---
title: 'TENG-BC: Unified Neural PDE Solver'
url: https://www.emergentmind.com/topics/teng-bc
type: topic
---

# TENG-BC: Unified Neural PDE Solver

Searching arXiv for TENG-BC and related TENG neural PDE solver papers.
TENG-BC, short for "Unified Time-Evolving Natural Gradient for Neural PDE Solvers with General Boundary Conditions," is a neural method for solving time-dependent partial differential equations by combining explicit time integration with a boundary-aware natural-gradient update [2603.00397]. In the reported formulation, each time step jointly enforces interior dynamics and boundary conditions, accommodating Dirichlet, Neumann, Robin, and mixed constraints within a unified framework. The method is positioned as a high-precision, mesh-free solver for settings in which long-time error accumulation and boundary enforcement are the dominant difficulties. A closely related precursor, "TENG++: Time-Evolving Natural Gradient for Solving PDEs With Deep Neural Nets under General Boundary Conditions" [2512.15771], extended the original Time-Evolving Natural Gradient framework from periodic to Dirichlet boundary conditions by combining natural-gradient optimization with explicit Euler and Heun schemes.

## 1. Origins, scope, and nomenclature

The defining idea behind the Time-Evolving Natural Gradient family is to replace a vanilla gradient step,
$$
\theta_{n+1}=\theta_n-\eta\,\nabla_\theta L(\theta_n),
$$
with a geometry-aware update,
$$
\theta_{n+1}=\theta_n-\eta\,F(\theta_n)^{-1}\nabla_\theta L(\theta_n),
$$
where $F(\theta)$ is a Fisher-information or generalized Gauss–Newton matrix. In the account given for TENG++, the original TENG framework was demonstrated under periodic boundary conditions, while TENG++ extended it to Dirichlet boundary conditions through a loss of the form $L(\theta)=L_{\rm PDE}(\theta)+\lambda L_{\rm BC}(\theta)$ and explicit time stepping [2512.15771]. TENG-BC further reformulates boundary handling so that interior and boundary constraints enter a single local least-squares problem, and the resulting update is interpreted directly as a natural-gradient step under an $L^2(\Omega)$-type metric augmented by boundary contributions [2603.00397].

This progression suggests a shift from penalty-based boundary enforcement toward an intrinsic operator-aware update. In TENG++, the Dirichlet condition is imposed by a boundary penalty weighted by $\lambda$; in TENG-BC, the boundary operator appears directly in the metric and residual construction, and the abstract explicitly states that the method operates “without delicate penalty tuning” [2603.00397].

The acronym should also be distinguished from an unrelated usage in power-electronics literature, where TENG denotes a triboelectric nanogenerator rather than a Time-Evolving Natural Gradient method [2102.04581]. In the present context, TENG-BC belongs to neural PDE solvers rather than energy-harvesting circuitry.

## 2. Continuous formulation and natural-gradient structure

In TENG-BC, the PDE is written as an evolution equation
$$
u_t(x,t)=L\,u(x,t),\qquad x\in\Omega,\quad t\in[0,T],
$$
subject to a general mixed boundary condition
$$
a(x)\,u(x,t)+b(x)\,\partial_n u(x,t)=v(x,t).
$$
A neural network $u_\theta(x)$ with parameters $\theta\approx\theta(t)$ represents the solution at each time step [2603.00397].

At a step of size $\Delta t$, the method forms a target field $u_{\rm target}$ determined by the chosen time integrator. The forward Euler target is
$$
u_{\rm target}=u_\theta+\Delta t\,L\,u_\theta,
$$
and higher-order methods, including Heun and RK4, define $u_{\rm target}$ through intermediate evaluations of $L\,u_\theta$ [2603.00397]. The network is then linearized around $\theta$, and one seeks an increment $\Delta\theta$ such that the induced variation $J\,\Delta\theta$ approximates the target increment $\Delta u\equiv u_{\rm target}-u_\theta$ in $L^2(\Omega)$:
$$
\Delta\theta_e=\arg\min_{\Delta\theta}\|\Delta u-J\,\Delta\theta\|^2_{L^2(\Omega)},
$$
where $J(x)=\partial_\theta u_\theta(x)$.

The normal equations are
$$
G(\theta)\,\Delta\theta=-\nabla_\theta\mathcal L_e,
$$
with
$$
G(\theta)=\int_\Omega J^T J\,dx,\qquad
\nabla_\theta\mathcal L_e=\int_\Omega J^T(u_\theta-u_{\rm target})\,dx.
$$
Hence
$$
\Delta\theta=-G(\theta)^{-1}\nabla_\theta\mathcal L_e.
$$
The paper identifies this update as “precisely the natural-gradient step under the $L^2(\Omega)$ metric on function space” [2603.00397].

This formulation places the method closer to a local function-space projection than to ordinary first-order optimization. A plausible implication is that the geometry of the network enters only through the Jacobian and the induced metric $G(\theta)$, which helps explain why the method is reported to maintain low stepwise error over long time horizons.

## 3. Unified treatment of boundary conditions

The distinctive feature of TENG-BC is its boundary-aware least-squares step. For a general boundary condition $a\,u+b\,\partial_n u=v$ on $\partial\Omega$, the mismatch is defined by
$$
\Delta v=v-\bigl(a\,u_\theta+b\,\partial_n u_\theta\bigr),
$$
with boundary Jacobians
$$
J_\partial(x)=\partial_\theta u_\theta(x),\qquad
K_\partial(x)=\partial_\theta[\partial_n u_\theta](x),\quad x\in\partial\Omega.
$$
The boundary-only update solves
$$
\Delta\theta_b=\arg\min_{\Delta\theta}\|\Delta v-[a\,J_\partial+b\,K_\partial]\Delta\theta\|^2_{L^2(\partial\Omega)}.
$$
TENG-BC combines the interior and boundary objectives into a single local least-squares problem over the closure $\overline\Omega$:
$$
\Delta\theta=\arg\min_{\Delta\theta}
\left\{
\|\Delta u-J\,\Delta\theta\|^2_{L^2(\Omega)}
+
\|\Delta v-(a\,J_\partial+b\,K_\partial)\Delta\theta\|^2_{L^2(\partial\Omega)}
\right\},
$$
or equivalently
$$
\Delta\theta=\arg\min_{\Delta\theta}\|R-\mathcal J\,\Delta\theta\|^2_{L^2(\overline\Omega)},
$$
where $R=(\Delta u;\Delta v)$ and $\mathcal J=(J;\,aJ_\partial+bK_\partial)$ [2603.00397].

The corresponding metric becomes
$$
G(\theta)=\int_\Omega J^T J\,dx
+\int_{\partial\Omega}(aJ_\partial+bK_\partial)^T(aJ_\partial+bK_\partial)\,ds,
$$
and the gradient is
$$
\nabla_\theta\mathcal L=
\int_\Omega J^T(u_\theta-u_{\rm target})\,dx
+
\int_{\partial\Omega}(aJ_\partial+bK_\partial)^T\Delta v\,ds.
$$
The update remains
$$
\Delta\theta=-G^{-1}\nabla_\theta\mathcal L.
$$

| Boundary type | Parameters |
|---|---|
| Dirichlet | $(a,b)=(1,0)$ |
| Neumann | $(a,b)=(0,1)$ |
| Robin | $a,b\neq 0$ on $\partial\Omega_R$ |
| Mixed | Spatially varying $(a(x),b(x))$ on different boundary segments |

This construction differs materially from the Dirichlet-only penalty approach used in TENG++, where
$$
L(\theta)=L_{\rm PDE}(\theta)+\lambda L_{\rm BC}(\theta)
$$
with
$$
L_{\rm PDE}(\theta)=\frac{1}{|\mathcal X_{\rm int}|}\sum_{x_i\in\mathcal X_{\rm int}}
|\mathcal N[u_\theta](x_i,t_n)|^2,
$$
and
$$
L_{\rm BC}(\theta)=\frac{1}{|\mathcal X_{\rm bdry}|}
\sum_{x_j\in\mathcal X_{\rm bdry}}
|u_\theta(x_j,t_n)-g(x_j,t_n)|^2.
$$
TENG++ therefore balances PDE and boundary enforcement through a scalar weight $\lambda$, whereas TENG-BC incorporates the boundary operator directly into the least-squares metric [2512.15771].

## 4. Time integration, linear algebra, and implementation

At the algorithmic level, TENG-BC receives current parameters $\theta^{(t)}$, a time increment $\Delta t$, the operator $L$, boundary data $(a,b,v)$, and the network $u_\theta$. It computes $u_{\rm target}$ by the selected time integrator, then performs a small number of least-squares iterations. In each iteration it samples interior points $\{x_i\}$ and boundary points $\{x_j\}$, computes
$$
\Delta u_i=u_{\rm target}(x_i)-u_\theta(x_i),
$$
and
$$
\Delta v_j=v(x_j)-[a\,u_\theta(x_j)+b\,\partial_n u_\theta(x_j)],
$$
forms the Jacobians, assembles $R=(\Delta u;\Delta v)$ and $\mathcal J=(J;\,aJ_\partial+bK_\partial)$, solves
$$
\Delta\theta=\arg\min_{\Delta\theta}\|R-\mathcal J\Delta\theta\|^2
$$
by, for example, truncated SVD, and updates $\theta\leftarrow\theta+\Delta\theta$ [2603.00397].

The practical details reported for TENG-BC are specific. Sampling uses fixed uniform grids of $N_\Omega$ interior and $N_{\partial\Omega}$ boundary points, with no resampling needed. Only a subset of parameters, for example 512, is active in each least-squares solve to improve conditioning, and the subset may be re-selected each step. Stability is supported by small singular-value truncation in the least-squares solver and double-precision arithmetic. The experiments use $\Delta t=5\times 10^{-4}$. Complexity is described as follows: assembling $\mathcal J$ costs
$$
O((N_\Omega+N_{\partial\Omega})\cdot d_{\rm model}\cdot d_{{\rm param\_active}}),
$$
while the least-squares solve via truncated SVD or QR costs
$$
O(d_{{\rm param\_active}}^2\cdot (N_\Omega+N_{\partial\Omega})).
$$
Overall, the method scales linearly in the number of samples [2603.00397].

The precursor TENG++ presents the same sequential-in-time philosophy in a more classical predictor-corrector language. Its Euler variant forms
$$
u^{\rm pred}(x)=u_\theta(x,t_n)+\Delta t\,\mathcal N[u_\theta](x,t_n),
$$
then fits $u^{\rm pred}$ together with the boundary condition by a few natural-gradient steps. Its Heun variant computes a predictor, evaluates the PDE residual at the predicted state, forms
$$
u^{\rm corr}(x)=u_\theta(x,t_n)+\frac{\Delta t}{2}\Bigl(\mathcal N[u_\theta](x,t_n)+\mathcal N[u^{\rm pred}](x,t_n)\Bigr),
$$
and then performs a second natural-gradient fit [2512.15771]. TENG++ states that Euler is cheap but first-order accurate in $\Delta t$, whereas Heun attains second-order accuracy while doubling the number of PDE-residual evaluations per time step.

## 5. Benchmarks and quantitative performance

The benchmark suite for TENG-BC spans diffusion, transport, and nonlinear PDEs under varied boundary conditions [2603.00397]. For the heat equation $u_t=\kappa\nabla^2u$ on the unit disk with $\kappa=0.1$, the tested boundary conditions include inhomogeneous Dirichlet, zero and nonzero Neumann, Robin $u+\lambda\partial_n u=0$ with $\lambda$ chosen via Bessel identities, and mixed Neumann–Dirichlet on a quarter-annulus. The transport equation
$$
\partial_t u+v(x)\cdot\nabla u=0
$$
is posed on the unit disk with inflow Dirichlet boundary and velocity
$$
v(x)=(\beta x_1-x_2,\beta x_2+x_1),\qquad \beta=0.2.
$$
The viscous Burgers equation
$$
u_t+u\,\nabla\cdot u=\nu\nabla^2u
$$
is tested on $[0,2\pi]^2$ with $\nu=10^{-3}$ and periodic boundaries [2603.00397].

The reported quantitative results are strong. For the heat equation, the full-time relative $L^2$ error at $T=1$ is approximately $2\times 10^{-4}$ for TENG-Euler, approximately $3\times 10^{-6}$ for TENG-Heun, and approximately $3$–$5\times 10^{-6}$ for TENG-RK4. For the transport equation, the per-step error remains $\lesssim 10^{-5}$. For Burgers, relative $L^2$ error is measured against spectral-1024:

| Method | $T=1$ | $T=2$ | $T=3$ | $T=4$ |
|---|---:|---:|---:|---:|
| TENG-Euler | $3.0\mathrm e{-4}$ | $1.7\mathrm e{-3}$ | $1.3\mathrm e{-3}$ | $1.6\mathrm e{-3}$ |
| TENG-Heun | $2.2\mathrm e{-6}$ | $2.5\mathrm e{-5}$ | $1.2\mathrm e{-4}$ | $2.2\mathrm e{-4}$ |
| TENG-RK4 | $2.6\mathrm e{-6}$ | $2.9\mathrm e{-5}$ | $1.4\mathrm e{-4}$ | $3.2\mathrm e{-4}$ |
| FEM | $3.2\mathrm e{-6}$ | $1.1\mathrm e{-3}$ | $4.8\mathrm e{-3}$ | $4.5\mathrm e{-3}$ |
| PINN-BFGS | $1.3\mathrm e{-1}$ | $2.0\mathrm e{-1}$ | $2.1\mathrm e{-1}$ | $1.7\mathrm e{-1}$ |

The paper summarizes these results by stating that TENG-BC variants maintain uniformly low error over time, outperform PINNs, and match or exceed fine-mesh FEM under comparable sampling budgets. It further characterizes the observed accuracy as “solver-level,” with errors at or below $10^{-6}$ for diffusion, approximately $10^{-5}$ for advection, and $10^{-4}$ near shocks in Burgers [2603.00397].

The precursor TENG++ reports a more restricted but informative heat-equation study on the unit disk $\Omega=B(0,1)\subset\mathbb R^2$ with
$$
u_t=\nu\Delta u,\qquad u|_{\partial\Omega}=0,\qquad \nu=1/10,
$$
and initial data given as a linear combination of Bessel-mode eigenfunctions. There, analytical solutions from Bessel expansions enable exact error computation. The stated findings are that TENG-Euler with step-size $h=0.005$ yields errors on the order of $10^{-3}$ by $t=0.8$, while TENG-Heun with the same $h$ achieves errors below $10^{-4}$ up to $t=4.0$; Euler is competitive for low to moderate accuracy demands, whereas Heun strongly outperforms for high precision [2512.15771].

## 6. Limitations, extensions, and interpretation

The limitations stated for TENG-BC are primarily linear-algebraic rather than conceptual. Least-squares solves become larger for very high-dimensional networks. The paper reports that the partial-update trick mitigates this issue, but also notes that further scalable solvers, such as randomized sketching, may be needed for extremely high-dimensional PDEs [2603.00397]. This suggests that the principal bottleneck is not the PDE formulation itself, but the cost and conditioning of repeatedly solving local Jacobian-based least-squares problems.

Several extensions are identified explicitly. TENG-BC lists extension to Cauchy-type and more general boundary-value problems, scalable approximations of the metric $G$ such as low-rank or block-diagonal forms, incorporation of data assimilation or parametric variations for operator learning, applications to coupled multi-physics or stochastic PDEs, and automatic selection of active-parameter subsets [2603.00397]. TENG++ likewise describes natural extensions to Neumann conditions through an additional penalty
$$
L_{\rm NC}(\theta)=\frac{1}{|\mathcal X_{\rm bdry}|}
\sum_{x_j\in\partial\Omega}
|\partial_n u_\theta(x_j)-h(x_j)|^2,
$$
to mixed boundary conditions by combining Dirichlet and Neumann terms on complementary segments, and to nonlinear, coupled, or higher-order PDEs by replacing the diffusion operator with a general $\mathcal N[\cdot]$; it also emphasizes that the networks remain mesh-free on complex geometries [2512.15771].

A recurrent misconception in this area is that general boundary conditions must be handled by manually tuned penalty weights or by carefully engineered trial spaces. The TENG-BC formulation directly contests that view: its discussion states that general boundary types are enforced intrinsically, avoiding penalty-weight tuning or trial-space construction [2603.00397]. Another misconception is that higher-order time integrators necessarily undermine the efficiency of neural solvers. The reported TENG++ trade-off is more nuanced: Heun requires roughly twice the PDE-residual evaluations per time step, but its superior temporal accuracy can allow larger $\Delta t$ for the same error tolerance, offsetting the extra cost [2512.15771].

In conceptual terms, TENG-BC is best understood as a localized time-stepping method in which the network supplies a continuous representation of the evolving field, while the Jacobian-induced metric supplies a geometry-aware correction at every step. The papers attribute its long-time behavior to local natural-gradient corrections that control stepwise error and suppress long-time drift, and its boundary robustness to a unified least-squares formulation that embeds the boundary operator directly into the update [2603.00397].

Source: https://www.emergentmind.com/topics/teng-bc