Pseudo-Constructions: Surrogates in Theory & Practice

• Pseudo-constructions are a family of surrogate technical devices that relax strict construction criteria while preserving essential universal, geometric, or trace features.
• They span diverse fields such as model theory, algebra, category theory, and machine learning by replacing fully realizable constructions with approximative frameworks that maintain core structural properties.
• Applications include constructing Hrushovski generics, implementing blow-up and blur methods in algebraic logic, developing pseudo-derivations in vertex operator algebras, and mining slot templates in language tasks.

Pseudo-constructions designate a family of technical devices in which a strict, direct, or fully realizable construction is replaced by a surrogate that preserves selected universal, geometric, or trace-theoretic features. In the literature represented here, the term ranges from Hrushovski generics controlled by predimension and free pseudospaces, to blow-up-and-blur atom structures, pseudo-derivations and pseudo-endomorphisms in vertex operator algebra, Lie pseudo-bialgebras, pseudo-Kan extensions, pseudo-derived categories, pseudocharacters on Brauer envelopes, and automatically mined slot templates for compositional generalization (Valizadeh et al., 2018, Tent, 2011, Ahmed, 2013, Li, 2010, Boyallian et al., 2011, Nunes, 2016, Positselski, 2017, Im et al., 2023, Katrapati et al., 24 Sep 2025).

1. Range of meanings

The prefix “pseudo-” does not mark a single invariant weakening. In some settings it indicates bicategorical coherence in place of strictness; in others it marks trace data that may or may not lift to genuine representations; elsewhere it denotes intermediate categories between established extremes, or induced templates that approximate reusable form–meaning pairings. This suggests a family resemblance rather than a single doctrine.

Domain	Pseudo-construction	Core role
Model theory	Hrushovski constructions, free pseudospaces	Generic structures via predimension; controlled ampleness
Algebraic logic	Blow up and blur	Weak representability without strong representability
VOA / algebra	Pseudo-derivations, pseudo-endomorphisms, pseudo-bialgebras	Coefficient-valued or pseudotensor generalizations
Category theory / homological algebra	Pseudo-Kan extensions, pseudo-derived categories	Bicategorical adjoints; intermediate triangulated categories
Topology / representation theory	Pseudocharacters, pseudo-holonomies	Trace-like data with lifting problems
Machine learning	Pseudo-constructions	Induced slot templates for compositional generalization

A recurrent pattern is that pseudo-constructions preserve enough structure to support equivalence, descent, trace, or generalization arguments, while relaxing canonicality, strict realizability, or dimension bounds.

2. Logical and model-theoretic pseudo-constructions

In Hrushovski-style model theory, pseudo-constructions arise from predimension-controlled generics. For a finite relational structure $A$ in a language with one symmetric anti-reflexive relation $R$, the predimension used in the main body of the pseudofiniteness analysis is

$\delta(A)=|A|-|R[A]|.$

This induces a strong substructure relation $\leq^\ast$, and the Fraïssé–Hrushovski limit $\mathfrak M$ is the $\langle K,\leq^\ast\rangle$-generic. The arity of $R$ is decisive: when $R$ is ternary, $\langle \mathbb Q^+,<\rangle$ is interpretable in the generic and the theory is not pseudofinite; when $R$ is binary and $R$0, the generic is an acyclic forest, $R$1 is complete, and the theory is decidable and pseudofinite (Valizadeh et al., 2018).

A second model-theoretic usage appears in the free pseudospace of dimension $R$2. The language $R$3 has unary predicates $R$4 and edges only between adjacent levels. The resulting theory $R$5 yields examples of $R$6-stable theories that are $R$7-ample but not $R$8-ample, and the prime models are buildings associated to certain right-angled Coxeter groups (Tent, 2011). The combinatorial geometry is organized by incidence, flags, residues, meet and join, and reduced paths. Ampleness is witnessed by maximal flags, while failure of $R$9-ampleness is forced by the bounded length of such flags.

Algebraic logic contributes a third, more indirect, logical sense. The blow up and blur method starts from a finite graph, relation algebra, or cylindric algebra, replaces each colour or atom by infinitely many copies, and adds finitely many blurs. The term algebra is representable because the blurs are sufficient at that level, but the complex algebra is not representable because the original finite obstruction reappears there, yielding weakly representable atom structures that are not strongly representable (Ahmed, 2013). The same framework places Monk-like constructions and the Hirsch–Hodkinson probabilistic constructions in a common scheme, and it underlies non-elementarity results for the class of strongly representable atom structures.

3. Algebraic pseudo-constructions

In the theory of pseudoalgebras over a cocommutative Hopf algebra $\delta(A)=|A|-|R[A]|.$0, a Lie $\delta(A)=|A|-|R[A]|.$1-pseudoalgebra replaces an ordinary bracket by a pseudobracket

$\delta(A)=|A|-|R[A]|.$2

A Lie $\delta(A)=|A|-|R[A]|.$3-coalgebra is an $\delta(A)=|A|-|R[A]|.$4-module with cobracket $\delta(A)=|A|-|R[A]|.$5 satisfying co-Jacobi, and a Lie pseudo-bialgebra is a triple $\delta(A)=|A|-|R[A]|.$6 satisfying the cocycle compatibility

$\delta(A)=|A|-|R[A]|.$7

Within this framework one obtains analogues of the classical Yang–Baxter equation, Manin triples, and Drinfeld’s double. The associated annihilation algebra is identified with a convolution algebra $\delta(A)=|A|-|R[A]|.$8, making the annihilation construction conceptually transparent (Boyallian et al., 2011).

A different algebraic usage appears in kite pseudo BL-algebras. Starting from a basic pseudo hoop $\delta(A)=|A|-|R[A]|.$9, sets $\leq^\ast$0 and $\leq^\ast$1, and injections $\leq^\ast$2, the construction forms a lower part $\leq^\ast$3 with reversed order and an upper part $\leq^\ast$4, ordered so that every lower element lies below every upper element. Multiplication and residuals are then defined piecewise using coordinatewise operations on $\leq^\ast$5 and the maps $\leq^\ast$6. The resulting algebra $\leq^\ast$7 is a pseudo BL-algebra; when $\leq^\ast$8 and $\leq^\ast$9 are bijections it is a pseudo MV-algebra; and goodness is characterized by $\mathfrak M$0 (Dvurečenskij, 2014). A notable feature is that a basic commutative hoop can yield a non-commutative pseudo BL-algebra or an algebra with non-commuting negations.

Riordan theory uses yet another variant. A pseudo-involution is a Riordan array $\mathfrak M$1 such that $\mathfrak M$2, equivalently $\mathfrak M$3, has order $\mathfrak M$4 for $\mathfrak M$5. One characterization is

$\mathfrak M$6

The paper gives two complementary constructions: starting from $\mathfrak M$7, Theorem 22 states that if $\mathfrak M$8 and $\mathfrak M$9, with $\langle K,\leq^\ast\rangle$0, there exists a unique $\langle K,\leq^\ast\rangle$1 such that $\langle K,\leq^\ast\rangle$2 is a pseudo-involution, namely

$\langle K,\leq^\ast\rangle$3

starting from $\langle K,\leq^\ast\rangle$4, if $\langle K,\leq^\ast\rangle$5 has compositional order $\langle K,\leq^\ast\rangle$6, the set of $\langle K,\leq^\ast\rangle$7 for which $\langle K,\leq^\ast\rangle$8 is a pseudo-involution forms an infinite subgroup under multiplication (Marshall et al., 2021).

4. Categorical, homological, and universal pseudo-constructions

In 2-category theory, pseudo-Kan extensions replace strict right adjoints to precomposition by bicategorical right reflections. For pseudofunctors $\langle K,\leq^\ast\rangle$9 and $R$0, a right pseudo-Kan extension $R$1 is the right biadjoint to precomposition

$R$2

When pointwise bilimits exist, it is computed by

$R$3

This formalism subsumes descent objects and categories of algebras: the descent object of a pseudocosimplicial diagram is the value at $R$4 of a right pseudo-Kan extension, hence a conical bilimit, and the theory is organized by pseudomonads, idempotent pseudomonads, and a 2-dimensional adjoint triangle theorem (Nunes, 2016).

Homological algebra uses “pseudo-” to mark interpolation. A pseudo-dualizing complex $R$5 for a pair of associative rings $R$6 is obtained from a dualizing complex by dropping the injective dimension condition while retaining strong finite presentedness on both sides and the homothety isomorphisms. From such a complex one defines Auslander and Bass classes and, from them, pseudo-coderived and pseudo-contraderived categories standing in between the conventional derived category and the coderived or contraderived category. The main result is a triangulated equivalence between a pseudo-coderived category and a pseudo-contraderived category induced by $R$7 (Positselski, 2017).

A universal-construction variant is provided by the Brauer envelope. For a small category $R$8, the Brauer envelope $R$9 is the free rigid symmetric monoidal category generated by $R$0; objects are finite signed sequences of objects of $R$1, and morphisms are decorated one-dimensional cobordisms modulo composition and rigid-symmetric relations. An evaluation

$R$2

yields, via universal construction, an $R$3-linear rigid symmetric monoidal category $R$4 and state-space functors. In this setting, pseudocharacters are evaluations satisfying antisymmetrizer degree conditions; for Brauer categories with at most countably many objects over an algebraically closed field of characteristic $R$5, pseudocharacters lift to characters of semisimple representations (Im et al., 2023).

5. Operatorial and geometric pseudo-constructions

Vertex-operator algebra theory uses pseudo-derivations and pseudo-endomorphisms as coefficient-valued analogues of derivations and endomorphisms. For a commutative associative $R$6-algebra $R$7 with derivation $R$8, an $R$9-valued pseudo-derivation is a linear map $\langle \mathbb Q^+,<\rangle$0 satisfying a twisted derivation identity, and an $\langle \mathbb Q^+,<\rangle$1-valued pseudo-endomorphism is a linear map $\langle \mathbb Q^+,<\rangle$2 compatible with the vertex structure. A fundamental example is

$\langle \mathbb Q^+,<\rangle$3

which is an $\langle \mathbb Q^+,<\rangle$4-valued pseudo-derivation. Under suitable hypotheses, $\langle \mathbb Q^+,<\rangle$5 is a pseudo-automorphism, and twisting a module by

$\langle \mathbb Q^+,<\rangle$6

produces new modules or twisted modules. The classical inner-automorphism twisting series

$\langle \mathbb Q^+,<\rangle$7

is reinterpreted as the exponential of a logarithmic pseudo-derivation (Li, 2010).

Low-dimensional topology provides a different, geometric usage centered on explicit pseudo-Anosov constructions. Using Dehn surgery on double twist knots $\langle \mathbb Q^+,<\rangle$8 and $\langle \mathbb Q^+,<\rangle$9, together with Turaev–Viro growth under Dehn filling, the paper constructs infinite families of $R$0-hyperbolic knots and explicit pseudo-Anosov mapping classes on $R$1. The monodromies

$R$2

are pseudo-Anosov for every $R$3, and the $R$4-hyperbolicity of their mapping tori implies that they satisfy the Andersen–Masbaum–Ueno conjecture about quantum representations (Kalfagianni et al., 2023).

6. Data-driven pseudo-constructions in compositional generalization

In sequence modeling, pseudo-constructions are computational approximations to Construction Grammar form–meaning pairings. A pseudo-construction is a partially specified template induced from training data, with fixed tokens and slot variables, such as _ around _ twice. The mining procedure on SCAN has three components: candidate extraction from contiguous spans of length at most $R$5, beam-search segmentation of source sentences into patterns and residual words, and a target-aware misalignment penalty (Katrapati et al., 24 Sep 2025).

For a candidate pattern $R$6, if $R$7 is the set of source sentences containing $R$8, each with target $R$9, the misalignment score is

$R$00

where $R$01 is aligned to the nearest-neighbour source sentence of $R$02. Low $R$03 favors patterns whose different instantiations map to similarly structured target sequences. The source and target sides are then rewritten with slot tokens $R$04, and a standard Transformer Seq2Seq model is trained on the abstracted pairs.

On SCAN, this preprocessing yields substantial improvements on out-of-distribution splits without architectural changes or additional supervision: accuracy rises to $R$05 on ADD JUMP and to $R$06 on AROUND RIGHT, and with about $R$07 of the original training data the model still reaches $R$08 on ADD JUMP (Katrapati et al., 24 Sep 2025). In this setting, “pseudo-” marks an induced, task-specific surrogate for conventionalized constructions rather than a formally universal weakening.

Across these domains, pseudo-constructions are best understood as structurally constrained surrogates. They preserve enough of the original construction to support representation, descent, trace, or generalization results, while shifting the technical burden from strict realization to coherence data, quotienting, interpolation, or induced schematic structure.