Chapter 48
Assembly Theory: Causal Depth and Copy Number
48.1 Assembly Index as Causal Depth
Assembly Theory assigns to each physical object \(M\) an assembly index \(\mathrm{AI}(M)\): the minimum number of joining steps required to construct \(M\) from copies of its constituent parts, where each step joins two objects already present into a new one. The theory also proposes that time is an object: the assembly index records not a static property of \(M\) but the accumulated computational history of its construction.
This proposal maps precisely onto the notion of causal depth developed in Part I.
Let \(\GSLT\) be a GSLT. A set \(E \subseteq \terms(\GSLT)\) of elementary terms is a designated collection of starting materials — the analogue of atoms or monomers in chemistry — from which all other terms are constructed by rewriting.
The assembly index of a term \(M \in \terms(\GSLT)\) with respect to elementary terms \(E\) is \[\mathrm{AI}(M) \;=\; d^o_{\min}(E, M) \;=\; \min\bigl\{ |\gamma| \;\mid\; \gamma \text{ is a context-labeled path in } \mathcal{ST}_o \text{ from some } e \in E \text{ to } M \bigr\}.\] That is, \(\mathrm{AI}(M)\) is the minimum path length in the open synchronization tree from the elementary terms to \(M\), where each edge is labeled by the minimal context required to trigger the corresponding transition.
Assembly steps require an environment to supply a reaction partner — a catalyst, a monomer source, an energy input. These are precisely the transitions that appear as edges in the open synchronization tree (labeled by minimal contexts) rather than the closed tree (which records only autonomous rewrites). Each edge in \(\mathcal{ST}_o\) labeled by a minimal context \(K\) represents one joining step: the context \(K\) is the environmental participant that enables the construction. The nesting depth of context labels along a path is therefore exactly the assembly index: each layer of context is one more joining operation, one more step of accumulated causal history. This makes precise Cronin and Walker’s intuition that assembly index measures “time as accumulating computational effects”: the context labels are the accumulated effects, and their nesting depth is the causal time elapsed. The notion of time as an object in Assembly Theory is not a metaphor; it is the minimum context-labeled path length in an open synchronization tree.
48.2 The Two Gaps in Assembly Theory
Despite the strength of the assembly index concept, Assembly Theory has attracted criticism on two grounds.
Gap 1: Copy number lacks a notion of location.
The copy number \(\mathrm{CN}(M)\) is supposed to count how many copies of \(M\) are present in a sample. But counting requires specifying where the count is taken. In a spatially extended system, the same molecule may be abundant locally and rare globally. Assembly Theory provides no spatial or locational framework in which to ground the count, making copy number undefined in any setting where location matters. The failure is worse than intractability: a count over an unspecified region is not a hard count but an ill-formed one, since there is no collection over which to range and no moment at which to range over it. The repair is therefore to supply the missing argument rather than to approximate a true value, and Chapter 18 supplies it.
Gap 2: The identity criterion for copies is undefined.
Before counting copies of \(M\), one must say what it means for two objects to be copies of the same thing. Assembly Theory implicitly appeals to structural identity (same molecular graph), but this criterion is both too strict (ignoring conformational flexibility) and too informal (not specifying the relevant equivalence relation).
The Rho calculus fills both gaps simultaneously.
48.3 Locations as Channels, Namespaces as Regions
Among all GSLTs, the Rho calculus is distinguished by providing an abstract but mathematically precise notion of location.
In the Rho calculus, a channel is a name of the form \(@P\) for a process \(P\). Channels serve as locations: the process \(P\) is said to reside at the channel \(@P\) when it is stored there via \(@P!(P)\) or accessed via \(\mathtt{for}(y \leftarrow @P)\,Q\). Since channels are themselves processes (programs), location is a computational notion: where something is stored is itself a piece of running code.
A namespace is a finite set of channels \(\mathcal{N} = \{@P_1, @P_2, \ldots, @P_k\}\). A namespace describes a region of the computational space: the collection of locations at which a population is hosted.
The finiteness is a convenience and will not survive. A region worth naming — the channels belonging to an organism, or of its organs, or of their cells, and so on down — is rarely one anybody can list, and Chapter 18 replaces the list by a predicate, so that a region may be finitely described and unboundedly large at once. Everything in this chapter goes through unchanged for the flat case; Chapter 53 takes up what changes when the region has strata.
Let \(P \in \terms(\Rho)\) be a process and let \(\mathcal{N}\) be a namespace. The copy number of \(P\) in \(\mathcal{N}\) is \[\mathrm{CN}(P, \mathcal{N}) \;=\; \bigl|\bigl\{ c \in \mathcal{N} \;\mid\; {*}c \bisim P \bigr\}\bigr|\] the number of channels in \(\mathcal{N}\) whose dereferenced content is bisimilar to \(P\).
Definition 48.4 is the flat, full-resolution case of a quantity with three arguments rather than two. Bisimilarity is not decidable for calculi of this expressiveness, so no budgeted learner computes this count; the repair is the one \(\chSci\) applies to the assay, namely to relativise it to the depth an observer can afford. Definition 53.1 does that, and Proposition 53.2 records the consequence, which is that a coarse observer does not miss copies but manufactures them.
Definition 48.4 answers Gap 2 directly: two processes are copies of each other if and only if they are bisimilar — experimentally indistinguishable by any program context. This is the correct identity criterion for the purposes of the theory: what matters is not structural identity but behavioral identity under interaction. Two molecules are copies in the relevant sense if and only if no experiment can tell them apart.
With Proposition 48.1 and Definition 48.4, the biosignature criterion of Assembly Theory becomes the mathematically well-posed statement: \[\mathrm{AI}(P) \gg 1 \quad \text{and} \quad \mathrm{CN}(P,\,\mathcal{N}) \gg 1\] for some accessible namespace \(\mathcal{N}\). High causal depth (a long construction history) concomitant with high copy number (abundant presence in a region) is indicative of a replication process.
Two qualifications are owed and are paid later. The copy number here is taken at full resolution, which no instrument has; Remark 53.2 gives the resulting bias and its direction. And the two conjuncts are not independent once the ecology uses its architecture; Proposition 53.9 gives the curve relating them.