1. Central Hypothesis: Balance in Inequality
In a $4\times 4$ magic square, the numbers $1$ through $16$ are arranged so that every row, column, and main diagonal adds up to the magic sum $\mu_4 = 34$. To see the hidden geometry inside these numbers, we place the $4\times 4$ grid flat on the $(x,y)$ plane centered at the origin $O = (0,0,0)$ and measure each number relative to the average value $\bar{s} = 8.5$:
High numbers $9\dots 16$ lift UP ($z_{ij} > 0$) and Low numbers $1\dots 8$ drop DOWN ($z_{ij} < 0$)
Each cell $(i,j)$ now defines a 3D vector from the origin $\mathbf{p}_{ij}(\lambda) = (x_j, y_i, \lambda z_{ij})$, where $\lambda > 0$ controls how tall we stretch the vertical heights. Wrapping a tight elastic skin (the 3D convex hull) around all 16 vector tips creates a faceted polyhedron we call a Magic Gem $\mathcal{G}_\lambda(S)$.
★ Gold border: trapped inside the hull ($P_S^{(1)}$)
2. Test 1 (Radial Spokes): Why Equalizing Spoke Lengths Breaks Balance & How Duality Restores It
Because every row and column of a magic square adds up to the same constant, the 16 natural 3D arrows $\mathbf{p}_{ij}(\lambda)$ radiating from the origin $O$ always cancel out completely: $\sum_{i,j}\mathbf{p}_{ij}(\lambda) = \mathbf{0}$. Likewise, in the flat 2D grid plane, weighting each cell's position vector $\mathbf{r}_{ij} = (x_j, y_i)$ by its signed height $z_{ij}$ gives $\sum_{i,j} z_{ij}\mathbf{r}_{ij} = \mathbf{0}$. If you place those 16 arrows head-to-tail in order of their compass angle around $O$, they form a perfectly closed 16-sided polygon.
But what happens if we change the lengths of the arrows? Suppose every cell pulls on the origin along its unit direction with length $L_{ij} = R_{ij}^\beta$ (in 3D) or $\ell(r_{ij}) = r_{ij}^\alpha$ (in 2D)—most simply, what if all 16 arrows are rescaled to unit length ($\beta=0$ on the unit sphere $S^2$, or $\alpha=0$ on the unit circle $S^1$)?
- Length–Weight Duality: Changing arrow lengths to $R_{ij}^\beta$ is mechanically identical to keeping the original arrows $\mathbf{p}_{ij}$ and reweighting the masses at their tips by $w_{ij} = R_{ij}^{\beta-1}$. At cubic 3D lengths ($\beta=3$), the vertical pull $V_z$ vanishes identically ($V_z = 0$) for every magic square. And on the 3D unit sphere ($\beta=0$), all $880$ squares can be balanced at $O$ using strictly positive tip weights $w_{ij}^* > 0$.
- The $\sqrt{5}$ Irrationality Barrier in 2D: A $4\times 4$ board consists of three concentric rings around $O$: 4 Inner cells ($r_1 = \frac{\sqrt{2}}{2}$), 8 Middle cells ($r_2 = \frac{\sqrt{10}}{2}$), and 4 Outer corners ($r_3 = \frac{3\sqrt{2}}{2}$). Because the middle ring is $\frac{r_2}{r_1} = \sqrt{5}$ times farther out than the inner ring—and $\sqrt{5} \approx 2.23607$ is irrational—16 unit arrows ($\alpha=0$) can never cancel out on any $4\times 4$ magic square. The closest any magic square can get is the famous Pell fraction $\frac{9}{4} = 2.25 \approx \sqrt{5}$ ($9^2 - 5\cdot 4^2 = 1$), achieved by 8 squares led by #70 with a tiny gap of $\|\mathbf{B}_{\mathrm{unit}}\|_2 \approx 0.07049$.
- Exact Non-Linear Polygon Closure at $\alpha^* = -1$ and $\alpha^* = +3$: Whenever a square's inner and outer ring dipoles point in the same direction ($204$ squares), there is a second exponent $\alpha^* \ne 1$ where the 16-gon snaps completely shut again—including exact inverse lengths $\ell(r)=1/r$ ($\alpha^*=-1$) on 25 squares (#45) and cubic lengths $\ell(r)=r^3$ ($\alpha^*=+3$) on 25 squares (#286).
Formal Statements: Length–Weight Duality, Three-Ring Formula & Pell Minimum (Proposition 2.2, Theorem 2.3, Corollary 2.4 & Theorem 2.5)
3D Length–Weight Duality & Cubic Vertical Vanishing (Proposition 2.2 & Theorem 2.5): Along unit rays $\hat{\mathbf{p}}_{ij}(\lambda) = \mathbf{p}_{ij}(\lambda)/R_{ij}(\lambda) \in S^2$, scaling lengths to $L_{ij}$ with weights $w_{ij}$ gives $\mathbf{V}_{w,L}(S;\lambda) = \sum_{i,j} w_{ij}L_{ij}\hat{\mathbf{p}}_{ij}(\lambda) = \sum_{i,j}\bigl(\frac{w_{ij}L_{ij}}{R_{ij}(\lambda)}\bigr)\mathbf{p}_{ij}(\lambda)$. For cubic 3D lengths ($\beta=3$, $L_{ij}=R_{ij}^3$), row–column separability of $r_{ij}^2 = x_j^2+y_i^2$ and $\sum z_{ij}^3 = 0$ force the vertical component to vanish identically for every semi-magic square: $$\mathbf{V}_3(S;\lambda) = \left(\lambda^2\sum_{i,j} z_{ij}^2 x_j,\;\; \lambda^2\sum_{i,j} z_{ij}^2 y_i,\;\; 0\right).$$ On the unit sphere $S^2$ ($\beta=0$), the minimum-variance balancing weight vector $\mathbf{w}^*(S) = \mathbf{1} - \hat{P}(\hat{P}^\top\hat{P})^{-1}\mathbf{V}_0(S)$ is strictly positive ($\mathbf{w}^*(S) > \mathbf{0}$) for all $880$ order-four magic squares.
Three-Ring Formula & Even-Lattice Pell Minimum (Theorem 2.3 & Corollary 2.4): Grouping $A_4 = R_{\mathrm{in}} \sqcup R_{\mathrm{mid}} \sqcup R_{\mathrm{out}}$ with ring dipoles $\mathbf{d}_k = \sum_{c\in R_k} z_c\mathbf{r}_c$ satisfies $\mathbf{d}_{\mathrm{in}} \in \mathbb{Z}^2\setminus\{\mathbf{0}\}$, $\mathbf{d}_{\mathrm{out}} \in (3\mathbb{Z})^2\setminus\{\mathbf{0}\}$, and $\mathbf{d}_{\mathrm{mid}} = -(\mathbf{d}_{\mathrm{in}}+\mathbf{d}_{\mathrm{out}})$, giving $$\mathbf{B}_{z,\ell}(S) = \left(\frac{\ell_1}{r_1} - \frac{\ell_2}{r_2}\right)\mathbf{d}_{\mathrm{in}}(S) + \left(\frac{\ell_3}{r_3} - \frac{\ell_2}{r_2}\right)\mathbf{d}_{\mathrm{out}}(S).$$ At unit length ($\ell_k=1$, $\alpha=0$), substituting $1/r_1=\sqrt{2}$, $1/r_2=\sqrt{2}/\sqrt{5}$, and $1/r_3=\sqrt{2}/3$ motivates $\mathbf{u}(S) = \mathbf{d}_{\mathrm{in}}(S) + \frac{1}{3}\mathbf{d}_{\mathrm{out}}(S) \in \mathbb{Z}^2$ and $\mathbf{v}(S) = \mathbf{d}_{\mathrm{in}}(S) + \mathbf{d}_{\mathrm{out}}(S) \in \mathbb{Z}^2$, so $\mathbf{B}_{\mathrm{unit}}(S) = \sqrt{2}\bigl(\mathbf{u}(S) - \frac{1}{\sqrt{5}}\mathbf{v}(S)\bigr)$. Because the four-cell diagonal halves $T_{\mathrm{diag}}$ and $R_{\mathrm{diag}}$ always have even sum across $\mathcal{M}_4$, both $\mathbf{u}(S) = 2\tilde{\mathbf{u}}(S)\in(2\mathbb{Z})^2$ and $\mathbf{v}(S) = 2\tilde{\mathbf{v}}(S)\in(2\mathbb{Z})^2$ lie in the even sublattice with $\tilde{\mathbf{u}}, \tilde{\mathbf{v}} \in \mathbb{Z}^2\setminus\{\mathbf{0}\}$, $\|\tilde{\mathbf{u}}(S)\|_\infty \le 8$, and $\|\tilde{\mathbf{u}}(S)\|_2 \le 4\sqrt{5} < 9$. Thus $\|\mathbf{B}_{\mathrm{unit}}(S)\|_2 = 2\sqrt{2}\|\tilde{\mathbf{u}}(S) - \tilde{\mathbf{v}}(S)/\sqrt{5}\|_2$ is minimized along a coordinate axis with $1 \le \tilde{u} \le 8 < 17$; by Lagrange's theorem on continued fractions for $\sqrt{5} = [2; \overline{4}]$ (convergents $2/1, 9/4, 38/17, \dots$), the unique best rational approximation in range is the fundamental Pell convergent $\tilde{v}/\tilde{u} = 9/4$ ($9^2 - 5\cdot 4^2 = 1$), giving $\|\mathbf{v}\|/\|\mathbf{u}\| = 18/8 = 9/4$ on $8$ squares (#70, #76, #121, #188, #452, #459, #490, #574) and achieving $$\min_{S\in\mathcal{M}_4}\|\mathbf{B}_{\mathrm{unit}}(S)\|_2 = \frac{18\sqrt{10}}{5} - 8\sqrt{2} = \frac{2\sqrt{10}}{5\varphi^6} \approx 0.07049108.$$
3. Test 2 (3D Rotation): Equal Masses Precess Smoothly, Unequal Entry Masses Flip $180^\circ$
If you spin a tennis racket, a book, or an irregular crystal in zero gravity around its intermediate axis of inertia ($I_1 < I_2 < I_3$), something famous and eerie happens: after a few turns, the object suddenly flips $180^\circ$ upside down, spins a few more turns, and flips back again. This is the Dzhanibekov (or tennis-racket) instability.
You might expect an irregular Magic Gem—whose 3D shape often has zero geometric symmetry—to tumble chaotically in space. Instead, the magic line-sum conditions produce a remarkable split between equal point masses and entry-weighted masses:
- Equal Masses at the 16 Vertices ($m_{ij} = 1$): Because the rows and columns balance, the horizontal moments of inertia are always equal ($I_1 = I_2 = I_\perp(\lambda)$) for every magic square. With no intermediate axis to go unstable, the gem never flips. Even better, at the magic vertical scale $\lambda_{\mathrm{iso}} = \frac{1}{\sqrt{17}} \approx 0.2425$, the vertical and horizontal moments become identical ($I_1 = I_2 = I_3$): every Magic Gem becomes a perfect spherical top that spins with zero wobble around any axis in 3D.
- Entry-Weighted Masses ($m_{ij} = s_{ij}$): If each cell instead weighs its number $s_{ij} \in \{1\dots 16\}$, the diagonal quadrupole moment $Q_{xy}(S) = \sum_{i,j} z_{ij}x_jy_i$ splits the two horizontal moments into $I_1 = 170 - |Q_{xy}|$ and $I_2 = 170 + |Q_{xy}|$. For $56$ squares (like Dürer #175), $Q_{xy}=0$ and spin stays smooth; for the other $824$ squares (led by #46 with $|Q_{xy}|=60$), spinning near the intermediate diagonal axis triggers dramatic $180^\circ$ Dzhanibekov flips.
-
How Magic Gems Compare to Classical Shapes (Sphere, Perfect Cube, Cone & Pyramid):
- Sphere & Perfect Cube ($I_1 = I_2 = I_3$): Counterintuitively, a uniform Perfect Cube is an exact spherical top ($I_{xx}=I_{yy}=I_{zz}=\frac{1}{6}Ma^2$), spinning with zero wobble around any axis just like a Sphere—and just like any equal-mass Magic Gem at $\lambda_{\mathrm{iso}} = 1/\sqrt{17} \approx 0.2425$.
- Right Circular Cone & Square Pyramid ($I_1 = I_2 \ne I_3$): Because of their rotational or $C_4$ symmetry, a Cone and a Square Pyramid are symmetric tops: their spin axis sweeps a smooth circular precession cone and never flips—matching an equal-mass Magic Gem at $\lambda \ne 1/\sqrt{17}$, even though the Magic Gem itself has no rotational symmetry.
Formal Statements: Universal Symmetric-Top Law & Quadrupole Split (Theorems 3.1 & 3.2)
Theorem 3.1 (Point-Mass Axisymmetric Rotor): For any arrangement with $\sum xz = \sum yz = 0$ carrying equal masses $M/n^2$ at $\mathbf{p}_{ij}(\lambda)$, the inertia tensor about $O$ is $I(\lambda) = M\operatorname{diag}(I_\perp(\lambda), I_\perp(\lambda), I_{zz})$ with universal ratio $$\frac{I_\perp(\lambda)}{I_{zz}} = \frac{1 + \lambda^2(n^2+1)}{2}.$$ At $n=4$, this ratio equals $21/25 = 0.84$ at the cubic 3D-print scale $\lambda_{\mathrm{cube}}=0.20$, equals $1$ at the spherical-top scale $\lambda_{\mathrm{iso}} = 1/\sqrt{17} \approx 0.2425$, and equals the integer $9$ at the canonical lift $\lambda=1$ (closing every free-precession orbit after $9$ spin revolutions).
Theorem 3.2 (Entry-Weighted Quadrupole Split & Central Inertia): Assigning masses $m_{ij} = s_{ij}$ on the planar grid gives principal moments $I_1(S) = 170 - |Q_{xy}(S)|$, $I_2(S) = 170 + |Q_{xy}(S)|$, and $I_3 = 340$, where $Q_{xy}(S) = \sum_{i,j} z_{ij}x_jy_i \in [-60, 60]$. For the $824$ squares with $Q_{xy}\ne 0$, intermediate-axis perturbations grow exponentially at dimensionless rate $\Gamma(S) = \sqrt{|Q_{xy}(S)|/170}$. In the 3D lift $\mathbf{p}_{ij}(\lambda)$, entry masses shift the center of mass vertically to $\mathbf{c}^{(s)}(\lambda) = (0, 0, \frac{b_n}{\bar{s}}\lambda) = (0, 0, \frac{n^2-1}{6}\lambda) = (0, 0, \frac{5}{2}\lambda)$; by Steiner's parallel-axis theorem, the central inertia tensor $I_{\mathrm{cm}}^{(s)}(\lambda) = I_O^{(s)}(\lambda) - M\|\mathbf{c}^{(s)}(\lambda)\|_2^2(\mathbf{I}_3 - \hat{\mathbf{e}}_z\hat{\mathbf{e}}_z^\top)$ retains the identical horizontal quadrupole split $I_{xy,\mathrm{cm}}^{(s)} = -Q_{xy}(S)$, vertical coupling $(I_{xz,\mathrm{cm}}^{(s)}, I_{yz,\mathrm{cm}}^{(s)}) = -\lambda^{-1}((\mathbf{V}_3)_x, (\mathbf{V}_3)_y)$, and central diagonal moments $I_{xx,\mathrm{cm}}^{(s)} = I_{yy,\mathrm{cm}}^{(s)} = Ma_n(1 + \frac{2(n^2+2)}{3}\lambda^2) = 170(1 + 12\lambda^2) = 340(\frac{1}{2} + 6\lambda^2)$ with $I_{zz,\mathrm{cm}}^{(s)} = 340$.
4. Test 3A (Convex Solids): Restoring Balance via "Diamond-in-a-Diamond" Core Counterweights
Suppose you 3D-print a Magic Gem $\mathcal{G}_0(S)$ out of uniform plastic and place it on a knife-edge fulcrum at the origin $O = (0,0,0)$. Even though the 16 discrete points have their average at $O$, $829$ of the $880$ solid Magic Gems tip over.
Why? Because the outer surface of the crystal only touches the outer corner vertices ($V$ points) and completely forgets the $16-V$ inner points trapped inside. If we wrap a second convex hull around those forgotten inner points, we get the Inner Core Gem $\mathcal{G}_1(S) \subset \mathcal{G}_0(S)$—a literal "diamond inside a diamond."
- 51 Gems Balance Naturally as Uniform Solids ($\mathbf{c}_0 = \mathbf{0}$): All $48$ centrally symmetric associative gems plus three remarkable non-associative gems in Group VI (#86, #739, #756).
- 33 Gems Balance via a Denser Inner Core ($\Delta\rho_1 > 0$): In Group VI, a $180^\circ$ rotation symmetry locks the outer solid centroid $\mathbf{c}_0$ and the inner core centroid $\mathbf{c}_1$ onto the same axis pointing in opposite directions. Printing the inner core slightly denser—starting at $\Delta\rho_1/\rho_0 = 5/4 = 1.25$ on Gem #673—brings the composite center of mass to $0.00^\circ$ level balance.
- 2 Gems Balance via a Lighter or Hollow Core Cavity ($\Delta\rho_1 < 0$): On 17 gems, volumetric simplex weighting pulls $\mathbf{c}_0$ to the same side of $O$ as $\mathbf{c}_1$. Two of these—#123 ($\Delta\rho_1/\rho_0 = -3/17$) and #54 ($\Delta\rho_1/\rho_0 = -57/58$, a nearly hollow cavity)—balance by printing the inner core lighter than the shell.
Formal Statements: Onion Peeling & Two-Density Balancing Formula (Propositions 4.1–4.2 & Theorem 4.3)
Composite Center of Mass (Theorem 4.3): Fabricating $\mathcal{G}_0(S)$ with baseline density $\rho_0 > 0$ and inner core $\mathcal{G}_1(S) = \operatorname{conv}(P_S\setminus\operatorname{vert}\mathcal{G}_0(S))$ with excess density $\Delta\rho_1 = \rho_1 - \rho_0$ yields composite centroid $$\mathbf{c}_{\mathrm{comp}} = \frac{\rho_0\operatorname{vol}(\mathcal{G}_0)\,\mathbf{c}_0 + \Delta\rho_1\operatorname{vol}(\mathcal{G}_1)\,\mathbf{c}_1}{\rho_0\operatorname{vol}(\mathcal{G}_0) + \Delta\rho_1\operatorname{vol}(\mathcal{G}_1)}.$$ Whenever $\mathbf{c}_0 \times \mathbf{c}_1 = \mathbf{0}$ and $\mathbf{c}_1 \ne \mathbf{0}$, exact balance $\mathbf{c}_{\mathrm{comp}} = \mathbf{0}$ is achieved at the unique rational ratio $$\frac{\Delta\rho_1}{\rho_0} = -\frac{\operatorname{vol}(\mathcal{G}_0)\,(\mathbf{c}_0\cdot\mathbf{c}_1)}{\operatorname{vol}(\mathcal{G}_1)\,\|\mathbf{c}_1\|_2^2}.$$ Across all $880$ gems, $528$ ($60.0\%$) also balance via non-negative point weights placed at the inner points $P_S^{(1)}$ ($-\mathbf{c}_0 \in \operatorname{cone}(P_S^{(1)})$).
5. Test 3B (Skeletons & Shells): Wireframes, Complete $K_{16}$ Networks & Papercraft Nets
A solid 3D print is only one way to build a Magic Gem. You can also build one from balls and rods, from a tangle of struts joining every pair of points, or from a sheet of folded paper. Each of these physical gems spreads its mass differently, so each one balances differently.
- Ball-and-rod wireframe Balls sit at the 16 points, with rods along the edges of the outer hull and the inner core. Rods weigh in proportion to their length, so the frame usually tips over. Making the inner rods thicker or thinner (rod-diameter ratio $d_1/d_0$ in the slender-rod limit $d/L_{uv}\to 0^+$) rebalances 158 Group VI gems at $\lambda = 1$ ($164$ at the cubic scale $\lambda = 0.20$), for example #673 at $d_1/d_0 \approx 0.793$ ($\lambda=1$) or $d_1/d_0 \approx 0.493$ ($\lambda=0.20$). That is more than four times as many gems as two-density solids can balance.
- Complete $K_{16}$ strut network Join all $\binom{16}{2} = 120$ pairs of points. If every rod has the same mass, the network balances at $O$ for every magic square and spins like the 16 point masses, with the same $1/\sqrt{17}$ spherical-top scale. With constant-thickness rods (mass proportional to length) it only balances for the $48$ associative squares, but it drifts $2.7$ to $3.6$ times less than the outer wireframe.
- Papercraft net Unfold the gem's surface into a flat, printable net. However you unfold it, you need one glue tab fewer than the number of corners ($N_{\mathrm{tabs}} = f_0 - 1$): 11 tabs for Dürer's #175 and 7 for #673. All $880$ gems unfold without any overlapping faces.
Formal Statements: Complete KN Networks, Nested Wireframes & Papercraft Nets (Theorem 4.4, Proposition 4.5 & Remark 4.6)
Theorem 4.4 (Complete $K_N$ Networks & Nested Wireframes): Let $K_N(P_S(\lambda))$ be the complete graph on the $N=n^2$ lifted points with edge midpoints $\mathbf{m}_{uv} = \frac{1}{2}(\mathbf{p}_u+\mathbf{p}_v)$ and lengths $L_{uv}(\lambda) = \|\mathbf{p}_u-\mathbf{p}_v\|_2$.
- Equal-Mass $K_N$ Universal Rotor: Because every vertex has degree $N-1$, assigning equal mass $m_e = M_{\mathrm{rods}}/\binom{N}{2}$ to each rod gives $\mathbf{c}_{=\text{mass}}(K_N) = \frac{1}{N}\sum_u\mathbf{p}_u = \mathbf{0}$ and continuous rod-network covariance $\Sigma_{\mathrm{rods}}^{(=\text{mass})}(K_N) = \frac{2N-3}{3(N-1)}\Sigma(\lambda)$ ($=\frac{29}{45}\Sigma(\lambda)$ at $n=4$). For any node-to-strut mass fraction $\theta\in[0,1]$, $K_N$ shares the universal axisymmetric ratio $I_\perp(\lambda)/I_{zz} = \frac{1}{2}(1+\lambda^2(n^2+1))$ and spherical-top scale $\lambda_{\mathrm{iso}} = (n^2+1)^{-1/2}$.
- Length-Weighted $K_{16}$ & Nested Wireframe Gauge Ratio (1D Slender-Rod Limit $d/L_{uv}\to 0^+$): Under uniform rod cross-section ($m_e = \rho_{\mathrm{lin}}L_e$), $\mathbf{c}_{\mathrm{len}}(K_{16}) = \mathbf{0} \iff S$ is associative ($48/880$), while across all $880$ gems $K_{16}$ reduces mean centroid drift by $2.7\times$–$3.6\times$ relative to $G(\mathcal{G}_0)$. Assigning linear densities $\rho_{\mathrm{lin},0}$ and $\rho_{\mathrm{lin},1}$ to the disjoint edge sets $E(\mathcal{G}_0)$ and $E(\mathcal{G}_1)$ with length moments $\mathbf{m}_{\mathrm{wire},k}(\lambda) = \sum_{uv\in E(\mathcal{G}_k)} L_{uv}(\lambda)\mathbf{m}_{uv}(\lambda)$ balances the composite wireframe whenever $\mathbf{m}_{\mathrm{wire},0}\times\mathbf{m}_{\mathrm{wire},1}=\mathbf{0}$ and $\mathbf{m}_{\mathrm{wire},0}\cdot\mathbf{m}_{\mathrm{wire},1} < 0$ at rod-diameter ratio $$\frac{d_1}{d_0} = \sqrt{\frac{\rho_{\mathrm{lin},1}}{\rho_{\mathrm{lin},0}}} = \sqrt{-\frac{\mathbf{m}_{\mathrm{wire},0}(\lambda)\cdot\mathbf{m}_{\mathrm{wire},1}(\lambda)}{\|\mathbf{m}_{\mathrm{wire},1}(\lambda)\|_2^2}},$$ balancing $158$ Group VI wireframes at $\lambda=1$ ($81$ with 3D cores, $37$ with 2D cores, $40$ with 1D cores; e.g. #673 at $d_1/d_0 \approx 0.793$) and $164$ at $\lambda=0.20$ (#673 at $d_1/d_0 \approx 0.493$).
- Remark 4.6 (Finite-Diameter Node/Strut Overlap Corrections): Physical 3D prints with finite rod diameter $d>0$ and node spheres/cubes of radius $r_{\mathrm{node}}>0$ experience pairwise cylinder-overlap and cap-truncation volume corrections of order $O((d/L_{\min})^2)$ and $O((r_{\mathrm{node}}/L_{\min})^3)$. Because $D_4$ and $C_2$ symmetries preserve all pairwise intersection volumes, the $48$ associative gems stay balanced for any finite $d, r_{\mathrm{node}}>0$, and the Group VI $C_2$ axis continues to constrain both wireframe moments to the same 1D axis.
- Proposition 4.5 (Papercraft Cutout Nets & $N_{\mathrm{tabs}} = f_0 - 1$): Cutting $\partial\mathcal{G}_\lambda(S)$ along the edges complement to a dual spanning tree $T^*\subset G^*$ of its $f_2$ facets leaves $e(T^*) = f_2-1$ interior fold hinges; by LP duality the primal cut edges form a spanning tree $T\subset G^{(1)}$ on the $f_0$ vertices, requiring $N_{\mathrm{tabs}}(S) = f_1 - (f_2 - 1) = f_0 - 1 \in \{7,\dots,15\}$ glue tabs (invariant under facial triangulation). All $880$ gems admit a strictly non-overlapping BFS spanning-tree unfolding at both $\lambda=1$ and $\lambda=0.20$, and $80$ Group VI nested thin shells $\partial\mathcal{G}_0\cup\partial\mathcal{G}_1$ ($\lambda=1$; $87$ at $\lambda=0.20$) balance in the zero-thickness limit $t\to 0^+$ via cardstock areal-density ratio $\sigma_1/\sigma_0 > 0$.
6. Test 4 (Freestanding Gravity): Why Identical Blocks Wedge Apart & Mirror Twins Lock Straight
Can we stack copies of a Magic Gem on top of each other to build a free-standing vertical tower under gravity? If you try to stack identical copies of a single gem $\mathcal{G}_\lambda(S)$ without rotating them, the top facet $F_+$ and bottom facet $F_-$ must have parallel slopes—a rare property shared by only $124$ of the $880$ gems. On most gems, stacking identical copies makes the column veer crookedly off to the side and topple.
Paper C solves this with a simple symmetry trick: take the complement magic square $\overline{S} = 17 - S$ and rotate the board $180^\circ$ ($r^2$). The resulting square $S^* = r^2\overline{S}$ is the Mirror Twin, and its 3D gem is the exact central inversion of the original gem: $$\mathcal{G}_\lambda(r^2\overline{S}) = -\mathcal{G}_\lambda(S).$$
- Automatic Slope & Shape Matching: Central inversion turns every upper facet $F_+$ (pointing up with normal $+\mathbf{n}_+$) into a congruent lower facet $-F_+$ on the mirror twin (pointing down with normal $-\mathbf{n}_+$). Whenever $F_+$ and $F_-$ are centrally symmetric polygons centered on the vertical $z$-axis ($48$ gems), alternating Primary Gem $\mathcal{G}_\lambda(S)$ and Mirror Twin $-\mathcal{G}_\lambda(S)$ builds a strictly vertical tower with $100\%$ polygon lock and zero center-of-mass drift (with the bottom facet $F_-$ seated on a complementary wedge plinth of slope $-\mathbf{n}_-$ when $F_-$ is oblique to the horizontal table).
- The Knight's-Move Square Champion (#763): Seventeen gems achieve the largest possible horizontal contact footprint—an area of $A_{xy} = 5$ ($55.6\%$ of the entire $3\times 3$ grid) formed by a tilted $2\times 1$ knight's-move square connecting $\{13,14,15,16\}$ on top and $\{1,2,3,4\}$ on the bottom. At the cubic 3D-print scale $\lambda = 0.20$, $74.5\%$ of all $14{,}824$ facets across the $880$ gems are gravitationally stable; a single block of #763 resting on its contact facet stands firm up to a table tilt of $\theta_{\mathrm{tip}}^{(1)} = 39.35^\circ$, while unbonded dry-stack joints of slope $\alpha = \arctan(\lambda)$ remain slip-free whenever static friction satisfies $\mu_s \ge \lambda = 0.20$ ($\theta_{\mathrm{joint}} = 11.31^\circ$).
Formal Statements: Facet Stability Margin & Mirror-Twin Tower Theorem (Proposition 5.2, Theorem 5.3 & Proposition 5.4)
Resting Stability (Definition 5.1 & Proposition 5.2): A facet $F \subset \partial\mathcal{G}_\lambda(S)$ is statically stable under gravity iff the orthogonal projection $\pi_F(\mathbf{c}(\lambda))$ lies in the strict interior of $F$ ($d_{\mathrm{stab}} > 0$), giving single-block critical tipping angle $\theta_{\mathrm{tip}}^{(1)} = \arctan(d_{\mathrm{stab}}/h_F(\lambda))$ toward the nearest edge of $F$. Rescaling from $\lambda=1$ to $\lambda_{\mathrm{cube}}=0.20$ increases stable facets across $\mathcal{M}_4$ from $6{,}378/14{,}824$ ($43.0\%$) to $11{,}044/14{,}824$ ($74.5\%$).
Mirror-Twin Alternating Towers & Base Plinths (Theorem 5.3 & Proposition 5.4): Given centrally symmetric facets $F_+, F_- \subset \mathcal{G}_\lambda(S)$ with centroids $\mathbf{c}(F_\pm) = (0,0,z_\pm)$, the unrotated blocks $B_{2m} = \mathcal{G}_\lambda(S) + m\mathbf{T}$ and $B_{2m+1} = -\mathcal{G}_\lambda(S) + 2\mathbf{c}(F_+) + m\mathbf{T}$ with vertical period $\mathbf{T} = (0,0,2(z_+-z_-))$ meet with $100\%$ polygon congruence at every interior joint and keep the tower's center of mass on the vertical $z$-axis (seated on a complementary base wedge plinth of upper normal $-\mathbf{n}_-(\lambda)$ when $\mathbf{n}_-\nparallel\hat{\mathbf{e}}_z$). Across an unbonded $M$-block column, each oblique joint of tilt $\alpha_\pm(\lambda) = \arccos(|\mathbf{n}_\pm(\lambda)\cdot\hat{\mathbf{e}}_z|)$ prevents sliding whenever $\mu_s \ge \tan\alpha_\pm(\lambda)$ ($\mu_s \ge \lambda = 0.20$ on the $17$ knight's-move towers), while monolithically bonded or pinned columns tip as a rigid body when the composite centroid projection exits the base footprint $F_-$.
7. Synthesis: Three Tiers of Balance in Inequality Across All 880 Magic Gems
Every theorem and census count in Paper C is verified in exact rational arithmetic across all $880$ order-four magic squares by the companion verification suite. Click any Load in Lab ↗ badge below to jump directly to that simulator with the corresponding canonical square pre-loaded.
Complete Reference Table (Table 4 of Paper C)
| Mechanical Property / Regime | Order-4 Census | Universal Law or Extremal Bound | Canonical Exemplars |
|---|---|---|---|
| Linear 3D & 2D Vectors ($\beta=1, \alpha=1$) | $880 / 880$ ($100\%$) | $\sum\mathbf{p}_{ij}(\lambda)=\mathbf{0}$ and $\mathbf{B}_{z,r}=\mathbf{0}$ identically | All semi-magic squares |
| Cubic 3D Vector Law ($\beta=3$, $L_{ij}=R_{ij}^3$) | $880 / 880$ ($100\%$) | $(\mathbf{V}_3(S;\lambda))_z = 0$ identically; $\mathbf{V}_3=\mathbf{0} \iff$ associative ($48$) | All semi-magic ($z=0$) |
| 3D Unit Sphere $\sum \hat{\mathbf{p}}_{ij}(\lambda) = \mathbf{0}$ | $48 / 880$ ($5.5\%$) | Exact $\mathbf{0}$ $\forall\lambda>0 \iff$ associative; $107$ have $\beta^*\ne 1$ | #175 ($\mathbf{0}$), #70 ($\beta^*\approx 2.24$), #376 |
| 3D Sphere Positive Weights $\mathbf{w}^*(S) > \mathbf{0}$ | $880 / 880$ ($100\%$) | $\sum w_{ij}^*\hat{\mathbf{p}}_{ij}=\mathbf{0}$ with $\|\mathbf{w}^*-\mathbf{1}\|_2 \in [0.005, 1.60]$ | All $880$ magic squares |
| Planar Unit-Vector Norm $\|\mathbf{B}_{\mathrm{unit}}(S)\|_2$ | $0 / 880$ vanish | $\min = \frac{18\sqrt{10}}{5} - 8\sqrt{2} \approx 0.070491$ ($v/u = 9/4$) | #70, #76, #121, #188, #452, #459, #490, #574 |
| Collinear Dipoles & Roots $\alpha^*(\kappa)\ne 1$ | $408 / 880$ ($46.4\%$) | $204$ with $\kappa>0$ ($25$ at $\alpha^*=-1$, $25$ at $\alpha^*=+3$) | #45 ($\alpha^*=-1$), #286 ($\alpha^*=+3$) |
| Point-Mass Rotor Ratio $I_\perp(\lambda)/I_{zz}$ | $880 / 880$ ($100\%$) | $\frac{1+17\lambda^2}{2}$ ($1$ at $\lambda=\frac{1}{\sqrt{17}}$, $9$ at $\lambda=1$) | Every semi-magic square |
| Entry-Weighted ($m=s$) Rotor | $56 / 880$ ($6.4\%$) | $Q_{xy}=0 \implies I_1=I_2=170$; $824$ flip ($|Q_{xy}|\le 60$) | #175 ($Q_{xy}=0$), #46 ($|Q_{xy}|=60$) |
| Naturally Balanced Solid $\mathbf{c}_0 = \mathbf{0}$ | $51 / 880$ ($5.8\%$) | $48$ associative + $3$ non-associative Group VI | #86, #739, #756 |
| Two-Density "Diamond-in-a-Diamond" | $35 / 511$ 3D cores | $33$ denser ($\Delta\rho_1/\rho_0 \ge 5/4$) + $2$ lighter/cavity | #673 ($5/4$), #123 ($-3/17$), #54 ($-57/58$) |
| Complete $K_{16}$ Network ($120$ rods) | $880 / 880$ (equal mass) | $\mathbf{c}=\mathbf{0}$, $\Sigma_{\mathrm{rods}}=\frac{29}{45}\Sigma(\lambda)$; length-weighted mean drift $0.025$ ($\lambda=0.2$) | Dürer #175 ($K_{16}$ strut model) |
| Nested Wireframes $G(\mathcal{G}_0)\cup G(\mathcal{G}_1)$ | $206 / 880$ ($\lambda=1$) | $48$ associative + $158$ Group VI ($164$ at $\lambda=0.20$) via $d_1/d_0$ | #673 ($d_1/d_0=0.793$ at $\lambda=1$; $0.493$ at $\lambda=0.20$) |
| Papercraft Nets & Shells $\partial\mathcal{G}_0$ | $880 / 880$ nets | $N_{\mathrm{tabs}}=f_0-1\in[7,15]$; $48+80$ Group VI shells balance | #175 ($11$ tabs, $13$ folds) |
| Resting Facet Stability ($\lambda=0.20$) | $11{,}044 / 14{,}824$ | $74.5\%$ stable at $\lambda=0.20$ ($43.0\%$ at $\lambda=1.0$) | Max $\theta_{\mathrm{tip}}^{(1)} = 47.3^\circ$ at $\lambda=0.20$ |
| Mirror-Twin Vertical Towers | $48 / 880$ ($5.5\%$) | Max projected footprint $A_{xy}=5$ ($17$ knight's-move) | #763, #603, #774 |
Open Questions for Future Exploration
1. Higher-Order Diophantine Ring Minima ($n \ge 5$)
At $n=5$, the grid rings have radii $\{1, \sqrt{2}, 2, \sqrt{5}, 2\sqrt{2}\}$, mixing $\sqrt{2}$ and $\sqrt{5}$. Does $\mathbf{B}_{\mathrm{unit}}(S)$ ever vanish for $n\ge 5$, or how fast does $\min_{S\in\mathcal{M}_n}\|\mathbf{B}_{\mathrm{unit}}(S)\|_2$ decay via simultaneous Diophantine approximation?
2. Onion Peeling Depth of Siamese Gems
By Paper A, the outer hull of an odd-order Siamese square strips off only $12$ vertices, leaving $n^2-12$ inner points on five sheared polygons. How does the onion peeling depth $d_{\mathrm{onion}}(S_n)$ grow with $n$?
3. Exact Solid-Isotropic Magic Gems
Point-weighted gems become exact spherical tops at $\lambda_{\mathrm{iso}}=(n^2+1)^{-1/2}$, whereas uniform solids $\mathcal{G}_\lambda(S)$ retain a tiny $0.004\%\text{–}2.4\%$ transverse split. Can a two-density composite $(\mathcal{G}_0,\mathcal{G}_1)$ achieve an exact scalar-identity solid inertia tensor?
4. Space-Filling & Interlocking 3D Packings
Beyond vertical mirror-twin columns, can a Magic Gem $\mathcal{G}_\lambda(S)$ and its $D_4\times\langle\tau\rangle$ symmetry images tile a horizontal slab or interlock in 3D space with full facet-to-facet contact?