Chemistry had the same structure
Mendeleev's periodic table is one face of an 8-dimensional object. The full element tensor has axes for proton number, neutron number, orbital quantum numbers (n, ℓ, mℓ, ms), charge state, and the relativistic parameter αZ. What we see on classroom posters is this tensor contracted along most of its axes — neutrons integrated out, charge frozen to zero, orbitals collapsed into columns. The Mendeleev table is a projection.
The Periodic Stack of Computation has the same structure. Six axes. Some depend on physics. Some are pure math. The projection from one to the other gives you two stacks, not one.
Which axes are math, which are physics
Three of the six axes describe properties of an operation that exist independent of any physical substrate — they're true whether the operation runs on silicon, photons, DNA, or paper and pencil:
- Axis A — Family (33 families). The algebraic structure. What laws does the operation obey? This is pure mathematics.
- Axis E — State topology (4 types). Does the operation remember? Stateless, finite, accumulating, feedback. This is a property of the function definition, not the hardware.
- Axis F — Information topology (4 types). How does information flow? Bijective (1-to-1), reductive (N-to-1), transformative (N-to-N), generative (1-to-N). This is a property of the map, not the machine.
The other three axes describe how the operation interacts with a physical substrate:
- Axis B — Abstraction level. Can this be a gate, a microarchitecture, an ISA instruction, or software only? This depends on what substrates exist today — it changes over time.
- Axis C — Thermodynamic cost. Is this L₀ (free), L₁ (entropy), or L₂ (Landauer)? This is determined by information topology (F axis) via Landauer's principle — but the actual energy depends on the substrate.
- Axis D — Concurrency topology. Scalar, vector, matrix, graph, embarrassingly parallel, sequential. This describes how the operation maps to parallel hardware. Pure math doesn't parallelize — physics does.
The projection
Retain axes A, E, F. Discard axes B, C, D. What remains is the Math Periodic Stack: a three-axis coordinate space where every operation is positioned by its algebraic family, its state topology, and its information topology. No physics. No hardware. Just the mathematics.
The math stack has 33 × 4 × 4 = 528 cells. Each cell is a (Family × State × Info) coordinate. A cell is occupied if at least one mathematical operation with that exact combination of properties has been identified and formalized. A cell is empty if no such operation is known.
The relationship between the two stacks
The math stack is the source of truth. It contains every mathematical operation that has demonstrated computational character — whether or not anyone has tried to compute it on physical hardware. It currently has 258 primitives across 33 families.
The compute stack is the applied subset. It takes every math-stack primitive and adds the three physics axes: how abstracted it is (B), what it costs (C), how it parallelizes (D). It adds the silicon target — the concrete description of what the hardware should look like. And it adds the existing status — whether that hardware has been built.
The compute stack is always a lagging indicator of the math stack. Math knows operations before engineering builds them. The gap between the two stacks is the roadmap for the next generation of hardware.
Why the math stack is older than the compute stack
Every primitive in the compute stack has a mathematical definition that existed before anyone tried to compute it. FFT was known in 1805 (Gauss). Matrix decomposition was known in the 1850s (Jacobi). The Fourier transform itself dates to 1807. The mathematics is older than the hardware by decades to centuries.
This means the math stack predicts the compute stack. If a mathematical operation exists and is well-formalized, it will eventually appear in the compute stack with (B, C, D) coordinates — and eventually a chip will implement it. The math→compute→silicon pipeline is a one-way function with a fixed delay.
The 2026 mathematics survey covered every major branch — algebra, analysis, topology, number theory, geometry (all flavors), category theory, mathematical physics, dynamical systems, measure theory, combinatorics, logic, formal languages, and more. The 258 primitives and 33 families are the result. Eight survey waves. Every exclusion justified explicitly.
The critical property: F determines C
The information topology axis (F) directly determines the thermodynamic cost axis (C), even across the math-to-physics boundary. This is a consequence of Landauer's principle:
- Bijective (F₁, 1-to-1) → L₀ free. No information is erased, so no energy is required. This is not a design choice — it is physics.
- Reductive (F_N1, N-to-1) → L₂ or L₂max. Information is concentrated. Bits are destroyed. Landauer's cost is non-negotiable.
- Transformative (F_NN, N-to-N) → Depends. If information is rearranged without loss, it can be L₀ (like a permutation). If information is lost in the process, it's L₂.
- Generative (F_1N, 1-to-N) → L₁ (if entropy source needed) or L₀ (if deterministic expansion).
This means the math stack, which has no physics axes, still contains the thermodynamic prediction for every operation. You can read it off the F axis without knowing anything about hardware. The physics is embedded in the mathematics. That's the deepest structural insight of the Periodic Stack.
The compute stack is a 6-axis tensor. Project away the 3 physics axes and you get the Math Periodic Stack: 258 primitives across 33 families in a 528-cell (Family × State × Info) coordinate space. The math stack is the source of truth — it predicts both what the compute stack will contain and what the thermodynamic cost of each operation will be, because information topology determines thermodynamic class by Landauer's principle. Open the instrument → Section II to see the 528-cell tensor face.