Physics-Integrated Operator Learning via Gaussian Splatting Representations

Jihao Zhang; Junyi Guo; Jian-Xun Wang

cs.LGAug 25, 2026

Physics-Integrated Operator Learning via Gaussian Splatting Representations

Authors: Jihao Zhang, Junyi Guo, Jian-Xun Wang

Organizations: Sibley School of Mechanical and Aerospace Engineering, Cornell University, Ithaca, NY, USA

Abstract

Neural operators provide efficient surrogates for spatiotemporal PDE systems, but purely data-driven formulations often accumulate substantial errors during long-horizon autoregressive prediction and may fail to exploit available governing-equation structure. Existing approaches incorporate physics primarily through residual-based training objectives or PDE-specific architectural constraints, which can introduce optimization difficulties or limit architectural generality. In this work, we introduce a representation-level approach to physics integration in which a feed-forward Gaussian splatting (FFGS) representation serves as a continuous interface between discretized solution fields and governing operators. The FFGS representation reconstructs the state as a continuous Gaussian field with closed-form spatial derivatives, allowing available physical PDE operators to be integrated directly within the learned evolution map without introducing a physics-residual loss. We evaluate the framework across two- and three-dimensional PDE systems, including advection, diffusion, nonlinear self-advection, and reaction dynamics. Over long-horizon autoregressive rollouts, the proposed framework reduces relative $\ell_2$ error by $1.5\times$ -- $2.2\times$ compared with the strongest purely data-driven baseline across the benchmark suite, while consistently improving spectral fidelity. The framework also remains effective when the governing equations are partially known, demonstrating robustness to incomplete physics. These results demonstrate that continuous field representations can provide a practical interface for incorporating known physical structure into generic neural-operator surrogates.

Figures & tables

Panel of Figure 1 — Figure 1: Method overview. (a) The FFGS encoder maps a normalized grid field to anisotropic Gaussian primitives, which are rendered locally into a continuous field. The encoder is trained in Stage 1 by field reconstruction and then frozen. (b) The physics-integrated neural operator $\mathcal{G}^{\theta}_{\Delta t}$ adds the embedded-physics component $\Phi^{\mathrm{FFGS}}_{\Delta t}$ to the neural-operator component $\mathcal{N}_{\theta}$ , with both acting on the same state. $\Phi^{\mathrm{FFGS}}_{\Delta t}$ evaluates the available governing terms with the filtered FFGS quantities and advances the state over $\Delta t$ . Each prediction is fed back as the next input, starting from $\mathbf{u}^{\mathrm{pred}}(t_{0})=\mathbf{u}(t_{0})$ . In Stage 2, both components are evaluated during the multi-step rollout, but only $\mathcal{N}_{\theta}$ is trained, and no gradients propagate through $\Phi^{\mathrm{FFGS}}_{\Delta t}$ .

Benchmark	Method	#Params	PSNR $\uparrow$ (dB)	$\mathrm{r}\ell_{2}\downarrow$	Spec. err $\downarrow$
Advection (2D)	FNO	57.8k	$40.04\pm 1.68$	$0.089\pm 0.011$	$0.0214$
	U-Net	55.7k	$36.20\pm 1.14$	$0.129\pm 0.025$	$0.0403$
	ResNet	61.2k	$29.42\pm 0.66$	$0.394\pm 0.028$	$0.2898$
	Ours	57.8k	$\mathbf{44.75\pm 1.92}$	$\mathbf{0.052\pm 0.014}$	$\mathbf{0.0031}$
Adv–Diffusion (2D)	FNO	57.8k	$41.41\pm 1.75$	$0.084\pm 0.011$	$0.0276$
	U-Net	55.7k	$36.37\pm 1.70$	$0.146\pm 0.037$	$0.0894$

Table 2: Aggregate test-rollout performance on the five full-physics PDE benchmarks.

Figure 2: Error evolution over the rollout on the Burgers benchmark. Relative

\ell_{2}

error is plotted on a logarithmic scale for the proposed surrogate and three purely data-driven baselines. Panel (a) shows rollouts on the training trajectories, and panel (b) shows rollouts on held-out trajectories with unseen initial conditions. The shaded region in panel (a) indicates extrapolation beyond the 50-step training horizon (

t>50

Figure 3: Three-dimensional advection–diffusion rollout for a representative held-out trajectory at large Péclet number. Positive and negative scalar structures are visualized using the isosurfaces $u=+0.4$ and $u=-0.4$ , respectively. Columns correspond to the ground truth, the proposed model, FNO, U-Net, and ResNet. Rows show the initial condition and the autoregressive predictions at $t=50$ , $100$ , and $200$ . All methods are initialized from the same ground-truth field at $t=0$ , and the viewing direction and isosurface levels are fixed across all panels.

Figure 4: Spectral fidelity on the 2D Burgers test set (

\nu=10^{-4}

; 30 held-out trajectories

\times

200 rollout steps). (a) Time-averaged radial energy spectrum

\overline{E}(\kappa)

for the reference solution and each model. (b) Ratio

\overline{E}(\kappa)/\overline{E}_{\mathrm{GT}}(\kappa)

over the energy-containing range. (c) Relative error of the time-averaged spectrum,

\|\overline{E}-\overline{E}_{\mathrm{GT}}\|_{2}/\|\overline{E}_{\mathrm{GT}}\|_{2}

, evaluated over

\kappa\geq 1

(Section 3.3 ).

Figure 5: Partial-physics regime on the advection–diffusion benchmark (30 held-out trajectories

\times

194 rollout steps). (a)

\mathrm{r}\ell_{2}

error of the proposed model and the baselines at matched and tenfold (

10\times

) trainable dynamics-map budgets. (b) Time-averaged energy spectra. The dotted line marks the band edge of the initial conditions.

Figure 6: Per-step inference cost measured as wall-clock time per autoregressive update. Values are the minimum over five timed rollouts after warm-up and are shown on a logarithmic scale.

Figure 7: Ablation in the full-physics regime. (a) Mean

\mathrm{r}\ell_{2}

error over the 200-step rollout with and without the embedded-physics component. Without this component, the model reduces to the FNO baseline, and the values coincide with those in Table 2 . (b)

\mathrm{r}\ell_{2}

error over the rollout on the Burgers benchmark with and without the low-pass filter. Both panels use a logarithmic scale.

Figure 8: Component ablation in the partial-physics regime. All variants use the same identified coefficients. FFGS-only is the embedded-physics component alone. FiLM-FNO is the neural-operator component trained alone, with the coefficients supplied through FiLM conditioning. (a)

\mathrm{r}\ell_{2}

error over the rollout. (b) Time-averaged energy spectra. The dotted line marks the band edge of the initial conditions. (c) Mean

\mathrm{r}\ell_{2}

error over the 200-step rollout.

Appendix figures & tables10 assets

Supplementary material from the paper’s appendix.

Appendix

Benchmark	$\hat{u}$	$\partial_{x}\hat{u}$	$\partial_{y}\hat{u}$	$\nabla^{2}\hat{u}$	Speedup
	Relative $\ell_{2}$ discrepancy
2D Adv	$1.8\times 10^{-6}$	$5.7\times 10^{-6}$	$1.0\times 10^{-5}$	$2.8\times 10^{-5}$	$13.9\times$
2D Adv–Diff	$1.9\times 10^{-6}$	$5.3\times 10^{-6}$	$1.2\times 10^{-5}$	$3.2\times 10^{-5}$	$14.3\times$
2D Burgers	$7.1\times 10^{-8}$	$1.1\times 10^{-7}$	$1.6\times 10^{-7}$	$5.8\times 10^{-7}$	$13.9\times$
2D Adv–Allen–Cahn	$7.5\times 10^{-8}$	$1.1\times 10^{-7}$	$2.3\times 10^{-7}$	$8.7\times 10^{-7}$	$14.6\times$

Appendix

Table 3: Local rendering with

W_{\mathrm{loc}}=4

compared with dense summation over all primitives. The table reports the relative

\ell_{2}

discrepancies of the reconstructed field and its derivatives and the speedup of the local rendering.

Benchmark	$e_{u}$	$e_{\nabla u}$	$e_{\nabla^{2}u}$
2D Adv	$1.41\times 10^{-3}$	$2.57\times 10^{-3}$	$6.51\times 10^{-3}$
2D Adv–Diff	$1.27\times 10^{-3}$	$2.48\times 10^{-3}$	$6.18\times 10^{-3}$
2D Burgers	$1.51\times 10^{-3}$	$5.13\times 10^{-3}$	$2.62\times 10^{-2}$
2D Adv–Allen–Cahn	$1.53\times 10^{-3}$	$1.17\times 10^{-2}$	$5.38\times 10^{-2}$
3D Adv–Diff	$5.87\times 10^{-3}$	$8.15\times 10^{-3}$	$1.34\times 10^{-2}$

Appendix

Table 4: Relative

\ell_{2}

errors of the frozen FFGS reconstruction and its closed-form derivatives, measured against pseudo-spectral references.

Benchmark	$N/2$	$\Delta t$	$\kappa_{c}$	Filter
2D Adv	80	1.0	6	Smooth, $\eta=2$
2D Adv–Diff	80	1.0	6	Smooth, $\eta=2$
2D Burgers	80	1.0	10	Sharp
2D Adv–Allen–Cahn	80	0.1	14	Sharp
3D Adv–Diff	32	1.0	8	Smooth, $\eta=2$
2D Adv–Diff (partial physics)	80	1.0	7	Sharp

Appendix

Table 5: Filter settings of the embedded-physics component.

N/2

is the largest Fourier index along each axis, and

\Delta t

is the integration step, which equals the frame spacing.

	Stage 1: FFGS encoder	Stage 2: rollout training
Optimizer	AdamW	Adam
Learning-rate schedule	Cosine decay	Linear warmup, cosine decay
Peak learning rate	$1.5\times 10^{-3}$	$1.0\times 10^{-3}$
Final learning rate	$1.0\times 10^{-6}$	$0$
Training length	80 epochs	10,000 steps
Batch size	32	20

Appendix

Table 6: Optimization settings of the two training stages. For the 3D benchmark, Stage 1 uses 40 epochs and batch size 8, and Stage 2 uses a global batch size of 12. For the partial-physics benchmark, Stage 1 uses 100 epochs and batch size 64.

Figure 9:

\mathrm{r}\ell_{2}

error over the 200-step rollout on the remaining full-physics benchmarks, averaged over the 30 held-out trajectories and shown on a logarithmic scale. (a) 2D advection, (b) 2D advection–diffusion, (c) 2D advection–Allen–Cahn, and (d) 3D advection–diffusion. All models are trained on the first 50 steps.

Figure 10: Representative rollout on 2D advection–Allen–Cahn. Columns show the initial condition and the predictions at $t=33$ , $66$ , $100$ , $133$ , $166$ , and $199$ . Rows show the reference solution, ResNet, FNO, U-Net, and the proposed model. The order parameter $u$ uses a common color scale across all methods and lead times.

Figure 11: Representative rollout on 2D advection–diffusion. Columns show the initial condition and the predictions at $t=33$ , $66$ , $100$ , $133$ , $166$ , and $199$ . Rows show the reference solution, ResNet, FNO, U-Net, and the proposed model. The scalar field $u$ uses a common color scale across all methods and lead times.

Figure 12: Representative rollout on 2D Burgers. Columns show the initial condition and the predictions at $t=33$ , $66$ , $100$ , $133$ , $166$ , and $199$ . Rows show the reference solution, ResNet, FNO, U-Net, and the proposed model. The displayed quantity is the vorticity normalized by the instantaneous reference standard deviation, $\omega/\sigma_{\omega}(t)$ , and the values of $\sigma_{\omega}(t)$ are given below each column.

Figure 13: Time-averaged radial energy spectra on the remaining full-physics benchmarks, averaged over the 30 held-out trajectories and 200 rollout steps. (a) 2D advection–Allen–Cahn, (b) 2D advection, (c) 2D advection–diffusion, and (d) 3D advection–diffusion. The

\kappa=0

mode is retained only for advection–Allen–Cahn (Section 3.3 ).

Figure 14: Model-capacity control on Burgers. (a)

\mathrm{r}\ell_{2}

error over the rollout for the proposed model and the baselines at matched (solid) and tenfold (dashed) trainable budgets, with parameter counts in the legend. (b) Spectral error, evaluated over

\kappa\geq 1

(Section 3.3 ).

Explore similar work

Feb 26, 2026

Learning Physical Operators using Neural Operators

Neural operators have emerged as promising surrogate models for solving partial differential equations (PDEs), but struggle to generalise beyond training distributions and are often constrained to a fixed temporal discretisation. This work introduces a physics-informed training framework that addresses these limitations by decomposing PDEs using operator splitting methods, training separate neural operators to learn individual non-linear physical operators while approximating linear operators with fixed finite-difference convolutions. This modular mixture-of-experts architecture enables generalisation to novel physical regimes by explicitly encoding the underlying operator structure. We formulate the modelling task as a neural ordinary differential equation (ODE) where these learned operators constitute the right-hand side, enabling continuous-in-time predictions through standard ODE solvers and implicitly enforcing PDE constraints. Demonstrated on incompressible and compressible Navier--Stokes equations, our approach achieves better convergence and superior performance when generalising to unseen physics. The method remains parameter-efficient, enabling temporal extrapolation beyond training horizons, and provides interpretable components whose behaviour can be verified against known physics.

Vignesh Gopakumar, Ander Gray, Dan Giles +5

UCL Centre for Artificial Intelligence · UK Atomic Energy Authority · LIX, CNRS, ´Ecole Polytechnique +2

PDE Surrogate Modeling Physics-Informed Neural Operators Computational Fluid Dynamics Modular Neural Networks

Ask PDF

May 25, 2026

Autoregression-Free Neural Operators for Time-Dependent PDEs

Neural operators learn mappings from function-dependent inputs to solutions, providing an effective framework for solving partial differential equations (PDEs). For time-dependent PDEs, existing methods typically perform long-horizon prediction through autoregressive rollout directly in high-dimensional physical field spaces, where each predicted state is recursively fed back as the input for the next step. Although effective for short-term prediction, this autoregressive rollout and the lack of continuous-time modeling lead to progressive error accumulation over long-horizon rollouts. In this work, we propose Autoregression-Free Neural Operators (AFNO), which map the time evolution of PDEs into a latent space and model continuous-time vector fields within it. AFNO uses flow matching to learn the latent vector field, thereby enabling continuous evolution over extended horizons, avoiding autoregressive rollout and capturing dynamics under varying parameter configurations through explicit conditioning on physical parameters. Theoretical analysis and extensive experiments on six PDEs demonstrate that AFNO improves long-horizon prediction stability and consistently reduces rollout errors compared with the baselines.

Jiaquan Zhang, Caiyan Qin, Haoyu Bian +7

School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu, Sichuan 611731, China · School of Robotics and Advanced Manufacture, Harbin Institute of Technology, Shenzhen, Guangdong 518055, China · School of Mathematical Sciences, Capital Normal University, Beijing 100048, China +3

Latent Dynamics Modeling PDE Operator Learning Spatiotemporal Forecasting

Ask PDF

Oct 4, 2026

Green-Routed Neural Operators: Physics Determines Where the Network Reads

We identify a mismatch between the physical role of transport fields in many PDEs and their usual role in neural operators: PDEs use them to select read coordinates, whereas neural operators typically treat them only as input values. We address this mismatch with the Green-Routed Neural Operator (GRNO), which uses the governing equation to determine where latent features are sampled. A parameter-free equation adapter evaluates the diagnostic relation and constructs a departure map whose values are the read coordinates. A multiscale encoder-decoder combines centered and routed reads of latent features to learn the complete finite-time update. Across five two- and three-dimensional PDE systems, GRNO achieves the lowest mean final relative

L^2

error on four under 40-step autoregressive evaluation and remains competitive on Keller-Segel. Fixed-weight route interventions reveal strong dependence on direction and spatial alignment in four systems, with weak dependence in Keller-Segel. In independently trained ablations, GRNO achieves lower mean errors than variants that supply the transport field only as an input feature, substitute a learned displacement for the equation-specified route, or apply the route with a spatial misalignment, across all five systems. It also substantially outperforms directly advecting the physical state and learning the remaining update, indicating that equation-specified read coordinates provide an effective structural prior for long-horizon PDE forecasting.

Chenhao Si, Ming Yan

School of Data Science The Chinese University of Hong Kong, Shenzhen Shenzhen, China

PDE Surrogate Modeling PDE Operator Learning Physics-Informed Neural Operators Neural Operators

Ask PDF