 4: No branch point gives A\ :sub:`n` ------------------------------------------- If :math:`G` is a tree with no vertex of degree 3, then every vertex has degree at most 2, so :math:`G` is a **path graph** :math:`A_n` on :math:`n` vertices. Its eigenvalues are :math:`2\cos\!\bigl(\frac{k\pi}{n+1}\bigr)` for :math:`k = 1, \ldots, n`, so: .. math:: \rho(A_{A_n}) = 2\cos\!\left(\frac{\pi}{n+1}\right) < 2 for all :math:`n \geq 1`. :math:`\checkmark` Step 5: The Diophantine inequality ------------------------------------ Suppose :math:`G` is a tree with exactly one branch point, having three arms of lengths :math:`p, q, r` edges (so :math:`p + q + r + 1` vertices total), with :math:`1 \leq p \leq q \leq r`. We denote this tree :math:`T_{p,q,r}`. **Theorem.** :math:`\rho(T_{p,q,r}) < 2` if and only if .. math:: :label: diophantine \frac{1}{p+1} + \frac{1}{q+1} + \frac{1}{r+1} > 1 *Proof.* The argument proceeds in three parts: we establish the characteristic polynomial recurrence for paths, factor the characteristic polynomial of :math:`T_{p,q,r}` at its branch point, then evaluate at :math:`\lambda = 2`. **Part A: Path polynomials are Chebyshev polynomials.** Let :math:`f_\ell(\lambda) = \det(\lambda I - A_{P_\ell})` be the characteristic polynomial of the path graph on :math:`\ell` vertices. By cofactor expansion along the first (or last) row: .. math:: f_0(\lambda) = 1, \quad f_1(\lambda) = \lambda, \quad f_\ell(\lambda) = \lambda \, f_{\ell-1}(\lambda) - f_{\ell-2}(\lambda) This is exactly the recurrence for the Chebyshev polynomials of the second kind :math:`U_\ell` under the substitution :math:`\lambda = 2\cos\theta`: .. math:: f_\ell(2\cos\theta) = U_\ell(\cos\theta) = \frac{\sin\bigl((\ell+1)\theta\bigr)}{\sin\theta} In particular, evaluating at :math:`\lambda = 2` (i.e. :math:`\theta \to 0`): .. math:: f_\ell(2) = \lim_{\theta \to 0} \frac{\sin\bigl((\ell+1)\theta\bigr)}{\sin\theta} = \ell + 1 **Part B: Factoring the characteristic polynomial of** :math:`T_{p,q,r}`. The tree :math:`T_{p,q,r}` has a central vertex :math:`v` of degree 3, connected to three arms which are paths of lengths :math:`p`, :math:`q`, :math:`r`. Expanding :math:`\det(\lambda I - A)` along the row and column of :math:`v`, the three arms decouple: .. math:: P_{T_{p,q,r}}(\lambda) = \lambda \cdot f_p \cdot f_q \cdot f_r \;-\; f_{p-1} \cdot f_q \cdot f_r \;-\; f_p \cdot f_{q-1} \cdot f_r \;-\; f_p \cdot f_q \cdot f_{r-1} where all polynomials are evaluated at :math:`\lambda`. The three subtracted terms correspond to the three edges from :math:`v` into each arm. Dividing through by :math:`f_p \cdot f_q \cdot f_r`: .. math:: \frac{P_{T_{p,q,r}}(\lambda)}{f_p \cdot f_q \cdot f_r} = \lambda - \frac{f_{p-1}}{f_p} - \frac{f_{q-1}}{f_q} - \frac{f_{r-1}}{f_r} So :math:`\lambda` is an eigenvalue of :math:`T_{p,q,r}` if and only if: .. math:: :label: secular \frac{f_{p-1}(\lambda)}{f_p(\lambda)} + \frac{f_{q-1}(\lambda)}{f_q(\lambda)} + \frac{f_{r-1}(\lambda)}{f_r(\lambda)} = \lambda **Part C: Evaluating at** :math:`\lambda = 2`. Using :math:`f_\ell(2) = \ell + 1` from Part A, the secular equation :eq:`secular` at :math:`\lambda = 2` becomes: .. math:: \frac{p}{p+1} + \frac{q}{q+1} + \frac{r}{r+1} = 2 Rewriting each fraction as :math:`1 - \frac{1}{\ell+1}`: .. math:: 3 - \left(\frac{1}{p+1} + \frac{1}{q+1} + \frac{1}{r+1}\right) = 2 which gives: .. math:: \frac{1}{p+1} + \frac{1}{q+1} + \frac{1}{r+1} = 1 Therefore **λ = 2 is an eigenvalue if and only if the sum of reciprocals equals 1**. Since the path polynomials :math:`f_\ell` are positive and increasing for :math:`\lambda > 2\cos(\pi/(\ell+1))`, the spectral radius is strictly less than 2 precisely when :math:`P_{T_{p,q,r}}(2) > 0`, which by the factored form above holds if and only if: .. math:: \frac{1}{p+1} + \frac{1}{q+1} + \frac{1}{r+1} > 1 :math:`\square` Step 6: Enumerating solutions ------------------------------ We now enumerate all integer solutions of :eq:`diophantine` with :math:`1 \leq p \leq q \leq r`. **Case** :math:`p = 1`: .. math:: \frac{1}{2} + \frac{1}{q+1} + \frac{1}{r+1} > 1 \quad\Longleftrightarrow\quad \frac{1}{q+1} + \frac{1}{r+1} > \frac{1}{2} - :math:`q = 1`: :math:`\frac{1}{2} + \frac{1}{r+1} > \frac{1}{2}` holds for all :math:`r \geq 1`. This gives :math:`T_{1,1,r} = D_{r+3}` for :math:`r \geq 1`, i.e. :math:`D_n` for :math:`n \geq 4`. :math:`\checkmark` - :math:`q = 2`: :math:`\frac{1}{3} + \frac{1}{r+1} > \frac{1}{2}`, so :math:`r+1 < 6`, i.e. :math:`r \in \{2, 3, 4\}`: - :math:`T_{1,2,2}`: :math:`\frac{1}{2} + \frac{1}{3} + \frac{1}{3} = \frac{7}{6} > 1` → **E**\ :sub:`6` :math:`\checkmark` - :math:`T_{1,2,3}`: :math:`\frac{1}{2} + \frac{1}{3} + \frac{1}{4} = \frac{13}{12} > 1` → **E**\ :sub:`7` :math:`\checkmark` - :math:`T_{1,2,4}`: :math:`\frac{1}{2} + \frac{1}{3} + \frac{1}{5} = \frac{31}{30} > 1` → **E**\ :sub:`8` :math:`\checkmark` - :math:`T_{1,2,5}`: :math:`\frac{1}{2} + \frac{1}{3} + \frac{1}{6} = 1` → **fails** (affine :math:`\tilde{E}_8`) - :math:`q \geq 3`: :math:`\frac{1}{q+1} + \frac{1}{r+1} \leq \frac{1}{4} + \frac{1}{4} = \frac{1}{2}`, which does not satisfy :math:`> \frac{1}{2}`. **No solutions.** **Case** :math:`p = 2`: .. math:: \frac{1}{3} + \frac{1}{q+1} + \frac{1}{r+1} > 1 \quad\Longleftrightarrow\quad \frac{1}{q+1} + \frac{1}{r+1} > \frac{2}{3} Since :math:`q \geq p = 2`, we have :math:`\frac{1}{q+1} \leq \frac{1}{3}`, so :math:`\frac{1}{r+1} > \frac{1}{3}`, requiring :math:`r < 2`. But :math:`r \geq q \geq 2`, a contradiction. **No solutions.** **Case** :math:`p \geq 3`: Then :math:`\frac{1}{p+1} + \frac{1}{q+1} + \frac{1}{r+1} \leq 3 \cdot \frac{1}{4} < 1`. **No solutions.** The complete classification ---------------------------- Combining all steps, the connected simple graphs with :math:`\rho(A) < 2` are exactly: .. list-table:: :header-rows: 1 :widths: 15 25 30 * - Diagram - Tree structure - Diophantine data * - :math:`A_n` (:math:`n \geq 1`) - Path on :math:`n` vertices - (no branch point) * - :math:`D_n` (:math:`n \geq 4`) - :math:`T_{1, 1, n-3}` - :math:`\frac{1}{2} + \frac{1}{2} + \frac{1}{n-2} > 1` * - :math:`E_6` - :math:`T_{1, 2, 2}` - :math:`\frac{1}{2} + \frac{1}{3} + \frac{1}{3} > 1` * - :math:`E_7` - :math:`T_{1, 2, 3}` - :math:`\frac{1}{2} + \frac{1}{3} + \frac{1}{4} > 1` * - :math:`E_8` - :math:`T_{1, 2, 4}` - :math:`\frac{1}{2} + \frac{1}{3} + \frac{1}{5} > 1` No other connected simple graphs produce a finite root system under the mutation game. :math:`\blacksquare` Connections ----------- The inequality :math:`\frac{1}{p+1} + \frac{1}{q+1} + \frac{1}{r+1} > 1` is the same constraint that classifies: - **Platonic solids** — via Euler's formula :math:`V - E + F = 2` applied to regular polyhedra with face valence :math:`p+1`, vertex valence :math:`q+1` - **Finite subgroups of SO(3)** — cyclic, dihedral, tetrahedral (E₆), octahedral (E₇), icosahedral (E₈) - **Finite subgroups of SU(2)** — the binary polyhedral groups (McKay correspondence) - **Spherical triangle groups** — triangles with angles :math:`\frac{\pi}{p+1}, \frac{\pi}{q+1}, \frac{\pi}{r+1}` on the sphere (angle sum > π) - **Simple singularities** — Arnold's classification of ADE singularities (the low-codimension cases are Thom's seven catastrophes) This ubiquity was highlighted by V. I. Arnold [Arnold1976]_ and is one of the most striking phenomena in mathematics. References ^^^^^^^^^^ .. [Arnold1976] V. I. Arnold, "Problems in present day mathematics," in *Mathematical Developments Arising from Hilbert Problems*, Proceedings of Symposia in Pure Mathematics, vol. 28, AMS, 1976.