🔭 astrophysics

Bayesian Doppler Imaging: Simultaneous Inference of Surface Maps and Geometric Parameters

This paper introduces a fully Bayesian, pixel-based Doppler imaging framework that simultaneously infers surface brightness maps and geometric parameters like inclination and rotation velocity, successfully applied to the brown dwarf Luhman 16B to reveal its surface features and constrain its physical properties without relying on fixed literature values.

Original authors: Yamato Ureshino, Hajime Kawahara, Hibiki Yama, Kento Masuda

Published 2026-05-12

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Yamato Ureshino, Hajime Kawahara, Hibiki Yama, Kento Masuda

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Picture: Taking a "Selfie" of a Spinning Star

Imagine you are trying to figure out what a spinning basketball looks like, but you can't see the ball itself. You only have a microphone recording the sound of the ball spinning. If there is a scuff mark on the ball, the sound changes slightly as that mark spins toward you and then away from you.

This paper introduces a new, highly sophisticated way to solve that puzzle for stars, brown dwarfs (failed stars), and planets. The authors, led by Yamato Ureshino and Hajime Kawahara, have created a computer method called Bayesian Doppler Imaging.

Their goal is to do two things at once:

Draw a map of the surface (where the dark clouds and bright spots are).
Measure the geometry (how fast it is spinning and how tilted it is relative to us).

The Problem: The "Blurry Photo" Dilemma

Traditionally, astronomers trying to map these spinning objects faced a tricky problem. It's like trying to guess the shape of a spinning top just by listening to the hum it makes.

The Blur: If a star spins fast, its light gets smeared out (like a long-exposure photo of a moving car).
The Confusion: You can't easily tell if the smearing is because the star is spinning very fast or because it is tilted toward you. Usually, astronomers had to guess the tilt or the speed based on other clues, rather than figuring it out from the light itself.
The Noise: The math involved is incredibly complex, making it hard to know how much you can trust the resulting map.

The Solution: A Smart Detective with a "Smoothness" Rule

The authors built a new detective tool that uses Bayesian statistics (a way of updating beliefs based on evidence) and Gaussian Processes (a mathematical rule that says "things close together on a map should look similar").

Think of it like this:

The Map as a Puzzle: They treat the surface of the star as a grid of thousands of tiny pixels.
The "Smoothness" Rule: They tell the computer, "Don't assume the surface is made of random static noise. Assume it's like a smooth painting." This helps the computer fill in the gaps without getting confused by random noise.
The Magic Trick: The computer is smart enough to separate the "easy" math (drawing the map) from the "hard" math (figuring out the tilt and speed). It solves the easy part instantly and then uses a powerful sampling method (called Hamiltonian Monte Carlo) to hunt down the best answers for the tilt and speed.

This allows them to figure out the tilt and speed directly from the light, without needing to guess them beforehand.

The Test Run: The Synthetic "Fake" Stars

Before looking at real data, the team tested their method on fake data they created on a computer.

They invented stars with specific patterns of dark spots.
They simulated the light these fake stars would send to Earth.
The Result: The method successfully found the spots and correctly guessed the speed and tilt of the fake stars.
The Limitation: They found that while the method is great at finding where a spot is horizontally (longitude), it struggles a bit to tell exactly how far north or south (latitude) it is. It's like being able to tell a car is on the highway, but having trouble telling if it's in the left or right lane if you are looking from a very steep angle.

The Real Deal: Luhman 16B

The team applied their method to real observations of Luhman 16B, a brown dwarf (a "failed star" that is too small to shine like a sun) located relatively close to Earth. They used data from the Very Large Telescope (VLT).

What they discovered:

The Map: They found a massive, dark "storm" or cloud region in the middle latitudes of the brown dwarf. This matches what other astronomers have seen before, but now they have a detailed map showing exactly where the uncertainty lies (where the map is blurry).
The Speed and Tilt: Without guessing, they calculated:
- Tilt (Inclination): The brown dwarf is tilted about 61 degrees relative to us.
- Speed: It spins at the equator at about 31 km per second.
The Radius: By combining the speed and the time it takes to spin once, they estimated the size of the brown dwarf. It fits well with what we expect from theories about how these objects grow and age.
A Mystery: They compared Luhman 16B to its twin, Luhman 16A. If both twins were born from the same cloud and spin at the same speed, but one looks tilted and the other doesn't, it suggests their "spinning axes" might be pointing in different directions. It's like two spinning tops that were dropped together but landed pointing in different ways.

The Bottom Line

This paper presents a new, robust way to take "pictures" of spinning, distant worlds. By treating the problem as a statistical puzzle and using a "smoothness" rule, the authors can now figure out the shape, speed, and tilt of these objects simultaneously.

They confirmed that Luhman 16B has a large dark cloud and provided the first simultaneous measurement of its tilt and spin speed without relying on outside guesses. While the method has limits (it's harder to see details near the poles or on the far side), it provides a much clearer and more honest picture of these mysterious cosmic objects than ever before.

Technical Summary: Bayesian Doppler Imaging: Simultaneous Inference of Surface Maps and Geometric Parameters

Problem Statement
Doppler imaging is a spectroscopic technique used to reconstruct two-dimensional surface maps of rotating stars, brown dwarfs, and exoplanets from time-dependent distortions in spectral line profiles. Traditionally, this has been treated as an ill-posed inverse problem solved via point estimation methods (e.g., maximum entropy or Tikhonov regularization). A fully Bayesian formulation is desirable to quantify uncertainties in the inferred maps; however, it remains computationally challenging due to the high dimensionality of the surface map parameters. Furthermore, physically critical geometric parameters—specifically the inclination ( $i$ ) and equatorial rotation velocity ( $v_{\text{rot}}$ )—are often fixed using external constraints or literature values, rather than being inferred jointly with the surface map. This paper addresses the need for a framework that simultaneously infers surface brightness distributions and these nonlinear geometric parameters while providing rigorous uncertainty quantification.

Methodology
The authors propose a fully Bayesian, pixel-based Doppler imaging framework that treats the inference as a Bayesian linear inverse problem conditioned on a small set of nonlinear geometric parameters.

Forward Model: The surface brightness map is discretized into pixels on a HEALPix grid. The forward model relates the surface map ( $\mathbf{a}$ ) to the observed spectral time series ( $\mathbf{d}$ ) via a design matrix ( $\mathbf{W}$ ), which encapsulates geometric projection, limb darkening, and Doppler broadening effects. The relationship is linear in the map ( $\mathbf{d} = \mathbf{W}\mathbf{a} + \boldsymbol{\epsilon}$ ) but nonlinear in the geometric parameters ( $\boldsymbol{\theta}$ ) that define $\mathbf{W}$ .
Gaussian Process Prior: To enforce spatial smoothness and parameterize the map's spatial scale, the surface map is modeled as a Gaussian Process (GP) prior over pixel intensities. This introduces a characteristic correlation length scale ( $\ell$ ) and amplitude ( $\sigma_a$ ).
Analytical Marginalization: By assuming Gaussian priors for the map and Gaussian likelihoods for the data, the authors analytically marginalize the high-dimensional linear map parameters ( $\mathbf{a}$ ). This reduces the posterior distribution to depend only on the low-dimensional nonlinear parameters ( $\boldsymbol{\theta} = \{i, v_{\text{rot}}, \text{limb-darkening}, \text{noise}, \text{GP hyperparameters}, \dots\}$ ).
Sampling: The marginal likelihood depends nonlinearly on $\boldsymbol{\theta}$ . The authors sample the posterior of these parameters using Hamiltonian Monte Carlo (HMC) with the No-U-Turn Sampler (NUTS), implemented in JAX/NumPyro. This allows for efficient exploration of the parameter space without the computational cost of sampling the full map at every step.
Map Recovery: Once posterior samples of $\boldsymbol{\theta}$ are obtained, the conditional posterior of the surface map is computed for each sample. The marginalized posterior of the map is approximated as a finite mixture of Gaussians weighted by the MCMC samples, allowing for the calculation of posterior means and spatially resolved uncertainties.

Key Contributions

Joint Inference Framework: The paper presents a method to simultaneously infer surface maps and geometric parameters ( $i$ , $v_{\text{rot}}$ ) without fixing them to literature values, disentangling the typical $v_{\text{rot}} \sin i$ degeneracy through time-dependent line profile distortions.
Efficient Bayesian Sampling: By analytically marginalizing the linear map parameters, the method enables efficient MCMC sampling of the nonlinear subspace, making fully Bayesian Doppler imaging computationally feasible for high-resolution maps.
Uncertainty Quantification: The framework provides spatially resolved uncertainty estimates for the surface maps, distinguishing between data-driven features and artifacts arising from the inversion process or prior assumptions.
Implementation: The code is implemented in JAX, leveraging automatic differentiation and GPU acceleration for efficient computation of the design matrix and likelihoods.

Results

Synthetic Data Validation: Experiments with synthetic data demonstrate that the method successfully recovers the longitudes of large-scale surface inhomogeneities and constrains $i$ and $v_{\text{rot}}$ under the adopted model assumptions. The synthetic tests confirm that while longitudinal structure is well-constrained, latitudinal sensitivity is intrinsically limited, particularly for low inclinations or isolated bright features, which may appear as reconstruction artifacts.
Application to Luhman 16B: The framework was applied to high-resolution VLT/CRIRES observations of the brown dwarf Luhman 16B.
- Geometric Parameters: The authors derived an inclination of $i = 61.0^{+14.3}_{-12.3}$ degrees and an equatorial rotation velocity of $v_{\text{rot}} = 31.2^{+5.3}_{-3.1}$ km s $^{-1}$ . These values were inferred without fixing $v_{\text{rot}} \sin i$ or $i$ to prior literature values.
- Surface Map: The analysis revealed a large-scale dark region at mid-latitudes, consistent with previous studies (e.g., Crossfield et al. 2014), but now accompanied by spatially resolved uncertainty estimates. The map also showed a spot-like feature near the equator and bright patches at high latitudes; however, the authors caution that the bright features may be artifacts of the limited latitudinal sensitivity and the smooth GP prior.
- Physical Implications: Combining the inferred $v_{\text{rot}}$ with the rotation period ( $P \approx 4.83$ hr) yields a radius estimate of $\approx 1.2 R_{\text{Jup}}$ , which is broadly consistent with evolutionary models for old brown dwarfs. The authors also note a potential spin-axis misalignment between Luhman 16A and 16B if they share similar equatorial rotation velocities.

Significance
The paper claims that this framework establishes a unified probabilistic model for Doppler imaging that treats the surface map as a latent variable, allowing for the joint constraint of atmospheric structure and fundamental geometric parameters. The ability to quantify uncertainties and identify the limits of latitudinal resolution is highlighted as a major advantage over traditional point-estimation methods. The authors suggest that while the current method relies on a fixed intrinsic line profile, the framework is extensible to jointly infer intrinsic spectral profiles in future work to further reduce systematic residuals. The code is made publicly available to facilitate further applications to substellar and exoplanetary atmospheres.