Physics-Grounded Symbolic Architecture (PGSA)

• PGSA is a design family that merges symbolic representations with explicit physical laws to achieve accurate simulation, prediction, and reasoning.
• It employs methods such as exact execution of causal bases, hierarchical soft-to-hard gating, and grammar-based PDE solutions to ensure physical consistency.
• Implementations across domains like robotics, high-energy physics, and motion forecasting demonstrate improved interpretability and reduced computational errors.

Searching arXiv for the cited PGSA-related papers to ground the article in current literature. Physics-Grounded Symbolic Architecture (PGSA) is a label used in recent arXiv literature for architectures that couple symbolic, discrete, or typed representations with explicitly physical structure, so that prediction, simulation, generation, or reasoning is organized around governing laws, geometric constraints, causal generators, or physically meaningful state variables rather than unconstrained statistical correlation alone. In current usage, the term covers several distinct but related constructions: Symbolic Kolmogorov-Arnold Networks with hard symbolic primitive selection, symbolic world models built from exact causal bases, physics-informed symbolic networks for PDEs, graph-network systems whose latent messages align with force vectors, retrieval-augmented symbolic regression for motion forecasting, deterministic geometric solvers for physics diagram synthesis, physics-aware scene graphs for robotics, and tool-constrained symbolic-computation pipelines for theoretical physics (Faroughi et al., 25 Mar 2026, Dobrin et al., 9 Jun 2026, Majumdar et al., 2022, Cranmer et al., 2019, Feng et al., 9 Jul 2025, Haque et al., 28 May 2026, Li et al., 7 Jun 2026, Menzo et al., 27 Mar 2026).

1. Definitional scope and recurring design commitments

A central formulation appears in the symbolic world-model literature. There, statistical World Models, including Joint-Embedding Predictive Architectures, learn an encoder $h\colon z\mapsto h(z)$ and a predictor $\hat W$ in latent space, and their theoretical identifiability holds only when $z$ is Gaussian and evolves under an OU process. By contrast, a PGSA “does not learn its forward operator from data,” “represents the world state directly in typed physical variables,” and “executes the known physical laws exactly (up to machine precision) at each step” (Dobrin et al., 9 Jun 2026). This establishes one strong interpretation of PGSA: symbolic grounding in the causal generator of dynamics.

A second interpretation appears in Symbolic-KAN. There, the architecture is “built on the Kolmogorov–Arnold representation theorem” and augments it “with an explicit library of analytic primitives and a hierarchy of soft-to-hard gating mechanisms so that, at the end of training, every hidden unit implements exactly one univariate primitive acting on exactly one learned linear projection” (Faroughi et al., 25 Mar 2026). The emphasis is not exact execution of a pre-specified law, but differentiable learning of a discrete symbolic structure that becomes closed-form after hardening.

A third interpretation appears in physics-informed symbolic networks. PISN assumes that an unknown PDE solution can be represented by a finite composition of primitive operators from a context-free grammar, relaxes production-rule choice into continuous weights, and trains the resulting symbolic network solely through physics-informed residuals, initial conditions, and boundary conditions (Majumdar et al., 2022). Here, physical grounding enters through residual minimization rather than through an externally given symbolic simulator.

Other PGSA formulations embed physical constraints into perception, graphics, and reasoning pipelines. DEM-NeRF integrates NeRF reconstruction with elasticity PDEs and variational energy (Tan et al., 28 Jul 2025). PhyDrawGen converts typed scene graphs into a deterministic Planar Straight-Line Graph whose constraints encode force balance, optical paths, and field topologies (Haque et al., 28 May 2026). PhysGraph constructs a hierarchical scene graph whose nodes carry geometry, mass, material, and articulation parameters (Li et al., 7 Jun 2026). Diagrammatica constrains an LLM agent to a convention-fixing diagram specification language and trusted symbolic backends (Menzo et al., 27 Mar 2026).

The literature therefore suggests that PGSA is not a single canonical architecture. A more precise reading is that it names a design family whose members bind symbolic structure to physically meaningful semantics, but do so with different commitments concerning learnability, determinism, and the division of labor between neural and non-neural components.

2. Formal representations

One mathematically explicit PGSA is the causal-basis construction. Let $z\in\mathbb{R}^n$ be the true latent state, $W\colon\mathbb{R}^n\to\mathbb{R}^n$ its transition map, $\mathcal{A}=\{a_1,\dots,a_m\}$ a finite set of deterministic functions, and $G=(V,E)$ a directed state graph whose nodes are typed physical variables. If $\mathcal{A}$ contains a causal basis for $W$, then there exists a finite composition

$$C = a_k\circ a_{k-1}\circ\cdots\circ a_1\in\mathcal{A}^*$$

such that $\hat W$0 on a dense domain. The PGSA predictor is then

$\hat W$1

This formulation makes the representation map the identity on typed variables and makes symbolic execution, rather than latent encoding, the core computational act (Dobrin et al., 9 Jun 2026).

Symbolic-KAN uses a different formalism. For input $\hat W$2, the scalar output is

$\hat W$3

and each hidden activation has the form

$\hat W$4

with

$\hat W$5

Here $\hat W$6 is a fixed library of analytic primitives, $\hat W$7 is a primitive-selection gate, and $\hat W$8 is an edge-selection mask. At convergence, $\hat W$9 and $z$0 become one-hot, so each unit retains exactly one projection and one primitive. For a single-layer model this yields the ridge-function form

$z$1

with each $z$2 selected from the library by gating (Faroughi et al., 25 Mar 2026).

PISN formalizes symbolic structure through a grammar. With nonterminal $z$3, alphabet $z$4, and production rules

$z$5

the discrete grammar is relaxed continuously as

$z$6

typically with a softmax over $z$7. Stacking such relaxations gives a multilayer symbolic network (Majumdar et al., 2022).

Graph-based PGSA uses a directed, fully connected graph $z$8 for $z$9-body systems. Messages are computed as

$z\in\mathbb{R}^n$0

incoming messages are pooled by

$z\in\mathbb{R}^n$1

and the node update yields $z\in\mathbb{R}^n$2. By fixing the message dimension $z\in\mathbb{R}^n$3, the model forces edge embeddings into the same dimensionality as physical force vectors (Cranmer et al., 2019).

PGSA also appears as a structured graph representation in robotics. PhysGraph defines a scene graph

$z\in\mathbb{R}^n$4

where object nodes $z\in\mathbb{R}^n$5 encode 3DGS geometry, AABB, fused CLIP+DINOv3 embedding, estimated mass, and a part subgraph, while part nodes $z\in\mathbb{R}^n$6 encode articulation type, joint axis, joint origin, Young’s modulus, Poisson’s ratio, and density (Li et al., 7 Jun 2026).

A more abstract grounding formalism is categorical. “Symbol Grounding via Chaining of Morphisms” models syntax, semantics, perception, and action as categories linked by functors

$z\in\mathbb{R}^n$7

so that the grounding map is $z\in\mathbb{R}^n$8. In that construction, grounding proceeds from linguistic parse, to logical semantics, to spatiotemporal relation, to action grammar (Lian et al., 2017).

3. Optimization, execution, and constraint enforcement

Symbolic-KAN is trained in two phases under a composite objective

$z\in\mathbb{R}^n$9

Primitive gates $W\colon\mathbb{R}^n\to\mathbb{R}^n$0 are implemented by a Gumbel-Softmax with annealed temperature $W\colon\mathbb{R}^n\to\mathbb{R}^n$1, while an entropy penalty

$W\colon\mathbb{R}^n\to\mathbb{R}^n$2

and a non-maximum-suppression penalty

$W\colon\mathbb{R}^n\to\mathbb{R}^n$3

drive selections toward sharp, non-redundant structure. Phase I optimizes projections, affine parameters, gating logits, and optional unit gates with Adam or SGD; once gates are sharp, Phase II hardens them to discrete one-hots and refines continuous parameters with a second-order method such as L-BFGS (Faroughi et al., 25 Mar 2026).

PISN and PINSN enforce physics by residual minimization. For a PDE $W\colon\mathbb{R}^n\to\mathbb{R}^n$4, the loss combines collocation residuals with initial and boundary terms: $W\colon\mathbb{R}^n\to\mathbb{R}^n$5 All derivatives required by the PDE are obtained by automatic differentiation through the symbolic network. Two extensions are integral to the framework: domain decomposition, where independent symbolic subnetworks are trained on subdomains with continuity constraints, and HyperPISN, where a hypernetwork outputs symbolic-network weights for PDE parameter values such as Reynolds number or viscosity. PINSN adds an MLP terminal at the end,

$W\colon\mathbb{R}^n\to\mathbb{R}^n$6

to model the residual of the symbolic network (Majumdar et al., 2022).

DEM-NeRF uses a staged optimization scheme. An instant-NGP NeRF $W\colon\mathbb{R}^n\to\mathbb{R}^n$7 is trained first on photometric loss over sparse multi-view images. A mesh is then extracted, particles are sampled from reconstructed density, geometry is frozen, and a second MLP $W\colon\mathbb{R}^n\to\mathbb{R}^n$8 is trained under either strong-form equilibrium or variational energy constraints. The joint loss includes a NeRF reconstruction term, total potential energy, Dirichlet boundary penalties, and Neumann traction penalties, with hyper-weights $W\colon\mathbb{R}^n\to\mathbb{R}^n$9 balancing vision and physics (Tan et al., 28 Jul 2025).

The motion-forecasting PGSA is explicitly “a fully inference-time pipeline.” It extracts trajectories with CoTracker, retrieves candidate equations via Normalized DTW from an equation bank, seeds an evolutionary symbolic-regression population with a fraction $\mathcal{A}=\{a_1,\dots,a_m\}$0 of retrieved equations, fits numeric parameters with L-BFGS, forecasts future trajectories by closed form or RK4, and conditions a frozen trajectory-guided diffusion model with Gaussian trajectory maps (Feng et al., 9 Jul 2025).

PhyDrawGen is organized as a symbolic propose-verify loop rather than a training objective. A scene-graph extractor yields a typed graph, a deterministic PSLG solver instantiates geometric primitives and exact constraints, an SVG renderer produces an intermediate diagram, and a fine-tuned Qwen-VL model proposes JSON patches to the scene graph until no violations remain or a maximum iteration count is reached (Haque et al., 28 May 2026).

Diagrammatica replaces free-form symbolic code generation with tool-constrained computation. The agent selects MCP-validated tools, emits a compact JSON diagram specification, and delegates exact symbolic or numerical manipulations to trusted backends. Two principal paths share the same diagram specification: NDA for order-of-magnitude estimates and EDA for tree-level symbolic calculations via automatic FeynCalc code generation and Mathematica execution (Menzo et al., 27 Mar 2026).

4. Major instantiations across domains

The term PGSA is deployed across markedly different technical domains. The following table organizes the principal instantiations represented in the cited literature.

Domain	PGSA instantiation	Core mechanism
Scientific machine learning	Symbolic-KAN	Analytic primitive library with hierarchical soft-to-hard gating
PDE solving	PISN / PINSN	Grammar-relaxed symbolic network trained by physics residuals
World modeling	Symbolic world model PGSA	Atom registry, typed variables, exact execution of a causal basis
Vision and elasticity	DEM-NeRF	NeRF reconstruction plus PINN/DEM elasticity constraints
Motion forecasting and I2V	Trajectory-guided PGSA	Retrieval-boosted symbolic regression and trajectory-conditioned diffusion
Physics diagram synthesis	PhyDrawGen	Typed scene graph, deterministic PSLG solver, propose-verify correction
Theoretical high-energy physics	Diagrammatica	Tool-constrained JSON DSL driving NDA and EDA backends
Robotics perception	PhysGraph	Physics-aware 3D scene graph with material, mass, and articulation
Physical law discovery	Graph-network PGSA	Pairwise locality, vector equivariance, linear superposition
Symbol grounding and robotics	Chaining of morphisms	Functorial mapping from language to semantics, perception, and action
Interpretable signal reasoning	Quantum Spectral Reasoning	Padé/Lanczos spectral extraction mapped to symbolic predicates

In graph-network law discovery, PGSA is a specialized Graph Network whose messages mirror pairwise physical interactions, and symbolic regression is then used to fit explicit algebraic equations to the learned message function; in the inverse-square experiment, the recovered symbolic form matches Newtonian structure up to a learned linear transform (Cranmer et al., 2019).

In diagram generation, PGSA takes the form of a typed symbolic scene representation plus exact geometric reasoning. Mechanics is encoded through vector closure and perpendicularity/parallelism constraints, optics through Snell’s law, thin-lens equations, and reflection, and electromagnetism through Coulomb and Lorentz-force direction constraints as well as planar no-cross field topology (Haque et al., 28 May 2026).

In robotics, PhysGraph grounds symbolic structure in reconstructed 3D object-centric geometry, part decomposition, material inference, adaptive voxelization, mass estimation $\mathcal{A}=\{a_1,\dots,a_m\}$1, and explicit revolute or prismatic articulation parametrizations. Its scene graph is then queried for affordance prediction or exported to MuJoCo for real-to-sim transfer (Li et al., 7 Jun 2026).

In high-energy physics, Diagrammatica’s PGSA places symbolic structure in a diagram specification language whose fields fix spins, masses, Lorentz structures, couplings, propagators, and color factors. This shifts correctness from implicit textual convention-following to auditable symbolic specification plus trusted execution (Menzo et al., 27 Mar 2026).

A plausible implication is that “physics-grounded” and “symbolic” are orthogonal axes in this literature. Some systems are symbolic because they harden learned primitive choices into analytic expressions; others are symbolic because they operate on typed scene graphs, formal grammars, Horn clauses, or categorical morphisms; still others are symbolic because the executable object is a finite composition of known physical atoms.

5. Theoretical results and reported empirical performance

One strong theoretical line is provided by the symbolic world-model PGSA. It proves three principal statements: exact linear identifiability for all physical regimes regardless of latent distribution; a per-step error bounded by numerical precision,

$\mathcal{A}=\{a_1,\dots,a_m\}$2

with $\mathcal{A}=\{a_1,\dots,a_m\}$3 for conservative or dissipative systems; and a temporal consistency horizon

$\mathcal{A}=\{a_1,\dots,a_m\}$4

contrasted with $\mathcal{A}=\{a_1,\dots,a_m\}$5 for statistical latent-space models and $\mathcal{A}=\{a_1,\dots,a_m\}$6 for pixel-space models (Dobrin et al., 9 Jun 2026).

Several PGSA systems report strong empirical results in their respective domains.

System	Reported result	Source
DEM-NeRF	Surface-displacement RMSE “< 1 mm avg.”; inference “~1 s” vs. FEM “∼1 140 s” and PAC-NeRF “∼600 s”; learned $\mathcal{A}=\{a_1,\dots,a_m\}$7 within 5%	(Tan et al., 28 Jul 2025)
Motion forecasting PGSA	Spring–mass: ReSR $\mathcal{A}=\{a_1,\dots,a_m\}$8 TED $\mathcal{A}=\{a_1,\dots,a_m\}$9, MSE $G=(V,E)$0; synthetic video: Kling with ReSR FVD $G=(V,E)$1, FID $G=(V,E)$2, TrajErr $G=(V,E)$3; human A/B preference “>70%” for physics alignment	(Feng et al., 9 Jul 2025)
Graph-network PGSA	In 3D inverse-square $G=(V,E)$4-body, test error remains flat “within 10% of its training-time loss” up to $G=(V,E)$5 when trained on $G=(V,E)$6 with $G=(V,E)$7	(Cranmer et al., 2019)
PINSN	“2–3 orders of magnitude” reduction in maximum pointwise error over standard PINN; Kovasznay at $G=(V,E)$8: $G=(V,E)$9	(Majumdar et al., 2022)
PhyDrawGen	On 1,449 textbook problems: VCSR $\mathcal{A}$0, LblCSR $\mathcal{A}$1, $\mathcal{A}$2, Blind $\mathcal{A}$3	(Haque et al., 28 May 2026)
PhysGraph	Replica mIoU $\mathcal{A}$4 vs. $\mathcal{A}$5; articulation joint accuracy $\mathcal{A}$6; affordance query success $\mathcal{A}$7	(Li et al., 7 Jun 2026)
Diagrammatica	For 19 tree-level single-vertex $\mathcal{A}$8 decays, “18/19 succeeded first”; $\mathcal{A}$9: $W$0 MeV vs. $W$1 MeV; $W$2: $W$3 MeV vs. $W$4 MeV	(Menzo et al., 27 Mar 2026)
Quantum Spectral Reasoning	SMAP F1 $W$5 vs. LSTM-AE $W$6; SWaT F1 $W$7 vs. CNN $W$8; CLEVR-X $W$9 vs. NS-CL $$C = a_k\circ a_{k-1}\circ\cdots\circ a_1\in\mathcal{A}^*$$0	(Kiruluta, 5 Aug 2025)

Symbolic-KAN’s reported contribution is more structural than metric-centric in the supplied text. It “reliably recovers correct primitive terms and governing structures in data-driven regression and inverse dynamical systems,” extends to “forward and inverse physics-informed learning of partial differential equations,” and produces “compact symbolic representations whose selected primitives reflect the true analytical structure of the underlying equations” (Faroughi et al., 25 Mar 2026).

PhyDrawGen and Diagrammatica illustrate a different form of empirical validation: not approximation to a latent function class, but reduction of symbolic-physical violations. PhyDrawGen reports large gains over GPT-5-image and Gemini baselines on force-arrow and angle correctness (Haque et al., 28 May 2026), while Diagrammatica validates a tool-constrained symbolic-computation stack through exhaustive decay catalogs, special-case simplifications, Standard Model checks, and agentic orchestration without human edits (Menzo et al., 27 Mar 2026).

6. Relationship to neighboring paradigms, limitations, and open issues

PGSA is consistently defined against neighboring approaches. Symbolic-KAN is contrasted with classical symbolic regression, which relies on combinatorial tree search, and with library-sparse methods such as SINDy and ADAM-SINDy, which only select from a fixed library rather than generating new functional compositions; it is also contrasted with MLP-based PINNs, which are said to reduce to opaque activations and suffer spectral bias (Faroughi et al., 25 Mar 2026). The symbolic world-model PGSA is contrasted with JEPA-style models whose exact linear identifiability requires Gaussian stationary OU latents and whose representation bias compounds over time in non-Gaussian regimes (Dobrin et al., 9 Jun 2026). Diagrammatica is contrasted with standalone FeynCalc workflows and unconstrained LLM code generation, where subtle convention handling leads to silent errors or high execution uncertainty (Menzo et al., 27 Mar 2026).

The literature also states several domain-specific limitations. Symbolic-KAN may miss “exotic functions not in the initial library,” training can be “more delicate due to gating dynamics,” hyperparameters such as annealing schedules and $$C = a_k\circ a_{k-1}\circ\cdots\circ a_1\in\mathcal{A}^*$$1’s require tuning, and interpretability “only holds after hardening and assumes gates converge cleanly to one-hots” (Faroughi et al., 25 Mar 2026). DEM-NeRF assumes a “Homogeneous, isotropic Neo-Hookean material,” is “Quasi-static (no dynamics/inertia),” has “No fracture or contact handling,” and “Requires clear NeRF reconstruction” (Tan et al., 28 Jul 2025). Quantum Spectral Reasoning notes that pole misidentification can yield incorrect predicates, current rule design is manual, and non-stationary or non-spectral data may require hybrid modules (Kiruluta, 5 Aug 2025). Diagrammatica is “currently tree-level only,” lacks one-loop and real-emission support, does not automatically handle flavor-matrix interference for non-diagonal Yukawa couplings, supports EFT operators of dimension $$C = a_k\circ a_{k-1}\circ\cdots\circ a_1\in\mathcal{A}^*$$2 only through NDA, and requires UFO-to-FeynGraph loading for BSM models (Menzo et al., 27 Mar 2026). PhysGraph’s modules “run in stages” with “no joint end-to-end training” (Li et al., 7 Jun 2026).

A common misconception would be to treat PGSA as synonymous with symbolic regression. The cited work does not support that reduction. Some PGSAs perform symbolic regression or symbolic selection, but others are exact symbolic simulators, typed scene-graph solvers, geometric constraint systems, graph-based inductive-bias architectures, or rule-based reasoning stacks. A second misconception would be to treat PGSA as anti-neural. Several implementations are explicitly neuro-symbolic: DEM-NeRF combines NeRF with PINN/DEM (Tan et al., 28 Jul 2025), PhyDrawGen combines GPT-4o, Qwen-VL, and deterministic geometry (Haque et al., 28 May 2026), and PhysGraph combines pretrained visual modules, GPT-5 reasoning, and structured 3D scene graphs (Li et al., 7 Jun 2026).

The literature suggests two open tensions. One is between exactness and coverage: architectures with exact symbolic execution typically require a known causal basis or solver, whereas architectures that learn symbolic structure from data depend on libraries, grammars, or gated primitive sets. The other is between interpretability and residual expressivity: PINSN explicitly appends an MLP residual to a symbolic core (Majumdar et al., 2022), and Symbolic-KAN uses continuous mixtures before hardening (Faroughi et al., 25 Mar 2026). These tensions do not negate the PGSA program; they define the principal trade space within which current implementations operate.