Active Catalytic Space in Computation & Catalysis

• Active Catalytic Space (ACS) is a polysemous term defining distinct phenomena in catalytic computation, autocatalytic networks, and topological catalysis.
• In catalytic computation, ACS involves actively manipulated memory that must be exactly restored, enabling advanced techniques like clean register programs and catalytic streaming.
• In network and materials science, ACS describes systems that maintain internal catalytic closure or exhibit symmetry-induced active sites, guiding catalyst design.

Active Catalytic Space (ACS) is not a single universally formalized concept in the arXiv literature. The term, or an explicitly related interpretation, appears in at least three technically distinct settings: catalytic computation, where a large auxiliary memory is actively manipulated under an exact restoration constraint; autocatalytic-network theory, where an ACS is a self-maintaining catalytic subgraph or reaction set; and crystalline catalysis, where ACS denotes the subset of real-space regions in which obstructed surface states or orbitally obstructed configurations generate catalytically active electronic structure. A separate direct-air-capture literature uses the same acronym for Alkalinity Concentration Swing, not for Active Catalytic Space (Alekseev et al., 24 Apr 2025, Zhang, 2013, Xu et al., 2021, Rinberg et al., 2021).

1. Terminological scope

The available literature suggests that “ACS” is best treated as a polysemous research term rather than as a single theory. In catalytic complexity, the phrase is used informally for catalytic memory that is “actively used” during computation but must be restored. In network chemistry and ecology, ACS usually abbreviates autocatalytic set. In topological materials and heterogeneous catalysis, ACS denotes a real-space map of surfaces, edges, or facets where symmetry-enforced electronic states produce catalytic activity. In direct air capture, ACS is explicitly defined as Alkalinity Concentration Swing, and the paper states that it is “not ‘Active Catalytic Space’” (Alekseev et al., 24 Apr 2025, Zhang, 2013, Xu et al., 2021, Rinberg et al., 2021).

Domain	ACS usage	Defining feature
Catalytic computation	Informal “active catalytic space”	catalytic tape is “actively transformed” and must be restored
Autocatalytic networks	autocatalytic set	every node has at least one incoming link from the same subgraph
Topological catalysis	active catalytic space	surfaces or edges with OSSs or obstructed orbitals
Direct air capture	Alkalinity Concentration Swing	concentration–dilution DAC cycle, not catalytic space

A recurring misconception is that ACS names a single formal object across these fields. The cited works do not support that reading. One computational paper states that it “does not introduce a new formal class called ‘Active Catalytic Space (ACS)’,” while the DAC paper makes the acronymic divergence explicit (Alekseev et al., 24 Apr 2025, Rinberg et al., 2021).

2. Catalytic computation and actively reusable memory

In catalytic complexity, the underlying formal model is the catalytic Turing machine with a standard work tape and a catalytic tape whose initial configuration must be exactly restored at the end. The associated class is $\CSPACE(s,c)$, and catalytic logspace is written as $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$ in one formulation and as $\CL := \bigcup_{d\in\mathbb{N}} \CSPACE[d\log n,n^d]$ in another. The computational significance of the model is that the catalytic tape is not passive ballast: it can be overwritten, shuffled, compressed, and reused, provided that the computation returns it to its original state (Alekseev et al., 24 Apr 2025, Agarwala et al., 14 Apr 2025).

The arithmetic abstraction for this “active” use is the clean register program. Over a finite ring $\mathbf{R}$, a register program updates registers by instructions of the form

$$R_i \leftarrow R_i + p(R_1,\dots,R_{i-1},R_{i+1},\dots,R_s)$$

or

$R_i \leftarrow R_i + \lambda x_j.$

A program is clean when only designated output registers differ from their initial values by a controlled increment, while all other registers are restored exactly. The Composition Lemma shows that clean programs compose without accumulating garbage: if $R_f$ and $R_g$ compute $f$ and $g$, then $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$0 uses $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$1 registers, $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$2 recursive calls, and $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$3 basic instructions. A clean register program over a finite ring can then be simulated by a catalytic Turing machine using $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$4 pure space and $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$5 catalytic space (Alekseev et al., 24 Apr 2025).

These tools support upper bounds beyond log-depth circuit evaluation. For all $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$6,

$\CL = \CSPACE(O(\log n), 2^{O(\log n)})$7

obtained by combining polynomial representations of depth-$\CL = \CSPACE(O(\log n), 2^{O(\log n)})$8 subcircuits with a layer-merging strategy. The same paper also proves a matrix-powering theorem: for all $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$9, there is a register program computing $\CL := \bigcup_{d\in\mathbb{N}} \CSPACE[d\log n,n^d]$0 with $\CL := \bigcup_{d\in\mathbb{N}} \CSPACE[d\log n,n^d]$1 recursive calls to $\CL := \bigcup_{d\in\mathbb{N}} \CSPACE[d\log n,n^d]$2, $\CL := \bigcup_{d\in\mathbb{N}} \CSPACE[d\log n,n^d]$3 basic instructions, and $\CL := \bigcup_{d\in\mathbb{N}} \CSPACE[d\log n,n^d]$4 registers over $\CL := \bigcup_{d\in\mathbb{N}} \CSPACE[d\log n,n^d]$5. These constructions are presented as steps toward understanding whether $\CL := \bigcup_{d\in\mathbb{N}} \CSPACE[d\log n,n^d]$6 (Alekseev et al., 24 Apr 2025).

The same active-memory viewpoint has recently yielded central algorithmic consequences. Bipartite maximum matching is in catalytic logspace with a polynomial time bound: $\CL := \bigcup_{d\in\mathbb{N}} \CSPACE[d\log n,n^d]$7. The algorithm uses a deterministic catalytic derandomization of the isolation lemma together with a compress-or-random framework, and the paper also proves $\CL := \bigcup_{d\in\mathbb{N}} \CSPACE[d\log n,n^d]$8. In data streams, a catalytic streaming model with clean space and catalytic memory yields exact multi-pass algorithms for frequency moments and graph counting: exact $\CL := \bigcup_{d\in\mathbb{N}} \CSPACE[d\log n,n^d]$9 in four passes using $\mathbf{R}$0 clean space, exact $\mathbf{R}$1 and $\mathbf{R}$2 in two and three passes respectively using $\mathbf{R}$3 clean space, exact $\mathbf{R}$4 in four passes using $\mathbf{R}$5 clean space, exact counting of induced occurrences of any fixed graph $\mathbf{R}$6 in four passes using $\mathbf{R}$7 clean space, and exact triangle counting in three passes using $\mathbf{R}$8 clean space. The same streaming paper proves that catalytic memory gives no additional power in the single-pass setting (Agarwala et al., 14 Apr 2025, Becker et al., 9 Jul 2026).

3. Autocatalytic sets in network and reaction-system theory

In autocatalytic-network theory, ACS most commonly means autocatalytic set. In the competitive percolation model, the system is a directed graph on $\mathbf{R}$9 nodes, with adjacency matrix $$R_i \leftarrow R_i + p(R_1,\dots,R_{i-1},R_{i+1},\dots,R_s)$$0, where $$R_i \leftarrow R_i + p(R_1,\dots,R_{i-1},R_{i+1},\dots,R_s)$$1 means that species $$R_i \leftarrow R_i + p(R_1,\dots,R_{i-1},R_{i+1},\dots,R_s)$$2 catalyzes species $$R_i \leftarrow R_i + p(R_1,\dots,R_{i-1},R_{i+1},\dots,R_s)$$3, and $$R_i \leftarrow R_i + p(R_1,\dots,R_{i-1},R_{i+1},\dots,R_s)$$4. An ACS is “a subgraph whose every node has at least one incoming link from a node that belongs to the same subgraph,” formally

$$R_i \leftarrow R_i + p(R_1,\dots,R_{i-1},R_{i+1},\dots,R_s)$$5

The paper also defines an independent ACS as “the maximal weakly-connected subgraph which is only composed of ACS-nodes,” and studies the largest independent ACS $$R_i \leftarrow R_i + p(R_1,\dots,R_{i-1},R_{i+1},\dots,R_s)$$6 as the analog of a giant component in percolation (Zhang, 2013).

The population dynamics couple graph structure to fitness. With absolute populations $$R_i \leftarrow R_i + p(R_1,\dots,R_{i-1},R_{i+1},\dots,R_s)$$7 and relative populations $$R_i \leftarrow R_i + p(R_1,\dots,R_{i-1},R_{i+1},\dots,R_s)$$8, the model uses

$$R_i \leftarrow R_i + p(R_1,\dots,R_{i-1},R_{i+1},\dots,R_s)$$9

Nodes with smaller stationary $R_i \leftarrow R_i + \lambda x_j.$0 are less fit and are more likely to mutate. The update rule is an $R_i \leftarrow R_i + \lambda x_j.$1-selection rule: among $R_i \leftarrow R_i + \lambda x_j.$2 least-fit candidate nodes, the system preferentially chooses rewiring moves that do not increase the size of the largest independent ACS, and only when every candidate would increase $R_i \leftarrow R_i + \lambda x_j.$3 is a growth event forced. The paper derives a threshold time $R_i \leftarrow R_i + \lambda x_j.$4 for abrupt ACS emergence,

$R_i \leftarrow R_i + \lambda x_j.$5

with $R_i \leftarrow R_i + \lambda x_j.$6, and reports that the maximal jump $R_i \leftarrow R_i + \lambda x_j.$7 grows linearly with $R_i \leftarrow R_i + \lambda x_j.$8. For $R_i \leftarrow R_i + \lambda x_j.$9, the fitted slopes are approximately $R_f$0, $R_f$1, and $R_f$2, whereas $R_f$3 shows a slope $R_f$4, corresponding to smooth rather than explosive growth (Zhang, 2013).

A chemically richer formulation appears in the artificial-chemistry thesis. There the chemistry contains ligation and cleavage reactions

$R_f$5

with catalysts multiplying both forward and reverse rates by $R_f$6. For a set $R_f$7 of catalyzed one-way reactions, let $R_f$8 be the products and $R_f$9 the reactants, excluding food molecules. Then $R_g$0 is an ACS if every reaction in $R_g$1 is catalyzed by some molecule in $R_g$2, and $R_g$3. In homogeneous uncatalyzed chemistry, large molecules are exponentially scarce; with an ACS, catalytic flux can focus resources onto ACS products, creating ACS-dominated steady states and bistability between low and high ACS states. For the example ACS65 with $R_g$4 and $R_g$5, the thesis reports $R_g$6 and $R_g$7 (Giri, 2013).

The thesis further identifies a central scaling obstacle: for isolated extremal ACSs, the maintenance threshold scales as a power law in catalyst length $R_g$8, but the initiation threshold scales exponentially,

$R_g$9

Its proposed resolution is the nested ACS or cascade structure, in which smaller ACSs reinforce larger ones by raising background and intermediate concentrations. In the ACS3+ACS8 example, once the smaller ACS crosses its own initiation threshold, the initiation threshold for the larger ACS drops by about four orders of magnitude. A cascade generated by nested ACSs can then support very large catalysts, including a reported $f$0 example culminating in $f$1, with catalytic strengths growing only as $f$2 for $f$3 and $f$4 (Giri, 2013).

4. Real-space topology and catalytically active regions in crystals

In topological materials and heterogeneous catalysis, ACS is a real-space notion tied to obstructed atomic insulators (OAIs), obstructed Wannier charge centers (OWCCs), obstructed surface states (OSSs), and real-space invariants (RSIs). An OAI is a topologically trivial insulator in which symmetric Wannier functions exist but some Wannier charge centers are pinned by symmetry at empty Wyckoff positions. RSIs are integer or mod-$f$5 indices defined per Wyckoff position, obtained from symmetry-data vectors $f$6 of irreducible representation multiplicities and Smith decompositions of induction matrices. Nonzero RSIs indicate that some Wannier orbitals cannot be removed from those positions in any symmetric, gapped gauge. In the catalytic interpretation, ACS is “the subset of real-space regions (typically particular surfaces, edges, or hinges) of a crystalline material where, due to real-space electronic topology encoded by RSIs and WCCs, obstructed surface states or orbitally obstructed configurations create metallic, charge-dense, symmetry-selected electronic states at or near the Fermi level” (Xu et al., 2021).

The catalytic principle is bulk–boundary in character. Bulk RSIs determine where OWCCs sit. Symmetry-preserving surfaces that intersect OWCCs exhibit filling anomalies and therefore host OSSs: metallic bands in the bulk gap, localized at the relevant surface. The papers argue that such surfaces have the defining properties of active catalytic sites—high stability, good electrical conductivity, and high charge carrier density near the Fermi level—and therefore identify a topologically determined ACS. One broad survey reports that, out of 34013 paramagnetic and 296 magnetic topologically trivial insulators, there are 3383 paramagnetic and 30 magnetic OAIs, together with a further 121 orbital-selected OAIs (OOAIs). A companion catalyst-discovery paper identifies 1788 unique OAIs (3383 ICSDs), of which 465 are potential high-quality catalysts, and derives the Miller indices of the active surfaces (Xu et al., 2021, Li et al., 2021).

The canonical proof-of-principle system is $f$7-MoS$f$8. The OAI diagnosis is a nonzero RSI,

$f$9

at empty Wyckoff position $g$0. The OWCC planes are intersected by side facets such as (100) and (010), but not by the ideal (001) basal plane without cutting through atoms. Surface-state calculations show metallic OSSs on the side surfaces and an absence of metallic OSSs on the basal plane. Electrochemical HER measurements on single crystals then show that edge-exposed samples retain almost the same activity as whole crystals, while activity is strongly reduced when edges are covered; hydrogen bubbles emerge exclusively from the edges, and edge sites have much lower charge-transfer resistance than the basal plane (Xu et al., 2021).

The same logic extends to $g$1-MoTe$g$2, $g$3-MoTe$g$4, and NiPS$g$5. For $g$6-MoTe$g$7 in space group 194, only Wyckoff $g$8 has nonzero RSI, $g$9, yielding OSSs on (100)-type edges but not on the (001) basal surface. For NiPS$\CL = \CSPACE(O(\log n), 2^{O(\log n)})$00 in space group 12, only Wyckoff $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$01 has nonzero RSI, $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$02, and the predicted active surfaces include $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$03, $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$04, $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$05, $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$06, $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$07, $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$08, $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$09, and $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$10, that is, side surfaces mostly perpendicular to the (001) basal plane. Conductive AFM maps show much higher conductivity at NiPS$\CL = \CSPACE(O(\log n), 2^{O(\log n)})$11 step edges than on flat basal regions, and HER measurements localize activity to the edge surfaces. By contrast, $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$12-MoTe$\CL = \CSPACE(O(\log n), 2^{O(\log n)})$13 is described as having metallic surface states on both basal and edge surfaces, so the catalytically active region is not confined to edges alone (Li et al., 2021).

5. Cross-domain structure

A plausible implication of these literatures is that the word “space” in ACS names three different mathematical objects. In catalytic computation it is memory space, constrained by exact restitution; in autocatalytic-network theory it is graph or reaction-network space, constrained by catalytic closure; in topological catalysis it is physical real space, constrained by crystal symmetry, Wyckoff geometry, and surface termination (Alekseev et al., 24 Apr 2025, Zhang, 2013, Xu et al., 2021).

A second commonality is the role of a global conservation or restoration principle. Catalytic Turing machines must return the catalytic tape to the initial configuration. Autocatalytic sets require internal catalytic support so that every member is sustained from within the set or from the food set. OAI-based ACS requires symmetry-pinned WCC structure, so that exposing the relevant surface necessarily produces OSSs. This suggests a shared organizational motif: ACS marks a structured domain that can be used intensively without losing the global constraint that defines it. In the computational papers the constraint is reversibility of memory; in the network papers it is catalytic closure; in the topological papers it is real-space symmetry obstruction (Alekseev et al., 24 Apr 2025, Giri, 2013, Li et al., 2021).

The quantitative control parameters are likewise domain-specific. In catalytic complexity they are clean space, catalytic space, time, passes, recursive calls, numbers of registers, and field size; in competitive autocatalytic percolation they are $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$14, $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$15, $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$16, $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$17, and $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$18; in topological catalysis they are RSIs such as $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$19, $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$20, and $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$21, together with Miller indices and the geometry of OWCC projections (Agarwala et al., 14 Apr 2025, Zhang, 2013, Xu et al., 2021).

6. Open problems, limitations, and terminological boundaries

In catalytic computation, the major open problem remains the characterization of $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$22. Current containments include $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$23, but the bounds are far apart. The specific questions emphasized are whether $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$24 contains higher levels of the $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$25 hierarchy, especially whether $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$26, whether the $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$27 upper bound can be driven down to $\CL = \CSPACE(O(\log n), 2^{O(\log n)})$28 work space, whether matching can be solved with much less catalytic space, and how broadly compress-or-random or catalytic derandomization methods extend beyond bipartite matching and current streaming tasks (Alekseev et al., 24 Apr 2025, Agarwala et al., 14 Apr 2025, Becker et al., 9 Jul 2026).

In autocatalytic-network theory, the competitive percolation paper does not rigorously determine whether the finite-size abrupt transition is discontinuous or continuous in the thermodynamic limit, and the thesis leaves open how far nested ACS cascades can be grounded in chemically realistic catalysts, rate heterogeneity, and compartmentalized settings. Both works nevertheless place the emergence, persistence, and controllability of large autocatalytic structures at the center of the problem (Zhang, 2013, Giri, 2013).

In crystalline catalysis, the OAI/OSS framework is presented as a symmetry-based method for locating catalytic activity, but it does not by itself determine adsorption energies, solvent effects, reconstruction pathways, or reaction barriers. Surface stability, passivation, strain, defects, and reaction-induced restructuring remain external to the topological diagnosis. The papers accordingly position ACS as a real-space descriptor that narrows the search space for catalytic sites rather than as a complete catalytic theory (Xu et al., 2021, Li et al., 2021).

Finally, the acronym itself has firm boundaries. In one strand of direct-air-capture research, ACS means Alkalinity Concentration Swing: a solution-based thermodynamic cycle in which an alkaline solution absorbs atmospheric CO$\CL = \CSPACE(O(\log n), 2^{O(\log n)})$29 as dissolved inorganic carbon, is concentrated so that the equilibrium CO$\CL = \CSPACE(O(\log n), 2^{O(\log n)})$30 partial pressure rises, outgasses CO$\CL = \CSPACE(O(\log n), 2^{O(\log n)})$31, and is then diluted and reused. That paper explicitly states that ACS there is “not ‘Active Catalytic Space’.” Any encyclopedia treatment of ACS therefore requires domain disambiguation before substantive analysis begins (Rinberg et al., 2021).