Monte-Carlo Imaging for Optical Interferometry
read more
Citations
Imaging the Surface of Altair
Accreting protoplanets in the LkCa 15 transition disk
LkCa 15: A Young Exoplanet Caught at Formation?
IMAGING AND MODELING RAPIDLY ROTATING STARS: α CEPHEI AND α OPHIUCHI
First Resolved Images of the Eclipsing and Interacting Binary Beta Lyrae
References
Markov Chain Monte Carlo: Stochastic Simulation for Bayesian Inference
Markov Chains: Gibbs Fields, Monte Carlo Simulation, and Queues
Markov Chain Monte Carlo: Stochastic Simulation for Bayesian Inference
High-resolution imaging of dust shells by using Keck aperture masking and the IOTA interferometer
An interferometry imaging beauty contest
Related Papers (5)
Analytic continuation of quantum Monte Carlo data by stochastic analytical inference.
Comparison of monte carlo methods for model probability distribution determination in sar interferometry
Frequently Asked Questions (15)
Q2. What is the general algorithm used by MACIM?
The general algorithm used by MACIM is a simulated annealing algorithm with the Metropolis sampler.1, 2 The image state space θj at iteration j consists of the set of pixel vectors {pi}j for all flux elements i with 1 ≤ i ≤ λ. λ is the total number of flux elements.
Q3. What are the main benefits of MACIM?
The main benefit of MACIM are the simulated annealing algorithm that can converge where self-calibration does not, and the flexibility in regularization techniques.
Q4. What is the default value for the annealing algorithm?
Starting from an initial map (by default a point source), the simulated annealing algorithm converges to a global minimum where as long as χ2r < γ (default γ = 4) the authors have T = 1.
Q5. What is the only model fitting option in MACIM?
the only implemented model fitting option is a centrally-located uniform disk (or point source) that takes up some fraction of the total image flux, and an over-resolved (background) flux component.
Q6. How many flux elements are in the top right corner of the map?
By adding up the flux elements in a 3× 3 pixel region for each step in the Markov Chain and calculating the fraction of time there is non-zero flux, the confidence level for the feature is only 54%.
Q7. What is the standard u and v coordinates for m?
The transform between image-space and complex visibility is stored in memory as vectors containing exp(iumxk) and exp(ivmyk) for baselines m and pixels k. um and vm are the standard u and v coordinates for baseline m.
Q8. What is the optimal number of image elements for this data set?
For this data set, one could argue that the optimal number of image elements is about 2000, because with 2000 elements the mean value of χ2r is 1.0 at unity temperature.
Q9. Why is the lower- value not well sampled?
the lower-λ value for mean χ2r = 1 can not be well sampled, because of the very high barriers to flux movement or adding/removing flux elements.
Q10. What is the main purpose of the regularizer?
Inspired by the Ising model, this regularizer encourages large regions of dark space in-between regions of flux and represents a means to utilize a priori knowledge of source structure.
Q11. What is the confidence level for the top right feature?
Given that there are 2000 flux elements, an appropriate question phrasing is “What is the confidence level for the top right feature containing more than 1/2000th of the flux”.
Q12. What is the probability of a point source in the top right corner?
There is a very small chance that many (in this case ∼100) flux elements can congregate in a single pixel, so the presence of a point-source becomes strong a priori knowledge that influences the final image.
Q13. How many pixel coordinates are needed to be stored in memory?
Splitting the pixel coordinates into xk and yk in this fashion means that only 2Mb √ n complex numbers need to be stored in memory (with Mb the number of baselines).
Q14. Why is the MACIM code more accurate for optical inferferometry than radio?
This is more true for optical inferferometry than radio interferometry, due to the general unavailability of absolute visibility phase.
Q15. What is the probability of the modification to the image state?
Given a temperature T , the modification to the image state is accepted with a probability:p(j, j + 1) = min(1, exp( χ2({pi}j) − χ2({qi})2T + α∆R)).