Causal Set Theory

The Program in Brief

Causal set theory begins with a slogan of unusual economy: order plus number equals geometry. Take a Lorentzian spacetime and throw away everything except which events causally precede which, and how many events there are. Sorkin’s claim, following the 1987 Bombelli–Lee–Meyer–Sorkin paper, is that you have thrown away nothing — the causal order recovers the metric up to a conformal factor, and the counting supplies the factor. So the fundamental object need not be a manifold. It can be a causal set: a locally finite partially ordered set, and nothing else.

The move that makes this more than a reformulation is the treatment of discreteness. Every naive discretization — a lattice, a grid, a cell complex — picks out a rest frame and breaks Lorentz invariance at the discretization scale. Causal set theory’s answer is to make the discreteness random. Sprinkle points into spacetime by a Poisson process at fixed density and the configuration is exactly Lorentz invariant, because the Poisson distribution depends only on the 4-volume and 4-volume is a Lorentz scalar. This is not an approximation or a suppression to high order. It is exact, and it is the most elegant result in the program.

On that foundation the program has built the Hauptvermutung, classical sequential growth models, the nonlocal causet d’Alembertian, and entanglement-entropy computations that recover area scaling. And in 1987 — eleven years before the supernova results — Sorkin used a number-fluctuation argument to predict that the cosmological constant should be small, nonzero, and of order the critical density. It is one of the few pre-registered successes in quantum gravity.

Discreteness That Does Not Break Lorentz Invariance

This framework does not argue with any of the above. It adopts it. The Causal Set Statistics derivation opens by verifying that the framework’s relational-invariant network satisfies the causal set axioms — partial order from the coherence dependency structure, local finiteness from the holographic bound — and then invokes the Bombelli et al. uniqueness result directly. The consequence is stated in the framework’s Lorentz Invariance derivation (Remark 10.2): the discrete substrate does not produce Lorentz violation, and the framework therefore predicts none will be found at any energy, including the Planck scale. That prediction is not the framework’s achievement. It is a causal-set result it is entitled to because it built the same kind of object.

What it does with it afterward is its own. Along a geodesic, the central limit theorem applied to Planck cells gives length variance proportional to Planck length times path length — the square-root scaling behind the holographic noise prediction. At cosmological scales, the same density fluctuations combine with loop-closure pressure to set a quantum Jeans mass and a Gaussian power-spectrum cutoff — the dark matter granularity prediction, whose mass scaling distinguishes it from warm dark matter. Two of the framework’s most concrete experimental targets are, at bottom, causal-set statistics evaluated at two different scales.

Where the Causal Order Comes From

The true divergence is not about what the substrate looks like but about which end of the argument is load-bearing.

Causal set theory posits the partial order. It is the primitive, and the research problem is to grow geometry, dynamics, and matter out of it. The framework runs the derivation the other way. Its primitives are observers — coherence conservation, a self/non-self boundary, a closed phase loop — and the partial order arrives as a theorem. In Time as Phase Ordering, interaction events are vertices and directed phase transfers are edges; each transfer costs strictly positive coherence, so no edge can run backward; positive cost around a closed circuit contradicts conservation, so the graph is acyclic; and reachability on a directed acyclic graph is exactly a partial order. The framework’s own summary is blunt: it arrives at the same mathematical structure causal set theory postulates as primitive.

The derivation adds a feature the causal-set axioms do not carry: the events being ordered are not featureless points but interactions of a definite kind. The Three Interaction Types classification is exhaustive — Passage, Fusion, Resonance — and only Resonance generates a new relational invariant. That gives the arrow of time a structural rather than statistical basis: relational-invariant depth is monotone non-decreasing along any directed path, so the arrow is a property of the causal architecture, not of coarse-graining or initial conditions.

The framework is equally explicit about what it has not paid for. The step from the discrete order to smooth Lorentzian causal structure rests, by its own Proposition 7.1, on the Hauptvermutung — and its listed open gap on recovering the metric from the order is, in its own words, the central open problem of causal set theory.

Two Roads to a Small Positive Lambda

Both programs predict a small positive cosmological constant, and both land near the same number by visibly different mechanisms.

Sorkin’s is a fluctuation argument. If the number of elements in the observed 4-volume is $N$, then $\Lambda$ fluctuates at order $1/\sqrt{N}$ in Planck units; with a Hubble-scale 4-volume this gives $\Lambda \sim 10^{-122}\,\ell_P^{-2}$, which is right. The cost is that the resulting $\Lambda$ is a residual: everpresent, fluctuating, with no fixed sign.

The framework’s route runs through observer accounting rather than volume counting. Observer Loop Viability establishes $\Lambda > 0$ strictly in two independent moves. Negative $\Lambda$ is excluded because it forces a Planck-density recollapse whose bounce destroys every observer structure, leaving conserved coherence with no carrier. Zero $\Lambda$ is excluded because integer quantization of the Cauchy-slice coherence total plus the holographic bound on an observer’s epistemic horizon make the per-observer budget $C_0$ finite, and the relation $\Lambda_n^{\text{eff}} = 3\pi/(S^{(n)}\ell_P^2)$ then forces a floor $\Lambda \geq 3\pi/(C_0 \ell_P^2) \approx 2 \times 10^{-122}\,\ell_P^{-2}$. The observed value sits about a factor of 1.43 above it.

So both programs put $10^{122}$ at the centre of the answer — Sorkin as the square root of a 4-volume element count, the framework as a horizon bit budget entering linearly. But the outputs differ in character, and the difference is testable. The framework’s $\Lambda$ is constant, strictly positive, and pinned to $w = -1$ exactly, with phantom behaviour forbidden outright (no phantom dark energy); Sorkin’s can in principle wander in sign. Any detected time variation in $w$ is fatal to the framework and comfortable for an everpresent-$\Lambda$ picture. The framework’s broader treatment of the 120-order puzzle — as an obstruction class of an observer-indexed family of spacetime patches rather than a number to be tuned — is developed in Cosmological Constant.

Matter, Dynamics, and the Price of Each Route

The hardest problem in causal set theory is well known inside the program: dynamics and matter. Classical sequential growth gives a stochastic law for growing causets one element at a time, and the quantum measure programme aims at its quantum successor, but neither yields a particle spectrum or a gauge group. Fields on causal sets are studied by writing continuum field theories onto the discrete substrate — a more modest ambition than deriving which fields there are.

This is where the framework’s inverted derivation pays. Because its primitive is an observer rather than an order relation, the Relational Invariants generated by Type III interactions carry structure — division-algebra content, winding charges, boundary geometry — from which spin-statistics, generation count, and gauge structure follow. Causal-set elements are by construction bare; framework events are the residue of interactions between structured things, and inherit that structure.

The price should be stated plainly. Causal set theory is the more radical program: it builds the manifold from the order with nothing else in hand. The framework’s causal set is sprinkled into an ambient pseudo-Riemannian structure it does not derive — its active speed-of-light postulate — with the Planck density supplied by the area-scaling postulate. Both keep several derivations, including Causal Set Statistics itself, at provisional status. On background independence, causal sets are doing the harder and cleaner thing.

The framework does carry a layer a sprinkling does not. Aperiodic Order argues the observer network can be neither periodic — which makes strong subadditivity informationally degenerate — nor disordered, since unbounded density fluctuations would give different observers different effective gravitational couplings. What survives is aperiodic order with finite local complexity: a Delone set with matching rules and a metallic-mean substitution structure. Causal set theory imposes no such second layer.

This second layer is also where the framework currently owes causal set theory an answer rather than offering one. A Delone set with Pisot inflation is not a Poisson process — it is hyperuniform, with number variance growing sub-extensively and a pure-point diffraction spectrum where a sprinkling has none. The framework’s position is that the two descriptions apply to different objects, the sprinkling to interaction events and the aperiodic order to the observers that generate them. But events are by construction interactions among observers, so the event statistics are not free to be posited independently of the generator, and no derivation yet shows that an aperiodically ordered network induces a Poisson event process. It is recorded as an open gap (Causal Set Statistics, Open Gap 6). The honest summary is that the framework inherits Sorkin’s Lorentz-invariance argument while carrying a second structure whose compatibility with it is unproved — and that if the residual order does survive into the event process, the suppressed variance is a candidate first-principles route to the holographic-noise amplitude the framework currently cannot compute.

The Pattern: Same Object, Opposite Direction of Derivation

Across every point of contact the shape is the same. The two programs share an object — a locally finite, Poisson-sprinkled partial order — and disagree about which end of it is the premise. Causal sets start at the order and push toward geometry, and the open problems accumulate on the far end: the continuum limit, the quantum dynamics, the matter content. The framework starts at the observer and derives the order, and the open problems accumulate on the near end: it assumes the ambient geometry it sprinkles into, and inherits the Hauptvermutung as a debt rather than repaying it.

The convergence carries some weight. Two programs with unrelated starting premises — one axiomatizing causal order, one axiomatizing observers — arrive at the same discrete substrate with the same statistics. Neither is evidence for the other, but the agreement is at least weak evidence that a Poisson-sprinkled causal order is robust to the route taken.

Where They Genuinely Part Ways

Three divergences are real rather than presentational. First, the status of the partial order — primitive in one program, theorem in the other — is not a matter of taste, because the framework’s derivation attaches physical content (positive coherence cost per transfer, permanent relational invariants) that fixes the arrow of time structurally, where causal sets must get their arrow from the growth dynamics. Second, the character of $\Lambda$: everpresent, fluctuating and sign-indefinite versus strictly positive, constant, $w = -1$ exactly. Third, background independence versus derivational reach: causal sets assume less, while the framework assumes a Lorentzian ambient structure and a Planck scale and in exchange reaches particle content, gauge groups, and black hole entropy as observer-relative loop counting on a horizon. Which trade is better is settled by which program closes its open problems first.

What the Framework Takes From It

More than from any other program on this list, and it should be said without hedging.

It takes the causal set axioms themselves — used as the definition of what the relational-invariant network is, not as an analogy. It takes the Bombelli–Lee–Meyer–Sorkin uniqueness theorem that Poisson sprinkling is the only Lorentz-invariant point process, on which the framework’s prediction of exact Lorentz invariance at all energies rests entirely. It takes the Hauptvermutung as the bridge from its derived partial order to continuum causal structure, and flags it honestly as a conjecture it has not proved. And it takes the causet length-estimator literature — Brightwell–Gregory, Myrheim–Meyer, the Dowker et al. numerical work — as the reference frame against which its own holographic-noise amplitude must be judged, reporting the awkward result rather than burying it: a direct Monte Carlo on the longest-chain estimator gives variance scaling far below what the framework’s central-limit picture requires, which is why that amplitude coefficient remains a heuristic rather than a theorem.

Last, it takes the strategic lesson. Sorkin’s $\Lambda$ prediction is the standing demonstration that a quantum-gravity program can commit to a number in advance and be right, on the strength of a counting argument rather than a model. That is the standard the framework’s predictions are trying to meet, and the reason its holographic-noise amplitude is presented with its tension against the Holometer bound stated in the open rather than smoothed over.