Laplace Guidance Field: Theory & Applications

• Laplace guidance field is a harmonic extension technique that uses Laplace’s equation to convert prescribed boundary or planar data into an interior field.
• In RF trap design, odd harmonic extension generates field-free guides by extending analytic functions, yielding precise RF null sets for various geometries.
• In robotic navigation, Laplace guidance fields create risk-aware and semantically modulated vector fields that enforce safe behavior near obstacles.

Laplace guidance field denotes a class of constructions in which Laplace’s equation is used to propagate prescribed data into a harmonic interior field that encodes guidance structure. In the literature considered here, the phrase appears in two distinct but related settings. In radio-frequency trap design, an odd harmonic extension maps an analytic generating function $P(x,y)$ on a symmetry plane to a three-dimensional potential whose in-plane radio-frequency null set is exactly $P(x,y)=0$ (Wesenberg, 4 May 2026). In risk-aware and semantically aware robot navigation, a vector field $v$ is obtained by solving a Dirichlet problem for Laplace’s equation with boundary values tied to obstacle normals, and in Safe-SAGE this construction is further augmented by an inner tangential interface and coupled to a Poisson safety function (Bahati et al., 29 Oct 2025, Yang et al., 5 Mar 2026).

1. Shared harmonic structure and terminological scope

All three works use Laplace’s equation as the governing PDE, but they prescribe different data and target different physical or algorithmic objects. In one case the unknown is a scalar RF potential $\Phi$ in a source-free region containing the plane $z=0$; in the others the unknown is a vector guidance field $v$ over a free-space domain $\Omega$ (Wesenberg, 4 May 2026, Bahati et al., 29 Oct 2025, Yang et al., 5 Mar 2026).

Context	Unknown	Prescribed data
RF trap networks	$\Phi(x,y,z)$	Planar Cauchy data $\Phi_0(x,y)$ and $\Psi_0(x,y)=\partial_z\Phi(x,y,0)$
Risk-aware safety filters	$P(x,y)=0$0	Boundary data $P(x,y)=0$1 on $P(x,y)=0$2
Safe-SAGE	$P(x,y)=0$3	$P(x,y)=0$4 on $P(x,y)=0$5, $P(x,y)=0$6 on $P(x,y)=0$7

The common mathematical motif is harmonic continuation. In the RF setting, analytic planar data determine a local three-dimensional continuation. In the navigation setting, boundary values on obstacle surfaces determine a smooth interior field whose direction and magnitude encode repulsion, conservatism, and, in Safe-SAGE, passing-side conventions. This suggests that “Laplace guidance field” functions less as the name of a single canonical object than as a family of Laplace-based guidance constructions.

2. Odd harmonic extension for field-free RF guides

For field-free RF trap networks, the starting point is a quasi-static RF potential $P(x,y)=0$8 satisfying

$P(x,y)=0$9

On the symmetry plane $v$0, one may prescribe two arbitrary analytic functions: the in-plane potential $v$1 and the normal derivative $v$2. Writing the Taylor expansion

$v$3

with $v$4, and using $v$5 implies $v$6, one obtains the decomposition

$v$7

where

$v$8

and

$v$9

In boundary-value language, $\Phi$0 enforces the Dirichlet data $\Phi$1, and $\Phi$2 enforces the Neumann data $\Phi$3 (Wesenberg, 4 May 2026).

To build a trap network whose RF field vanishes exactly on $\Phi$4 in the plane, one sets

$\Phi$5

The resulting odd extension is

$\Phi$6

By construction, $\Phi$7, $\Phi$8, and $\Phi$9. Hence on $z=0$0 the full RF field $z=0$1 has only an $z=0$2-component proportional to $z=0$3, and the in-plane null set $z=0$4 is carried into three-space as an RF-free “guide” (Wesenberg, 4 May 2026).

The same construction has a Fourier representation. If

$z=0$5

then

$z=0$6

with $z=0$7. An equivalent form is the real part of an “evanescent-wave” superposition (Wesenberg, 4 May 2026).

3. Geometric repertoire and design-space parametrization in trap networks

The odd-extension construction yields explicit field-free guide networks beyond smooth straight-line intersections, including cusp guides, cotangential contacts, and periodic lattices. For polynomial generators, the series can truncate because repeated application of $z=0$8 eventually vanishes (Wesenberg, 4 May 2026).

For cusp guides, the polynomial generator is

$z=0$9

Its zero set is the semi-cubic parabola $v$0. Since $v$1 and $v$2, the series truncates after two steps:

$v$3

At $v$4, $v$5 vanishes iff $v$6, giving the cusp guide (Wesenberg, 4 May 2026).

For cotangential contacts, one takes

$v$7

Then $v$8, $v$9, and $\Omega$0, so

$\Omega$1

At $\Omega$2 the null curves are $\Omega$3, two parabolas tangent to the $\Omega$4-axis, realizing a field-free “cotangential contact” (Wesenberg, 4 May 2026).

Periodic networks arise when $\Omega$5 is jointly periodic with periods $\Omega$6 and is expanded as

$\Omega$7

The odd extension becomes

$\Omega$8

which is the unique harmonic function vanishing at $\Omega$9 with $\Phi(x,y,z)$0. The in-plane RF-null set remains the contour $\Phi(x,y,z)$1 (Wesenberg, 4 May 2026).

A square-lattice family with tunable angle $\Phi(x,y,z)$2 and rounding $\Phi(x,y,z)$3 is obtained by defining

$\Phi(x,y,z)$4

and then

$\Phi(x,y,z)$5

Its zero-contours form a square-grid network of guides crossing at angle $\Phi(x,y,z)$6, with $\Phi(x,y,z)$7 controlling how sharply or smoothly the four branches join. The paper further states several general principles: any analytic $\Phi(x,y,z)$8 immediately yields a three-dimensional harmonic field $\Phi(x,y,z)$9 whose in-plane RF-null set is exactly $\Phi_0(x,y)$0; local intersection geometry is read off directly from the factorization or low-order Taylor expansion of $\Phi_0(x,y)$1 near its zeroes; smoothness and analytic-continuation require $\Phi_0(x,y)$2 to be analytic; and intersections sharper than cusps are excluded. Because the full field is known in closed form, one can compute ponderomotive potentials, Hessians, and transport barriers algebraically, providing a compact, algebraic design-space for RF trap networks in QCCD architectures (Wesenberg, 4 May 2026).

4. Risk-aware safety filters and tunable flux boundary data

In risk-aware safety filtering, the Laplace guidance field is a vector field synthesized on the free-space region in which the robot moves. The domain is $\Phi_0(x,y)$3, described as the open, bounded, connected free-space region, and the boundary $\Phi_0(x,y)$4 is the union of obstacle surfaces. On each boundary point $\Phi_0(x,y)$5, the outward unit normal $\Phi_0(x,y)$6 is known, and a user-specified scalar $\Phi_0(x,y)$7 encodes the desired negative flux magnitude at each boundary point. This $\Phi_0(x,y)$8 is chosen by mapping semantic, probabilistic or dynamic features of obstacles into a $\Phi_0(x,y)$9 risk value and then onto negative flux magnitudes (Bahati et al., 29 Oct 2025).

The guidance field $\Psi_0(x,y)=\partial_z\Phi(x,y,0)$0 is defined componentwise by

$\Psi_0(x,y)=\partial_z\Phi(x,y,0)$1

with boundary condition

$\Psi_0(x,y)=\partial_z\Phi(x,y,0)$2

Equivalently,

$\Psi_0(x,y)=\partial_z\Phi(x,y,0)$3

Because $\Psi_0(x,y)=\partial_z\Phi(x,y,0)$4, $\Psi_0(x,y)=\partial_z\Phi(x,y,0)$5 points normally inward, towards the interior of $\Psi_0(x,y)=\partial_z\Phi(x,y,0)$6, and $\Psi_0(x,y)=\partial_z\Phi(x,y,0)$7 on $\Psi_0(x,y)=\partial_z\Phi(x,y,0)$8. The paper describes this as a Dirichlet problem on the vector field whose boundary prescription simultaneously enforces a Neumann-type condition on any potential function’s gradient through $\Psi_0(x,y)=\partial_z\Phi(x,y,0)$9 (Bahati et al., 29 Oct 2025).

The interpretation is explicitly risk-aware. At $P(x,y)=0$00, $P(x,y)=0$01 is colinear with the inward normal, so it pushes trajectories away from the obstacle. Inside $P(x,y)=0$02, the harmonic extension smoothly blends these normal directions into a globally continuous repulsive field. Larger $P(x,y)=0$03 implies stronger repulsion and more cautious behavior, and regions of higher-flux boundary produce larger interior magnitudes, so the robot feels obstacles with higher risk from farther away (Bahati et al., 29 Oct 2025).

The same work couples the guidance field to a Poisson safety function $P(x,y)=0$04, obtained from

$P(x,y)=0$05

so that $P(x,y)=0$06 is the safe set and $P(x,y)=0$07. For a first-order system $P(x,y)=0$08, the risk-aware safety filter replaces the usual $P(x,y)=0$09 term in the control barrier formulation with the guidance field directly, using the quadratic program

$P(x,y)=0$10

with $P(x,y)=0$11. The forward-invariance proof proceeds by noting that $P(x,y)=0$12 implies $P(x,y)=0$13 on $P(x,y)=0$14, since $P(x,y)=0$15 there (Bahati et al., 29 Oct 2025).

The examples emphasize how the boundary scaling changes activation geometry. With three circular obstacles and three different boundary flux scalings $P(x,y)=0$16, the zero-level set of the QP activation function $P(x,y)=0$17 shifts outward as $P(x,y)=0$18 increases. For a moving obstacle whose speed $P(x,y)=0$19 determines $P(x,y)=0$20, increasing speed produces a larger, asymmetric activation region in the direction of motion. In a scene segmented by YOLO into wall, chair, and person classes with priorities $P(x,y)=0$21, $P(x,y)=0$22, and $P(x,y)=0$23, exponential risk-mapping and linear interpolation to $P(x,y)=0$24 produce small activation zones around walls, medium around chairs, and very large around humans (Bahati et al., 29 Oct 2025).

5. Safe-SAGE: social-semantic adaptive guidance in two dimensions

Safe-SAGE generalizes the navigation setting by introducing a two-layer boundary structure and explicitly linking semantic perception to safety-critical control. Its motivation is that traditional safety-critical control methods, such as control barrier functions, suffer from semantic blindness, exhibiting the same behavior around obstacles regardless of contextual significance (Yang et al., 5 Mar 2026).

The free-space domain is $P(x,y)=0$25, with smooth outer boundary $P(x,y)=0$26, the union of all obstacle surfaces. An inner social interface $P(x,y)=0$27 is introduced by buffering $P(x,y)=0$28 inward by a distance $P(x,y)=0$29:

$P(x,y)=0$30

On each connected component of $P(x,y)=0$31, the framework prescribes purely normal flux of class-dependent magnitude $P(x,y)=0$32; on $P(x,y)=0$33 it prescribes purely tangential flow of magnitude $P(x,y)=0$34. The guidance field $P(x,y)=0$35 is then the unique solution of the vector Dirichlet problem

$P(x,y)=0$36

with

$P(x,y)=0$37

and

$P(x,y)=0$38

Here $P(x,y)=0$39 is the outward unit normal on $P(x,y)=0$40, $P(x,y)=0$41 the unit tangent on $P(x,y)=0$42, and $P(x,y)=0$43 are smooth, class-dependent scalar functions (Yang et al., 5 Mar 2026).

The numerical scheme is specified on a uniform Cartesian grid of spacing $P(x,y)=0$44. For interior nodes, the standard five-point finite-difference approximation is imposed:

$P(x,y)=0$45

On nodes adjacent to $P(x,y)=0$46 or $P(x,y)=0$47, $P(x,y)=0$48 is overwritten by the prescribed Dirichlet data $P(x,y)=0$49 or $P(x,y)=0$50. Collecting all unknowns into a vector $P(x,y)=0$51 yields a sparse linear system

$P(x,y)=0$52

with rows replaced by boundary rows wherever a node lies on $P(x,y)=0$53 or $P(x,y)=0$54. The field can then be solved with a standard sparse-linear solver such as conjugate-gradient or a direct Cholesky factorization (Yang et al., 5 Mar 2026).

Safe-SAGE modulates the Poisson safety function by replacing the unmodulated forcing term with the divergence of the guidance field. The unmodulated Poisson scheme solves

$P(x,y)=0$55

for a negative forcing term $P(x,y)=0$56. Safe-SAGE instead sets

$P(x,y)=0$57

and solves

$P(x,y)=0$58

By Hopf’s lemma, on $P(x,y)=0$59 the gradient of any positive-inside Poisson solution satisfies $P(x,y)=0$60 and $P(x,y)=0$61, while $P(x,y)=0$62 with magnitude $P(x,y)=0$63. Thus locally near an obstacle of class $P(x,y)=0$64,

$P(x,y)=0$65

so larger $P(x,y)=0$66 produces a steeper ascent of $P(x,y)=0$67 away from that obstacle, that is, a larger semantic safety margin (Yang et al., 5 Mar 2026).

The full framework combines perception and filtering. The environment is perceived by fusing multi-sensor point clouds with vision-based instance segmentation and persistent object tracking to maintain up-to-date semantics beyond the camera’s field of view. A multi-layer safety filter consisting of both a model predictive control layer and a control barrier function layer uses the Poisson safety function and flux modulation of the guidance field to introduce varying levels of conservatism and multi-agent passing norms for different obstacles in the environment (Yang et al., 5 Mar 2026).

6. Comparative interpretation, limitations, and recurring misconceptions

A recurring misconception is that the phrase “Laplace guidance field” names a single boundary-value problem. The cited works show a more differentiated picture. In field-free RF trap networks, the construction is a scalar harmonic continuation from planar Cauchy data, and the key design object is the analytic generating function $P(x,y)=0$68 whose zero set prescribes the in-plane RF-null geometry (Wesenberg, 4 May 2026). In the safety-filter literature, the construction is a vector Dirichlet problem on a free-space domain, with boundary data specified directly from obstacle normals and, in Safe-SAGE, from both normals and tangents (Bahati et al., 29 Oct 2025, Yang et al., 5 Mar 2026).

The design semantics also differ. In the RF setting, the object of interest is the field-free guide itself: a guide line, cusp, cotangential contact, or periodic lattice obtained as the null set of the harmonic extension. Local intersection geometry is read off directly from the factorization or low-order Taylor expansion of $P(x,y)=0$69 near its zeroes, and periodic networks arise by insisting that $P(x,y)=0$70 be doubly periodic (Wesenberg, 4 May 2026). In the navigation setting, the object of interest is not a null set but a repulsive or socially biased vector field whose magnitude is tunable through boundary flux parameters $P(x,y)=0$71 or $P(x,y)=0$72 and, in Safe-SAGE, through the tangential bias $P(x,y)=0$73 (Bahati et al., 29 Oct 2025, Yang et al., 5 Mar 2026).

The analytic requirements are likewise domain-specific. For trap networks, smoothness and analytic-continuation require $P(x,y)=0$74 to be analytic, and intersections sharper than cusps are excluded (Wesenberg, 4 May 2026). For risk-aware safety filters, the existence narrative is tied to elliptic theory: under mild regularity, specifically $P(x,y)=0$75, classical elliptic theory guarantees a unique smooth solution $P(x,y)=0$76, and convergence is second-order in the grid resolution $P(x,y)=0$77 for standard five- or seven-point stencils (Bahati et al., 29 Oct 2025).

Taken together, these works indicate a common operational principle: prescribed boundary or planar data are transformed by a harmonic extension into an interior field that is then interpreted as guidance. In one branch this yields a compact parametrization for the design space for QCCD architectures; in the other it yields risk-aware or semantically aware safety margins and passing behavior for robotic systems in dynamic environments (Wesenberg, 4 May 2026, Yang et al., 5 Mar 2026).