Dynamical Boundary Constraint (DBC)

• Dynamical Boundary Constraint (DBC) is a framework where boundary conditions evolve dynamically and are integrated into the system’s internal control.
• It is implemented across various fields—using rapid switching in quantum mechanics, constrained PDAE formulations in PDEs, and internal constraints in PINNs—to couple bulk and boundary behavior.
• DBC ensures that boundary conditions actively influence system dynamics, enabling adaptive control, improved numerical stability, and innovative topology-changing experiments.

Searching arXiv for recent and foundational papers on dynamical boundary constraints and related formulations. Dynamical Boundary Constraint (DBC) denotes a class of constructions in which a boundary condition is not treated as a fixed external prescription, but as part of the dynamics, as an explicit constraint, or as an internal control mechanism. In current arXiv usage, the term appears in several technically distinct settings: rapidly switched self-adjoint boundary conditions for quantum particles, dynamical boundary variables in AdS$_2$, bulk–surface PDE systems rewritten as constrained PDAEs, hyperbolic dynamical boundary conditions for wave equations, and internal pseudo-boundary penalties in Physics-Informed Neural Networks (PINNs) (Asorey et al., 2013, Kim et al., 2023, Altmann, 2018, Vitillaro, 2015, Martínez-Esteban et al., 29 Jul 2025). These works suggest a common theme: the boundary is not passive, and the admissible boundary relation is itself subject to evolution, coupling, or optimization.

1. Conceptual range and canonical forms

Across the literature, DBC is implemented in markedly different mathematical languages, but each instance replaces a static boundary prescription by a structure that is evolved, composed, constrained, or enforced dynamically.

Setting	Representative relation	Role of the boundary
Quantum mechanics on an interval	$i(I+U)\varphi'=(I-U)\varphi$, with rapid switching $U\star V$	boundary condition becomes an effective dynamical object
AdS$_2$ boundary mechanics	$\dot T-e^\varphi=0$ and $\dot T=1-\kappa(I_++I_-)$	boundary trajectory is a dynamical variable
Parabolic and hyperbolic PDEs	$p-\gamma u=0$ or $u_{tt}+\partial_\nu u-\Delta_\Gamma u+\cdots=0$	bulk and boundary evolve as a coupled constrained system
PINNs	internal DBC loss at $N_{\text{DBC}}$ points	intermediate pseudo-boundaries reshape optimization

In operator-theoretic quantum models, DBC appears as rapid alternation between self-adjoint boundary conditions and leads to a new effective boundary condition through a dynamical composition law (Asorey et al., 2013). In low-dimensional gravity, the boundary embedding is itself dynamical and obeys a constraint tied to bulk matter fluxes (Kim et al., 2023). In parabolic and wave problems, the boundary can carry its own evolution equation and can be represented explicitly by a Lagrange-multiplier constraint in a PDAE formulation (Altmann, 2018, Vitillaro, 2015). In PINNs, DBC denotes an algorithmic device that inserts internal boundary-like constraints during training in order to suppress trivial or misleading solutions (Martínez-Esteban et al., 29 Jul 2025).

2. Quantum-mechanical and field-theoretic formulations

A concrete realization of DBC is given by a non-relativistic spinless quantum particle in a bounded domain, specialized in detail to $\Omega=[0,1]$, where the free Hamiltonian is $i(I+U)\varphi'=(I-U)\varphi$0 and self-adjoint boundary conditions are parameterized by $i(I+U)\varphi'=(I-U)\varphi$1 through

$i(I+U)\varphi'=(I-U)\varphi$2

This is the Asorey–Ibort–Marmo classification on the interval (Asorey et al., 2013). Dirichlet, Neumann, Robin, mixed Dirichlet–Robin, and pseudo-periodic boundary conditions all arise as particular choices of $i(I+U)\varphi'=(I-U)\varphi$3. Rapid alternation between two self-adjoint extensions $i(I+U)\varphi'=(I-U)\varphi$4 and $i(I+U)\varphi'=(I-U)\varphi$5 is then studied through

$i(I+U)\varphi'=(I-U)\varphi$6

with the effective boundary condition encoded by a commutative and associative product

$i(I+U)\varphi'=(I-U)\varphi$7

The analysis proceeds at the level of quadratic forms rather than operator sums, using the Kato–Lapidus generalization of the Trotter formula (Asorey et al., 2013).

Two structural facts are central. First, the evolution remains unitary; rapid switching does not introduce decoherence. Second, the Dirichlet boundary condition $i(I+U)\varphi'=(I-U)\varphi$8 is absorbing: $i(I+U)\varphi'=(I-U)\varphi$9 More generally, incompatible partially Dirichlet constraints can force the intersection of form domains to collapse to full Dirichlet. A particularly transparent example is rapid switching between two pseudo-periodic boundary conditions with distinct phases $U\star V$0, which yields

$U\star V$1

The paper interprets this as a simple instance of superposition of different topologies, and explores possible implementations with superconducting quantum interference devices (Asorey et al., 2013).

A complementary field-theoretic version treats the boundary condition itself as a dynamical field. For a scalar field $U\star V$2, the general self-adjoint boundary condition can be written as

$U\star V$3

with $U\star V$4 a Hermitian operator on the boundary Hilbert space. In the boundary-action formulation,

$U\star V$5

and $U\star V$6 is promoted to a bilocal dynamical field through an additional action $U\star V$7 (Karabali et al., 2015). Local Robin conditions correspond to $U\star V$8, while time-dependent $U\star V$9 produces radiation from a time-dependent boundary. After integrating out the boundary field $_2$0, the effective action acquires a $_2$1 contribution, so the boundary constraint field develops induced mass and kinetic terms (Karabali et al., 2015). This suggests a DBC interpretation in which the boundary relation is no longer merely imposed but governed by its own effective dynamics.

3. Dynamical boundary constraints in AdS$_2$2

In the AdS$_2$3 setting based on the Almheiri–Polchinski modification of Jackiw–Teitelboim gravity, the boundary is taken to be a moving curve $_2$4, so the boundary time coordinate $_2$5 becomes a dynamical variable (Kim et al., 2023). Matching the dilaton to the Poincaré asymptotics yields

$_2$6

The boundary stress tensor is

$_2$7

and this is identified with the Hamiltonian of the boundary quantum mechanics.

Engelsøy–Mertens–Verlinde introduce the variable $_2$8 and write the boundary Hamiltonian as

$_2$9

The corresponding Lagrangian contains an explicit constraint term,

$\dot T-e^\varphi=0$0

so the DBC is

$\dot T-e^\varphi=0$1

Geometrically, this enforces that the boundary embedding be compatible with AdS$\dot T-e^\varphi=0$2 asymptotics; algebraically, it ties the boundary variables $\dot T-e^\varphi=0$3 and $\dot T-e^\varphi=0$4 in the one-dimensional theory (Kim et al., 2023).

Dirac analysis yields two primary constraints,

$\dot T-e^\varphi=0$5

with no secondary constraints, and the Dirac matrix is invertible, so the system is fully second-class. The Batalin–Tyutin extension introduces auxiliary fields $\dot T-e^\varphi=0$6 and replaces these by first-class constraints

$\dot T-e^\varphi=0$7

together with an involutive Hamiltonian and a Wess–Zumino-like term

$\dot T-e^\varphi=0$8

In unitary gauge, the BT-extended system reduces to the original second-class system. The authors emphasize an open holographic question: whether a well-defined extended bulk theory corresponding to the extended boundary theory exists (Kim et al., 2023).

4. Constrained variational principles and holographic boundary data

In constrained mechanical systems, boundary conditions are themselves constrained by the Hamiltonian structure. A naive variational principle that fixes all configuration variables at the boundary can impose more conditions than there are physical degrees of freedom, so extrema of the action need not coincide with solutions of the Dirac-constrained dynamics (Izumi et al., 2023). The proposed remedy is a Hamiltonian procedure: identify all constraints, perform a canonical transformation to variables

$\dot T-e^\varphi=0$9

rewrite the boundary term as

$\dot T=1-\kappa(I_++I_-)$0

add the counterterm $\dot T=1-\kappa(I_++I_-)$1, and fix only the physical variables $\dot T=1-\kappa(I_++I_-)$2 at the boundary (Izumi et al., 2023). In this sense, the admissible boundary data are dictated by the dynamical constraint structure rather than selected independently.

A related but terminologically distinct use appears in AdS/BCFT, where the end-of-the-world brane $\dot T=1-\kappa(I_++I_-)$3 admits conformal boundary condition (CBC), Dirichlet boundary condition (DBC), and Neumann boundary condition (NBC). Here DBC means Dirichlet boundary condition, not dynamical boundary constraint (Chu et al., 2021). The conformal boundary condition fixes the conformal class of the induced metric and the trace of the extrinsic curvature,

$\dot T=1-\kappa(I_++I_-)$4

while Dirichlet sets

$\dot T=1-\kappa(I_++I_-)$5

and Neumann imposes

$\dot T=1-\kappa(I_++I_-)$6

The perturbation analysis shows that CBC and DBC are interpreted as fluctuations of the extrinsic curvature of $\dot T=1-\kappa(I_++I_-)$7, whereas NBC is interpreted as fluctuation of the induced metric of $\dot T=1-\kappa(I_++I_-)$8. In all cases, the fluctuation modes are massive, and the boundary central charges are exactly the same for CBC and DBC (Chu et al., 2021). This usage is adjacent to DBC in the dynamical-constraint sense because boundary data again determine which geometric variables propagate, but the acronym denotes a different object.

5. PDE, PDAE, and hyperbolic dynamic boundary conditions

In parabolic problems, dynamic boundary conditions can be formulated as bulk–surface systems with an explicit coupling constraint. For

$\dot T=1-\kappa(I_++I_-)$9

the dynamic boundary equation is written by introducing a boundary variable $p-\gamma u=0$0. For the nonlocal case,

$p-\gamma u=0$1

The condition $p-\gamma u=0$2 is the boundary coupling condition, and it is incorporated explicitly as a constraint with a Lagrange multiplier in the PDAE system

$p-\gamma u=0$3

Well-posedness follows from inf-sup stability of the coupling operator and a Gårding inequality on the constraint kernel, using the operator differential-algebraic framework of Emmrich–Mehrmann (Altmann, 2018). This formulation makes the boundary constraint explicit rather than building it into the ansatz space.

For wave equations, the boundary can itself carry a hyperbolic evolution law. The model studied on a bounded domain $p-\gamma u=0$4 with boundary decomposition $p-\gamma u=0$5 is

$p-\gamma u=0$6

$p-\gamma u=0$7

$p-\gamma u=0$8

The boundary term $p-\gamma u=0$9 makes the boundary condition itself hyperbolic, so the trace on $u_{tt}+\partial_\nu u-\Delta_\Gamma u+\cdots=0$0 is a genuine dynamical variable (Vitillaro, 2015). The natural phase space is $u_{tt}+\partial_\nu u-\Delta_\Gamma u+\cdots=0$1, with energy containing both interior and boundary kinetic and potential contributions. The paper proves local Hadamard well-posedness, a blow-up alternative, and an energy identity. When damping is linear and the source terms are superlinear in the sense

$u_{tt}+\partial_\nu u-\Delta_\Gamma u+\cdots=0$2

negative-energy initial data yield finite-time blow-up. When the sources are at most linear at infinity, or are dominated by the damping terms, the solution is global and defines a dynamical system (Vitillaro, 2015). In the PDE literature, this is the most literal use of DBC: the boundary obeys its own evolution equation and is coupled to the bulk by normal flux and tangential operators.

6. DBC as an optimization constraint in PINNs

In PINNs, DBC is introduced as a training strategy rather than a physical boundary law. The method adds internal boundary-like constraints at selected points inside the computational domain in order to prevent convergence to trivial or misleading solutions, especially for oscillatory and multi-scale problems (Martínez-Esteban et al., 29 Jul 2025). The total loss takes the form

$u_{tt}+\partial_\nu u-\Delta_\Gamma u+\cdots=0$3

with DBC points

$u_{tt}+\partial_\nu u-\Delta_\Gamma u+\cdots=0$4

placed at internal locations. A generic DBC term is

$u_{tt}+\partial_\nu u-\Delta_\Gamma u+\cdots=0$5

Two training organizations are distinguished: an individual subinterval method, where the domain is split into intervals separated by DBC points, and a cumulative method, where previously imposed DBCs remain active as new ones are added (Martínez-Esteban et al., 29 Jul 2025). In the harmonic oscillator example, the ODE loss $u_{tt}+\partial_\nu u-\Delta_\Gamma u+\cdots=0$6 is evaluated at $u_{tt}+\partial_\nu u-\Delta_\Gamma u+\cdots=0$7 training points and on a validation set with $u_{tt}+\partial_\nu u-\Delta_\Gamma u+\cdots=0$8, and cases $u_{tt}+\partial_\nu u-\Delta_\Gamma u+\cdots=0$9 are compared over ten independent trainings. The box-plot of MAE shows that the median MAE decreases as $N_{\text{DBC}}$0 increases. In the light-trajectory example around a black hole, reference trajectories are computed by a Runge–Kutta scheme, DBCs are placed at several radii, and increasing $N_{\text{DBC}}$1 leads to smaller MAE and more consistent performance (Martínez-Esteban et al., 29 Jul 2025).

This usage is algorithmic rather than operator-theoretic. The “boundary” is internal to the training protocol, but the formal role is parallel: the solution is anchored not only by the physical boundary and differential residual, but also by dynamically enforced interior constraints.

7. Terminological ambiguity and disambiguation

DBC is not a uniform acronym across arXiv. In "Analysis of Generalizability of Deep Neural Networks Based on the Complexity of Decision Boundary," DBC means "Decision Boundary Complexity," a scalar score defined from the normalized Shannon entropy of eigenvalues of adversarial points sampled on or near a classifier’s decision boundary: $N_{\text{DBC}}$2 That construction is explicitly unrelated to dynamical boundary constraints; it measures decision-boundary complexity as a proxy for generalization in DNNs (Guan et al., 2020).

A second ambiguity appears in AdS/BCFT, where DBC abbreviates "Dirichlet boundary condition" in contrast to CBC and NBC (Chu et al., 2021). In that context, DBC refers to the fixed induced metric condition $N_{\text{DBC}}$3, not to a dynamical constraint on the boundary itself.

Accordingly, the term “Dynamical Boundary Constraint” is best read contextually. In quantum mechanics and field theory it concerns self-adjoint extensions, bilocal boundary kernels, and topology-changing composition laws; in gravity it refers to boundary embedding variables and constraint conversion; in PDE analysis it denotes genuine boundary evolution equations or explicit coupling constraints; and in PINNs it names an internal loss mechanism that dynamically inserts boundary-like anchors into the training process (Asorey et al., 2013, Kim et al., 2023, Altmann, 2018, Martínez-Esteban et al., 29 Jul 2025).