Attenuation Law: Concepts & Applications

• Attenuation law is a quantitative relation that defines how radiation, wave amplitude, or transmitted power decreases with propagation through a medium.
• In astrophysics, it explains the effective dimming of starlight in galaxies by integrating factors like dust composition, scattering, and geometry.
• In acoustics, photonic transport, and viscoelasticity, attenuation laws provide predictive models for frequency and length-dependent losses, enhancing material analysis.

Searching arXiv for recent/relevant papers on attenuation law across major domains. An attenuation law is a quantitative relation that specifies how radiation, wave amplitude, or transmitted power is reduced by propagation through a medium. The term is strongly context dependent. In galaxy studies it denotes the effective wavelength-dependent dimming of integrated starlight, $A_\lambda$, after dust composition, scattering, and star–dust geometry have all been folded together (Salim et al., 2020). In disordered photonic transport it often denotes the dependence of $\langle \ln T\rangle$ on device length $L$ (Baron et al., 2011). In acoustics it frequently denotes a frequency law such as $\alpha(f)=\alpha_0 f^\gamma$ (Jiménez et al., 2014). Across these domains, the same phrase therefore refers not to a single formula, but to a family of constitutive, radiative-transfer, and effective-medium relations.

1. General mathematical forms

Several canonical attenuation laws recur across disciplines, but they describe different observables and should not be conflated.

Domain	Representative quantity	Representative law
Galaxy dust attenuation	$A_\lambda$	$A_\lambda = m_\lambda - m_{\lambda,0}$
Waveguide transport	$\langle \ln T\rangle$	$\langle \ln T\rangle = -\alpha L$
Acoustic attenuation	$\alpha(f)$	$\alpha(f)=\alpha_0 f^\gamma$
Viscoelastic wave propagation	$\langle \ln T\rangle$0	$\langle \ln T\rangle$1

In the astrophysical notation reviewed by Salim and Narayanan, attenuation is defined by

$\langle \ln T\rangle$2

with common normalizations

$\langle \ln T\rangle$3

and a useful UV–optical slope parameter

$\langle \ln T\rangle$4

(Salim et al., 2020).

In periodic-waveguide transport, the standard figure of merit is often written

$\langle \ln T\rangle$5

which encodes exponential loss only if $\langle \ln T\rangle$6 is actually linear in $\langle \ln T\rangle$7 (Baron et al., 2011). In soft-tissue acoustics, the empirical target law is usually

$\langle \ln T\rangle$8

where $\langle \ln T\rangle$9 is tissue dependent (Jiménez et al., 2014). In viscoelasticity with positive relaxation spectrum, attenuation and dispersion are encoded by an admissible dispersion–attenuation function

$L$0

which is more restrictive than an arbitrary frequency power law (Seredyńska et al., 2010).

These forms are mathematically analogous only at a high level. A plausible implication is that “attenuation law” is best understood as a structural role in a model—how attenuation scales with the control variable—rather than as a fixed equation type.

2. Dust attenuation laws in galaxies

In extragalactic astrophysics, attenuation law and extinction curve are not synonymous. Extinction refers to absorption plus scattering out of a single line of sight, whereas attenuation is the net, galaxy-scale loss of light after scattering into the line of sight, mixed optical depths, unobscured stars, and age-dependent obscuration are included (Salim et al., 2020). This distinction is foundational because two galaxies with the same underlying grain extinction curve can exhibit different attenuation laws.

A standard quantitative framework uses

$L$1

together with the UV–optical slope

$L$2

and the 2175 Å bump strength $L$3 (Salim et al., 2020). The review reports a strong empirical slope–opacity relation for local galaxies,

$L$4

such that lower-$L$5 systems tend to have steeper curves and higher-$L$6 systems greyer ones (Salim et al., 2020).

Radiative-transfer calculations can generate large attenuation-law diversity even when the underlying extinction curve is held fixed. In a simulation suite spanning roughly three orders of magnitude in stellar mass, Narayanan et al. found that increasing fractions of unobscured young stars flatten normalized attenuation curves, whereas increasing fractions of unobscured old stars steepen them; the 2175 Å bump strength varies in tandem and is primarily influenced by the fraction of unobscured O and B stars, with scattered light playing a secondary role (Narayanan et al., 2018). The same study predicts substantial dispersion in attenuation laws at low redshift that decreases toward $L$7 (Narayanan et al., 2018).

This theoretical picture places geometry, age-selective obscuration, and radiative transfer on equal footing with grain physics. It also undercuts the common simplification that an attenuation law is a direct observable proxy for a dust composition alone.

3. Empirical galaxy parameterizations and observed diversity

Flexible attenuation-law parameterizations are now standard because fixed laws such as Calzetti or Milky Way templates often fail to reproduce the diversity seen in galaxy SEDs. One widely used modified-Calzetti form is

$L$8

with $L$9 recovering the original Calzetti prescription (Buat et al., 2019). To reproduce the near-infrared flattening predicted by radiative-transfer models, Buat et al. proposed

$\alpha(f)=\alpha_0 f^\gamma$0

with

$\alpha(f)=\alpha_0 f^\gamma$1

(Buat et al., 2019).

Composite SED analysis at $\alpha(f)=\alpha_0 f^\gamma$2 found that dust-law slope and UV-bump strength are correlated,

$\alpha(f)=\alpha_0 f^\gamma$3

so that steeper attenuation laws have stronger bumps; more active galaxies have shallower curves and weaker bumps (Kriek et al., 2013). Individual-galaxy constraints from SHARDS at $\alpha(f)=\alpha_0 f^\gamma$4 yielded an anti-correlation

$\alpha(f)=\alpha_0 f^\gamma$5

and showed that attenuation-law diversity can bias the UV-slope diagnostic by

$\alpha(f)=\alpha_0 f^\gamma$6

(Tress et al., 2018).

Recent JWST results sharpen the distinction between ensemble averages and object-by-object diversity. An empirical attenuation law derived from stacked JADES/NIRSpec and MIRI data for a mass-selected sample at $\alpha(f)=\alpha_0 f^\gamma$7 gave

$\alpha(f)=\alpha_0 f^\gamma$8

found no significant evidence for a 2175 Å bump in the average curve, and concluded that the ensemble-average law is broadly consistent with the local starburst relation, though flatter in the ultraviolet than several intermediate-redshift determinations (Rodighiero et al., 10 Apr 2026). By contrast, flexible BAGPIPES fits to three JWST galaxies at $\alpha(f)=\alpha_0 f^\gamma$9–8 found non-local attenuation curves on an object-by-object basis, including one case with a Milky Way-like bump (Markov et al., 2023).

The observational state of the field therefore supports both of the following propositions: attenuation laws vary strongly among galaxies, and the ensemble-average law of a controlled sample can nevertheless appear Calzetti-like.

4. Path-length attenuation in structured and disordered media

In guided-wave transport, attenuation law usually means a length dependence of transmission rather than a wavelength-dependent opacity. For ordinary translation-invariant waveguides, exponential decay is commonly assumed: $A_\lambda$0 For periodic monomode waveguides, however, this law is not generally valid (Baron et al., 2011).

Hugonin et al. showed that a cell-by-cell scattering formulation produces a length-dependent incremental loss because the added-cell term depends on the accumulated rear-reflection amplitude $A_\lambda$1 (Baron et al., 2011). Only after $A_\lambda$2 reaches a stationary distribution does a constant asymptotic damping rate $A_\lambda$3 emerge. In the perturbative regime, where $A_\lambda$4, linear behavior is recovered; beyond that regime, coherent accumulation of backscattering near the Brillouin-zone boundary invalidates the naive constant-loss-per-cell picture (Baron et al., 2011).

The paper’s two benchmark geometries demonstrate the point sharply. For the silicon photonic-crystal waveguide, $A_\lambda$5 remained essentially linear for $A_\lambda$6 from 5 to 70. For the grating nanowire, the law was distinctly nonexponential: at $A_\lambda$7, near-linear behavior emerged only for

$A_\lambda$8

and at $A_\lambda$9 the normalized local damping rate reached

$A_\lambda = m_\lambda - m_{\lambda,0}$0

(Baron et al., 2011). The attenuation coefficient in such systems is therefore asymptotic and geometry dependent, not a universally extractable constant.

A related breakdown of single-law intuition appears in generalized radiative transfer for stochastic binary mixtures. For Markovian mixtures, deterministic-origin paths and collision-origin paths obey different attenuation statistics. The uncorrelated free-path density is

$A_\lambda = m_\lambda - m_{\lambda,0}$1

whereas the intercollision density is obtained from the second derivative of the deterministic-origin transmittance,

$A_\lambda = m_\lambda - m_{\lambda,0}$2

after normalization (d'Eon, 2019). The mean correlated free path remains classical,

$A_\lambda = m_\lambda - m_{\lambda,0}$3

but the free-path distributions are nonexponential and origin dependent (d'Eon, 2019). This rules out a single Beer–Lambert attenuation law for reciprocal transport in Markovian binary mixtures.

5. Frequency-dependent attenuation in acoustics and viscoelasticity

In soft-tissue acoustics, attenuation law is often specified empirically as

$A_\lambda = m_\lambda - m_{\lambda,0}$4

with $A_\lambda = m_\lambda - m_{\lambda,0}$5 in typical biological tissues (Jiménez et al., 2014). Because nonlinear propagation generates harmonics, the attenuation law must be represented over a finite band rather than at a single carrier frequency. One practical strategy is to approximate the target power law by a small number of relaxation mechanisms, yielding

$A_\lambda = m_\lambda - m_{\lambda,0}$6

For ultrasound applications over 1–20 MHz, two relaxation processes were reported to be sufficient for most soft tissues, including the fundamental and first ten harmonics (Jiménez et al., 2014).

Recent work generalized this to spatially heterogeneous media by targeting

$A_\lambda = m_\lambda - m_{\lambda,0}$7

with both amplitude and exponent varying voxelwise (Sode et al., 9 Jun 2026). An automated two-relaxation calibration over $A_\lambda = m_\lambda - m_{\lambda,0}$8–1.0 dB/(MHz$A_\lambda = m_\lambda - m_{\lambda,0}$9 cm) and $\langle \ln T\rangle$0–2.0 achieved mean errors below 3% over 1–20 MHz, with dispersion error $\langle \ln T\rangle$1 m/s in the clinically relevant core region $\langle \ln T\rangle$2–1.4 (Sode et al., 9 Jun 2026). The same relaxation formulation was extended into the convolutional perfectly matched layer so that the boundary inherited the local dispersive attenuation law rather than acting as a mismatched absorber (Sode et al., 9 Jun 2026).

Viscoelastic theory imposes a much stronger constraint on admissible attenuation laws. In a medium with positive relaxation spectrum, the attenuation/dispersion function must admit the positive-measure representation

$\langle \ln T\rangle$3

which implies

$\langle \ln T\rangle$4

and enforces sublinear high-frequency growth,

$\langle \ln T\rangle$5

Superlinear power-law attenuation is therefore incompatible with positive relaxation spectrum and with finite propagation speed (Seredyńska et al., 2010). This is a direct theoretical limitation on attenuation-law choice, not merely a fitting preference.

6. Inference, identifiability, and systematic bias

Attenuation laws are often effective parameters inferred from incomplete data, and their estimation is typically entangled with geometry, stellar populations, apertures, or source statistics. In galaxy photometric SED fitting, mock-catalog experiments with BAGPIPES found that $\langle \ln T\rangle$6 and the slope parameter $\langle \ln T\rangle$7 can be recovered without inducing strong spurious population-level correlations when optical-to-IR coverage is available; with all bands, the residual scatter was about 0.12 in $\langle \ln T\rangle$8 and 0.17 in $\langle \ln T\rangle$9, whereas optical-only fits were substantially less stable (Meldorf et al., 2023). The same study concluded that observed $\langle \ln T\rangle = -\alpha L$0–$\langle \ln T\rangle = -\alpha L$1 correlations are unlikely to be pure fitting artifacts, though $\langle \ln T\rangle = -\alpha L$2 remains intrinsically difficult to constrain (Meldorf et al., 2023).

Emission-line approaches face a different identifiability problem. A three-line nebular method based on H$\langle \ln T\rangle = -\alpha L$3, H$\langle \ln T\rangle = -\alpha L$4, and Pa$\langle \ln T\rangle = -\alpha L$5 models the optical depth as

$\langle \ln T\rangle = -\alpha L$6

so that the line ratios can in principle constrain both $\langle \ln T\rangle = -\alpha L$7 and $\langle \ln T\rangle = -\alpha L$8 (Prescott et al., 2022). In practice, most galaxies in the test sample preferred implausibly steep or shallow slopes because elevated Pa$\langle \ln T\rangle = -\alpha L$9/H$\alpha(f)$0 ratios can be reproduced either by slightly sub-unity covering fractions, typically $\alpha(f)$1, or by slit/aperture mismatches between Balmer and Paschen measurements (Prescott et al., 2022). The paper’s conclusion was that a consistent spectroscopic aperture is essential if one wants to measure the optical–infrared attenuation-law slope in individual galaxies (Prescott et al., 2022).

For heavily obscured JWST-selected systems, the degeneracy becomes astrophysically consequential. Flexible Prospector fits to three optically dark galaxies showed a significant degeneracy between dust attenuation-curve slope, total dust attenuation, and stellar mass, with the authors emphasizing that one can hide substantially more stellar mass under flatter attenuation laws (Lapasia et al., 13 Jan 2026). A plausible implication is that attenuation law is often not a nuisance parameter but a dominant uncertainty in population inference.

Across fields, the recurring misconception is that an attenuation law is a fixed material constant. The literature instead shows that it is frequently an effective law: asymptotic rather than local in periodic waveguides, origin dependent in stochastic transport, geometry dependent in galaxies, and causality constrained in viscoelastic media.