-
Simulation-based Inference for Exoplanet Atmospheric Retrieval: Insights from winning the Ariel Data Challenge 2023 using Normalizing Flows
Authors:
Mayeul Aubin,
Carolina Cuesta-Lazaro,
Ethan Tregidga,
Javier Viaña,
Cecilia Garraffo,
Iouli E. Gordon,
Mercedes López-Morales,
Robert J. Hargreaves,
Vladimir Yu. Makhnev,
Jeremy J. Drake,
Douglas P. Finkbeiner,
Phillip Cargile
Abstract:
Advancements in space telescopes have opened new avenues for gathering vast amounts of data on exoplanet atmosphere spectra. However, accurately extracting chemical and physical properties from these spectra poses significant challenges due to the non-linear nature of the underlying physics.
This paper presents novel machine learning models developed by the AstroAI team for the Ariel Data Challe…
▽ More
Advancements in space telescopes have opened new avenues for gathering vast amounts of data on exoplanet atmosphere spectra. However, accurately extracting chemical and physical properties from these spectra poses significant challenges due to the non-linear nature of the underlying physics.
This paper presents novel machine learning models developed by the AstroAI team for the Ariel Data Challenge 2023, where one of the models secured the top position among 293 competitors. Leveraging Normalizing Flows, our models predict the posterior probability distribution of atmospheric parameters under different atmospheric assumptions.
Moreover, we introduce an alternative model that exhibits higher performance potential than the winning model, despite scoring lower in the challenge. These findings highlight the need to reevaluate the evaluation metric and prompt further exploration of more efficient and accurate approaches for exoplanet atmosphere spectra analysis.
Finally, we present recommendations to enhance the challenge and models, providing valuable insights for future applications on real observational data. These advancements pave the way for more effective and timely analysis of exoplanet atmospheric properties, advancing our understanding of these distant worlds.
△ Less
Submitted 17 September, 2023;
originally announced September 2023.
-
Variational analysis of HF dimer tunneling rotational spectra using an ab initio potential energy surface
Authors:
Oleg L. Polyansky,
Roman I. Ovsyannikov,
Jonathan Tennyson,
Sergei P. Belov,
Mikhail Yu. Tretyakov,
Vladimir Yu. Makhnev,
Nikolai F. Zobov
Abstract:
A very accurate, (HF)$_2$ potential energy surface (PES) due to Huang et al. (J. Chem. Phys., 150, 154302 (2019)) is used to calculate the energy levels of the HF dimer by solving the nuclear-motion Schrödinger equation using variational program WAVR4. Calculations on an extended range of rotational states show very good agreement with experimental data. In particular the known empirical rotationa…
▽ More
A very accurate, (HF)$_2$ potential energy surface (PES) due to Huang et al. (J. Chem. Phys., 150, 154302 (2019)) is used to calculate the energy levels of the HF dimer by solving the nuclear-motion Schrödinger equation using variational program WAVR4. Calculations on an extended range of rotational states show very good agreement with experimental data. In particular the known empirical rotational constants for the ground and some observed excited vibrational states are reproduced with an accuracy of about 50 MHz. This level of accuracy is shown to extend to higher excited inter-molecular vibrational states $v$ and higher excited rotational quantum numbers $(J,K_a)$. These calculations allow the assignment of 3 new $J$-branches in an HF dimer tunneling-rotation spectra recorded 30 years ago. These branches belong to excited $K_a$ = 4 state of the ground vibrational state, and $K_a$ = 0 states of excited inter-molecular vibrational states.
△ Less
Submitted 14 July, 2021;
originally announced July 2021.
-
Highly accurate HF dimer ab initio potential energy surface
Authors:
Roman I. Ovsyannikov,
Vladimir Yu. Makhnev,
Nikolai F. Zobov,
Jacek Koput,
Jonathan Tennyson
Abstract:
A very accurate, (HF)$_2$ potential energy surface (PES) is constructed based on \ai\ calculations performed using the MOLPRO package at the CCSD(T) level of theory with an aug-cc-pvQz-F12 basis set at about 161~000 points. a higher correlation correction is computed at CCSDT(Q) level for 2000 points and is considered alongside other more minor corrections due to relativity, core-valence correlati…
▽ More
A very accurate, (HF)$_2$ potential energy surface (PES) is constructed based on \ai\ calculations performed using the MOLPRO package at the CCSD(T) level of theory with an aug-cc-pvQz-F12 basis set at about 161~000 points. a higher correlation correction is computed at CCSDT(Q) level for 2000 points and is considered alongside other more minor corrections due to relativity, core-valence correlation and Born-Oppenheimer failure. The analytical surface constructed uses 500 constants to reproduce the \ai\ points with a standard deviation of 0.3 \cm. Vibration-rotation-inversion energy levels of the HF dimer are computed for this PES by variational solution of the nuclear-motion Schrödinger using program WAVR4. Calculations over an extended range of rotationally excited states show very good agreement with the experimental data. In particular the known empirical rotational constants $B$ for the ground vibrational states are predicted to better than about 2 MHz. $B$ constants for excited vibrational states are reproduced several times more accurately than by previous calculations. The experimental dissociation energy of the HF dimer is reproduced \ai\ within the experimental accuracy of about 1 \cm\ for the first time. This level of accuracy is shown to extend to higher excited inter-molecular vibrational states $v$ and higher excited rotational quantum numbers $(J,K_a)$.
△ Less
Submitted 23 July, 2021; v1 submitted 13 July, 2021;
originally announced July 2021.
-
A new spectroscopically-determined potential energy surface and \emph{ab initio} dipole moment surface for high accuracy HCN intensity calculations
Authors:
Vladimir Yu. Makhnev,
Aleksandra A. Kyuberis,
Oleg L. Polyansky,
Irina I. Mizus,
Jonathan Tennyson,
Nikolai F. Zobov
Abstract:
Calculations of transition intensities for small molecules like H$_2$O, CO, CO$_2$ based on s high-quality potential energy surface (PES) and dipole moment surface (DMS) can nowadays reach sub-percent accuracy. An extension of this treatment to a system with more complicated internal structure -- HCN/HNC (hydrogen cyanide/hydrogen isocyanide) is presented. A highly accurate spectroscopically-deter…
▽ More
Calculations of transition intensities for small molecules like H$_2$O, CO, CO$_2$ based on s high-quality potential energy surface (PES) and dipole moment surface (DMS) can nowadays reach sub-percent accuracy. An extension of this treatment to a system with more complicated internal structure -- HCN/HNC (hydrogen cyanide/hydrogen isocyanide) is presented. A highly accurate spectroscopically-determined PES is built based on a recent \aipes\ of the HCN/HNC isomerizing system. 588 levels of HCN with $J$~=~(0,~2,~5,~9,~10) are reproduced with a standard deviation from the experimental values of $σ=0.0373$ \cm\ and 101 HNC levels with $J$~=~(0,~2) are reproduced with $σ=0.37$ \cm. The dependence of the HCN rovibrational transition intensities on the PES is tested for the wavenumbers below 7200 \cm. Intensities are computed using wavefunctions generated from an \ai\ and our optimized PES. These intensities differ from each other by more than 1\%\ for about 11\% of the transitions tested, showing the need to use an optimized PES to obtain wavefunctions for high-accuracy predictions of transition intensities. An \ai\
DMS is computed for HCN geometries lying below 11~200 \cm. Intensities for HCN transitions are calculated using a new fitted PES and newly calculated DMS. The resulting intensities compare much better with experiment than previous calculations. In particular, intensities of the H--C stretching and bending fundamental transitions are predicted with the subpercent accuracy.
△ Less
Submitted 22 October, 2018;
originally announced October 2018.
-
High accuracy water potential energy surface for the calculation of infrared spectra
Authors:
Irina I. Mizus,
Aleksandra A. Kyuberis,
Nikolai F. Zobov,
Vladimir Yu. Makhnev,
Oleg L. Polyansky,
Jonathan Tennyson
Abstract:
Transition intensities for small molecules such as water and CO$_2$ can now be computed with such high accuracy that they are being used to systematically replace measurements in standard databases. These calculations use high accuracy ab initio dipole moment surfaces and wavefunctions from spectroscopically-determined potential energy surfaces. Here an extra high accuracy potential energy surface…
▽ More
Transition intensities for small molecules such as water and CO$_2$ can now be computed with such high accuracy that they are being used to systematically replace measurements in standard databases. These calculations use high accuracy ab initio dipole moment surfaces and wavefunctions from spectroscopically-determined potential energy surfaces. Here an extra high accuracy potential energy surface (PES) of the water molecule (\hato) is produced starting from an ab initio PES which is then refined to empirical rovibrational energy levels. Variational nuclear motion calculations using this PES reproduce the fitted energy levels with a standard deviation of 0.011 \cm, approximately three times their stated uncertainty. Use of wavefunctions computed with this refined PES is found to improve the predicted transition intensities for selected (problematic) transitions. A new room temperature line list for H2(16)O is presented. It is suggested that the associated set of line intensities is the most accurate available to date for this species.
△ Less
Submitted 3 March, 2018;
originally announced March 2018.
