Time-dependent P1 and the telegrapher equation#

P1 is an angular approximation. Steady state is a separate assumption. Keeping time derivatives in the first two P1 moment equations produces a system for fluence and current that can be written as a telegrapher equation. Neglecting the time derivative of current produces the usual time-dependent diffusion equation.

Important

TransientP1 solves the coupled time-dependent P1 equations on finite samples. It supports monochromatic surface sources, a sampled pulse, and time-resolved contact-detector output. Frequency-domain modulation and a time-dependent diffusion solver are not implemented.

Adding time to radiative transfer#

Let \(v\) be the energy propagation speed in a homogeneous, nondispersive medium. The common approximation \(v=c_0/n\) uses the vacuum light speed \(c_0\) and refractive index \(n\). For dispersive media, the appropriate energy or group speed needs a separate physical choice; a steady refractive-index parameter does not settle that question.

The time-dependent transport equation is

\[\frac{1}{v}\frac{\partial L}{\partial t} +\mathbf s\cdot\nabla L+(\mu_a+\mu_s)L =\mu_s\int_{4\pi}p(\mathbf s,\mathbf s')L(\mathbf s',t)\,d\Omega' +Q.\]

The angular moments use the same definitions as in Transport theory and approximations, now with time as an additional argument. For constant \(v\), the radiation energy density is \(u=\Phi/v\), with units \(\mathrm{J\,m^{-3}}\). Consequently \((1/v)\partial_t\Phi\) represents a local rate of energy storage rather than a loss mechanism.

For an isotropic volume source, \(\mathbf q_1=\mathbf 0\). Define \(\mu_{\mathrm{tr}}=\mu_a+\mu_s'\). Applying the P1 closure gives

\[\frac{1}{v}\partial_t\Phi+\nabla\cdot\mathbf J+\mu_a\Phi=q_0, \qquad \frac{1}{v}\partial_t\mathbf J+\frac{1}{3}\nabla\Phi +\mu_{\mathrm{tr}}\mathbf J=\mathbf 0.\]

The first equation conserves radiant energy. The second describes how the directional current changes: a fluence gradient drives flow, while absorption and transport scattering relax its directional memory. Derivations retaining these two time derivatives are presented in the radiative transfer P1 study hosted by OSTI.

Current relaxation and the diffusion approximation#

Dividing the current equation by \(\mu_{\mathrm{tr}}\) and defining \(D=1/(3\mu_{\mathrm{tr}})\) gives

\[\tau_J\partial_t\mathbf J+\mathbf J=-D\nabla\Phi, \qquad \tau_J=\frac{1}{v\mu_{\mathrm{tr}}}.\]

The current relaxation time \(\tau_J\) is the time to travel one transport mean free path. Current does not adjust instantly to a new gradient in this system. For an approximately fixed local gradient, the homogeneous part of its relaxation decays as \(\exp(-t/\tau_J)\).

If variations occur over times \(T\) for which \(\tau_J/T\ll1\), dropping the current time derivative gives \(\mathbf J\approx-D\nabla\Phi\). Substituting into energy conservation then yields the parabolic diffusion equation

\[\frac{1}{v}\partial_t\Phi -\nabla\cdot(D\nabla\Phi)+\mu_a\Phi=q_0.\]

There are two common diffusion-coefficient conventions:

\[D=\frac{1}{3\mu_{\mathrm{tr}}}\quad[\mathrm m], \qquad \mathcal D=vD\quad[\mathrm{m^2\,s^{-1}}].\]

Multiplying the whole diffusion equation by \(v\) gives \(\partial_t\Phi-\nabla\cdot(\mathcal D\nabla\Phi) +v\mu_a\Phi=vq_0\) when \(v\) is spatially constant. MomentRT uses the first convention, so its steady \(D\) must not be inserted as a diffusivity in \(\mathrm{m^2\,s^{-1}}\).

Eliminating current without dropping its time derivative#

The scalar derivation below assumes spatially and temporally constant \(v\), \(\mu_a\), and \(\mu_{\mathrm{tr}}\), and an isotropic source. With heterogeneous coefficients, spatial derivatives of those coefficients must be accounted for; the coupled moment equations are the safer starting point.

Differentiate energy conservation with respect to time:

\[\frac{1}{v}\partial_t^2\Phi +\nabla\cdot\partial_t\mathbf J +\mu_a\partial_t\Phi=\partial_t q_0.\]

Take the divergence of the current equation:

\[\nabla\cdot\partial_t\mathbf J =-\frac{v}{3}\nabla^2\Phi -v\mu_{\mathrm{tr}}\nabla\cdot\mathbf J.\]

Use the original conservation law to replace \(\nabla\cdot\mathbf J=q_0-\mu_a\Phi-(1/v)\partial_t\Phi\). After collecting terms and multiplying by \(v\), the result is

\[\boxed{ \partial_t^2\Phi +v(\mu_{\mathrm{tr}}+\mu_a)\partial_t\Phi -\frac{v^2}{3}\nabla^2\Phi +v^2\mu_{\mathrm{tr}}\mu_a\Phi =v\partial_tq_0+v^2\mu_{\mathrm{tr}}q_0. }\]

Both absorption contributions must be retained: absorption appears in energy conservation and in the loss of directional current. This is why the damping coefficient contains \(\mu_{\mathrm{tr}}+\mu_a\). The derivative of the source also follows from the elimination; it cannot be omitted for an arbitrary time-varying source.

In a nonabsorbing source-free homogeneous region, this simplifies to

\[\tau_J\partial_t^2\Phi+\partial_t\Phi =\mathcal D\nabla^2\Phi.\]

This is the classical damped-wave, or telegrapher, form. It is related to the equations for signals on electrical transmission lines through its mathematical structure. It describes propagation and relaxation of an intensity distribution, not electromagnetic phase oscillations.

Propagation speed and validity#

For the homogeneous P1 system, the characteristic propagation speed is

\[v_{\mathrm{P1}}=\sqrt{\frac{\mathcal D}{\tau_J}} =\frac{v}{\sqrt{3}}.\]

The finite front speed corrects the mathematical instantaneous tails of a parabolic diffusion solution. It is still smaller than the transport speed \(v\) of unscattered photons. Retaining current inertia therefore does not make P1 an accurate ballistic or early-arrival model. The angular closure still suppresses strongly directional structure. The difference between diffusion, conventional P1, and alternative transient closures is examined in Diffusion, P1, and other approximate forms of radiation transport.

For times much longer than current relaxation, diffusion can describe the slowly varying part of the solution. With nonzero absorption, it is clearer to take that approximation in the coupled current equation: merely deleting the second derivative from the general scalar equation can leave an extra absorption-dependent factor on \(\partial_t\Phi\).

Initial and boundary data#

The coupled transient system requires initial fluence and current. A scalar second-order equation instead requires initial fluence and its initial time derivative, constrained by conservation:

\[\partial_t\Phi(\mathbf r,0) =v\left[q_0(\mathbf r,0)-\mu_a\Phi(\mathbf r,0) -\nabla\cdot\mathbf J(\mathbf r,0)\right].\]

An arbitrary choice of both scalar initial data may not correspond to the intended initial angular field. A transient boundary model must also be specified. The partial-current reasoning in Finite-sample P1 solver can be applied to instantaneous P1 moments, but extending the solver requires a consistent treatment of time-dependent sources, interfaces, and boundary conditions rather than adding one time derivative to the steady matrix.

Finite-element implementation#

TransientP1 solves the coupled equations above rather than applying the homogeneous scalar telegrapher equation across a heterogeneous sample. The fluence uses continuous linear tetrahedral nodal basis functions, while the three current components are constant within each tetrahedron. Optical coefficients are sampled at element centroids; they may vary between elements. The discrete current is not smoothed across interfaces. Internal refractive indices must match and the energy speed is an explicit uniform input.

The instantaneous outward P1 partial-current condition on a Robin face is

\[\mathbf J\cdot\mathbf n = \beta\Phi - \frac{2I_\mathrm{in}}{1+R}, \qquad \beta=\frac{1-R}{2(1+R)}.\]

Here \(R\) is the supplied diffuse effective reflectance and \(I_\mathrm{in}\) is diffuse incoming partial current already coupled into the sample. Reflecting faces have zero normal current. The face term is inserted in the weak conservation equation at each time step; it is not a normal-incidence Fresnel model. Boundary escaping irradiance and contact detectors retain the existing Lambertian angular approximation.

The method starts with zero fluence and current. A dimensionless nonnegative power_profile multiplies the monochromatic source power at the supplied times and must start at zero. It is linearly interpolated over each time interval. Implicit midpoint time integration advances both P1 moments together; the element current can be eliminated locally for each step, leaving one sparse nodal solve. The time step may vary. Midpoint is second-order for smooth solutions but does not suppress all rapidly oscillating numerical modes; resolve the pulse and current-relaxation timescale, and inspect signed results.

The returned TransientResult contains times and per-time frames with nodal fluence, element-centered current, and boundary fields. A simulation with named detectors also provides detector_power arrays. It reports incoming, absorbed, and escaping powers and stored energy at each time; the interval_energy_residual checks midpoint energy conservation. Negative values are not clipped. Storing every frame costs memory proportional to the number of time samples and mesh elements.

For each interval, the reported residual is

\[\frac{U_{n+1}-U_n}{t_{n+1}-t_n} +\overline{P_\mathrm{abs}}+\overline{P_\mathrm{escape}} -\overline{P_\mathrm{in}}, \qquad U=\int_\Omega \frac{\Phi}{v}\,dV,\]

where overbars denote endpoint averages. The difference between escaping and incoming power is the signed net outward current. Detector-aperture power is reported separately and is not part of this whole-boundary energy balance.

Use the pulsed layered sample example for a complete calculation. The solver handles one monochromatic source configuration at a time; spectral sweeps, broadband integration, nonzero initial fields, and the steady-state detector-convergence helper are not implemented for transient results. Refine the mesh, source and detector footprints, and time grid independently. Compare against an appropriate transport reference for the intended geometry: energy conservation or convergence of the discrete P1 equations alone does not prove P1 is accurate for early-time or near-source light transport.

Steady state and modulation#

Setting all time derivatives to zero recovers

\[\mathbf J=-D\nabla\Phi,\qquad -\nabla\cdot(D\nabla\Phi)+\mu_a\Phi=q_0.\]

This is the implemented bulk model. In steady state there is no additional telegrapher time scale for the solver to resolve.

For reference, with the harmonic convention \(e^{i\omega t}\), the full homogeneous P1 current law becomes

\[\widehat{\mathbf J}=-D_\omega\nabla\widehat\Phi,\qquad D_\omega=\frac{1}{3(\mu_{\mathrm{tr}}+i\omega/v)}.\]

Its fluence equation is

\[-\nabla\cdot(D_\omega\nabla\widehat\Phi) +(\mu_a+i\omega/v)\widehat\Phi=\widehat q_0.\]

The approximation \(|\omega|\tau_J\ll1\) replaces \(D_\omega\) by the real steady coefficient \(D\). This is the distinction between frequency-domain P1 and frequency-domain diffusion. These complex equations are explanatory; the current real-valued steady solver does not implement them.