Public PhD Defense
Joel Ong
Aug 2 2022
Animation: NASA
Power spectra of MDI dopplergrams
\[ \begin{aligned} \Delta\nu &\sim \sqrt{M / R^3} \\ \nu_\text{max} &\sim M / R^2 \sqrt{T_\text{eff}} \end{aligned} \]
Mode frequencies satisfy \(\nu_{nl} \sim \Delta\nu\left(n + {l \over 2} + \epsilon_l(\nu)\right) + \mathcal{O}(1/\nu)\)
Three quantum numbers \(n, l, m\): \[ \begin{aligned} l &= 0, 1, 2, \ldots \\ m &= -l, -l+1, \ldots, l-1, l \end{aligned} \]
Zonal (\(m = 0\))
Prograde sectoral
(\(m = +l\))
Retrograde sectoral
(\(m = -l\))
\[\huge {\partial^2 \over \partial t^2}f = c_s^2 \nabla^2 f\]
\[\huge -\omega^2 f = c_s^2 \nabla^2 f\]
\[\large \nabla^2 f = -{\omega^2 \over c_s^2} f \equiv -|\mathbf{k}|^2 f\]
Dispersion relation: \[\omega^2 = c_s^2 |\mathbf{k}|^2\]
\[\huge {\partial^2 \over \partial t^2}f = \left(c_s^2 \nabla^2 - \omega^2_\text{ac}(\mathbf{x})\right) f\]
\[\huge -\omega^2 f = \left(c_s^2 \nabla^2 - \omega^2_\text{ac}(\mathbf{x})\right) f\]
\[\large \nabla^2 f = -\left({\omega^2 - \omega_\text{ac}^2\over c_s^2}\right) f \equiv -k^2 f\]
Wave propagates where \(k^2(r, \omega) > 0\) and decays where \(k^2(r, \omega) < 0\).
\[{c_s^2 k_r^2 \sim \omega^2 \left(1 - {{\color{blue} S_l}^2 \over \omega^2}\right)\left(1 - {{\color{orange}N}^2 \over \omega^2}\right)}\]
\[\small\color{orange}N^2 = {- g}\left.{\partial \log \rho \over \partial s}\right|_P{\mathrm d s \over \mathrm d r}\]
\[\small\color{blue}S_l^2 = c_s^2 k_h^2 = {l(l+1) c_s^2 \over r^2}\]
\[{c_s^2 k_r^2 \sim \omega^2 \left(1 - {{\color{blue} S_l}^2 \over \omega^2}\right)\left(1 - {{\color{orange}N}^2 \over \omega^2}\right)}\]
\[\Large \omega_+^2 \sim c_s^2 |\mathbf{k}|^2\]
\[{\color{gray} \omega_p > S_l, N}\]
\[\Large \omega_+^2 \sim c_s^2 |\mathbf{k}|^2\]
\[\large\mathbf{v}_g = {\partial \omega \over \partial \mathbf{k}} = c_s {\mathbf{k} \over |\mathbf{k}|}\]
\[ \begin{aligned} \oint \mathbf{k} \cdot \mathrm d \mathbf{r} &= 2 \pi \left(n' - {1 \over 2}\right) \\ 2 \omega \int_0^R {\mathrm d r \over c_s} &= 2 \pi \left(n + {l \over 2} + \epsilon_l\right) \end{aligned} \]
\[\large\nu = {1 \over 2T}\left(n + {l \over 2} + \epsilon\right)\]
\[\Large \omega_-^2 \sim N^2 {k_h^2 \over |\mathbf{k}|^2}\]
\[{\color{red} \omega_g < N, S_l}\]
\[\Large \omega_-^2 \sim N^2 {k_h^2 \over |\mathbf{k}|^2}\]
\[ \begin{aligned} \oint \mathbf{k} \cdot \mathrm d \mathbf{r} &= 2 \pi n' \\ \implies {2 \sqrt{l(l+1)} \over \omega} &\int_{N^2 > 0} {N \over r} \mathrm d r \\&= 2 \pi \left(n + {l \over 2} + \epsilon_{g, l}\right) \end{aligned} \]
\[\large{1 \over \nu} = \Delta\Pi_l \left(n + {l \over 2} + \epsilon_{g, l}\right)\]
Data: \(y_\text{obs} \in Y\)
Models: \(x_i \in X\);\[F: X \to Y\]
Best-fitting model: \[x = \mathop{\mathrm{argmax}}_{x_j \in X}\ \mathcal{L}\left(x_j\right)\]
\[F: \underbrace{\left(M, t, Y_0, Z_0, \alpha_\text{mlt}, \ldots\right)}_{x \in X} \mapsto \underbrace{\left(L, T_\text{eff}, [\text{M/H}], \log g, \ldots\right)}_{y \in Y}\]
\[ \color{red} \text{and } \Delta\nu, \nu_{\text{max}}, \left\{\nu_{n,l}\right\} \]
Hare “Zebedee”, Cunha+ 2021
Precise measurements of field stars: \[ {\sigma_R \over R} \lesssim 2 \%; {\sigma_M \over M} \lesssim 5 \% \]
For the Sun, even standard solar models don’t give the right frequencies!
The asteroseismic “surface term”: a (mostly) smooth function of frequency.
Lacunae in literature:
16 Cyg B: from Ong, Basu, McKeever (2021)
For each choice of surface term correction,
we construct a posterior distribution
for some quantities of interest
(e.g. \(M\), \(R\), age) for each star in our samples
For each parameter \(P\), we consider normalised differences \[z_{P, \text{BG14 vs. other}} = {\mu_{P, \text{other}} - \mu_{P, \text{BG14}} \over \sqrt{\sigma^2_{P, \text{BG14}} + \sigma^2_{P, \text{other}}}}\]
We examine the distribution of these
differences
over a large* sample of stars, all else being equal
*Not actually very large
Main-sequence stars
\[\tiny{(\text{Kepler LEGACY sample}: N = 66)}\]
General agreement on the
inferred masses…
\(z\)-score for mass
…and on the inferred radii.
\(z\)-score for radius
Parameter estimates generally
agree quite robustly…
\(z\)-score for age
… but not for all parameters. \(\tiny(p \sim 10^{-8})\)
\(z\)-score for initial helium abundance
First-ascent red giants
\[\tiny{(\text{NGC 6791}: N = 29; l=0,2\text{ only})}\]
Nonparametric methods appear
to agree with each other…
\(z\)-score for initial helium abundance
…but not with our
fiducial parameterisation…
\(z\)-score for mass
…and these offsets appear
to be systematic.
\(z\)-score for age
NGC 6791 is an open cluster — what about the
actual distribution of stellar properties?
Stars should be coeval, so red giants should be of similar mass and age.
Disagreements on the age scale?
Reference values: McKeever+ 2019 EB’s
Disagreements on the mass distribution?
vs. isochronal mass range
Disagreements on composition?
Reference values: McKeever+ 2019 EB’s
Other parametric methods yield comparable internal scatter!
Different treatments of the surface term exhibit qualitative systematic differences for main-sequence stars vs. red giants.
What happens in between?
Why have we excluded dipole (\(l = 1\)) modes in this analysis?
Over the course of post-main-sequence evolution, \(N^2\) increases dramatically —
an interior g-mode cavity develops.
(proxy for age \(\to\))
Mixed modes exhibit
avoided crossings
between underlying p- and g-modes.
Pure p-modes: \[\nu_{n,l} \sim \Delta\nu \left(n_p + {l \over 2} + \epsilon_{n,l}\right)\]
Pure g-modes: \[{1 \over \nu_{n,l}} \sim \Delta\Pi_l \left(n_g + {l \over 2} + \epsilon_{g, n,l}\right)\]
\[{k_r^2 \sim {\omega^2 \over c_s^2} \left(1 - {{\color{blue} S_l}^2 \over \omega^2}\right)\left(1 - {{\color{orange}N}^2 \over \omega^2}\right)}\]
\[\tiny {\color{gray}\sin \left[\theta_p(\nu)\right] = 0};\ {\color{red}\cos \left[\theta_g(\nu)\right] = 0}\]
\[\tiny {\color{gray}\sin \left[\int_{r_{1,p}}^{r_{2,p}} k_r \mathrm d r\right] = 0};\ {\color{red}\cos \left[\int_{r_{1,g}}^{r_{2,g}} k_r \mathrm d r\right] = 0}\]
\[\tiny {\color{gray}\tan \left[\int_{r_{1,p}}^{r_{2,p}} k_r \mathrm d r\right]} {\color{red}\cot \left[\int_{r_{1,g}}^{r_{2,g}} k_r \mathrm d r\right]} = {1 \over 4} \exp \left[-2 \int_{r_{2,g}}^{r_{1,p}} |k_r| \mathrm d r\right]\]
(Coupling between oscillators)
Relationship between \((\omega_i, \alpha_i)\) and stellar structure unclear…
\[\Large \hat H \psi_n = \left(\hat T + \hat V\right)\psi_n = E_n \psi_n\]
\[\large {\color{blue}{\hat H_1}} \psi_n = \left(\hat T + \hat V_1\right)\psi_n = {\color{blue}E_n \psi_n}\]
\[\large {\color{orange}{\hat H_2}} \psi_n = \left(\hat T + \hat V_2\right)\psi_n = {\color{orange}E_n \psi_n}\]
\[\huge \hat H = \hat T + \hat V_1 + \hat V_2 = {\color{blue}{\hat H_1}} + \hat V_2 = \hat V_1 + {\color{orange}{\hat H_2}}\]
\[\Large\psi \sim {1 \over \sqrt{2}} \left({\color{blue}\psi_1} + {\color{orange}\psi_2}\right)\] \[\Large E \sim E_1 - \left|\int {\color{blue}\psi_1^*} \hat H {\color{orange}\psi_2}\ \mathrm dV\right|\]
\[\Large\psi \sim {1 \over \sqrt{2}} \left({\color{blue}\psi_1} - {\color{orange}\psi_2}\right)\] \[\Large E \sim E_1 + \left|\int {\color{blue}\psi_1^*} \hat H {\color{orange}\psi_2}\ \mathrm dV\right|\]
\[\psi_\text{mol} = {\color{blue}\sum_i c_{1,i} \psi_{1,i}} + {\color{orange}\sum_j c_{2,j} \psi_{2,j}}.\]
Energy eigenvalues and mixing coefficients are solutions to the
Generalised Hermitian Eigenvalue Problem
\[ \mathbf{H} \begin{bmatrix} {\color{blue}\mathbf{c}_1} \\ {\color{orange}\mathbf{c}_2} \end{bmatrix} = E \cdot \mathbf{D} \begin{bmatrix} {\color{blue}\mathbf{c}_1} \\ {\color{orange}\mathbf{c}_2} \end{bmatrix} \] \[ \small H_{{\color{blue}i}{\color{orange}j}} = \langle {\color{blue}\psi_i}|\hat{H}|{\color{orange}\psi_j}\rangle;~~~~D_{{\color{blue}i}{\color{orange}j}} = \langle {\color{blue}\psi_i}|{\color{orange}\psi_j}\rangle \]
\[\Large \hat{\mathcal{L}} \xi_n = -\omega_n^2 \rho_0 \xi_n\]
\[\Large \hat{\mathcal{L}} = {\color{red}\hat{\mathcal{L}}_\gamma} + \hat{\mathcal{R}}_\gamma\]
\[\Large {\color{red}\hat{\mathcal{L}}_\gamma \xi_{\gamma,n} = -\omega^2_{\gamma,n}\xi_{\gamma,n}}\]
\[\Large \hat{\mathcal{L}} = \hat{\mathcal{R}}_\pi + {\color{grey}\hat{\mathcal{L}}_\pi}\]
\[\Large {\color{grey}\hat{\mathcal{L}}_\pi \xi_{\pi,n} = -\omega^2_{\pi,n}\xi_{\pi,n}}\]
\[\Large \hat{\mathcal{L}} = {\color{grey}\hat{\mathcal{L}}_\pi} + \hat{\mathcal{R}}_\pi = \hat{\mathcal{R}}_\gamma + {\color{red}\hat{\mathcal{L}}_\gamma}\]
\[\Large \xi_\text{mixed} \sim {\color{grey}c_\pi \xi_\pi} + {\color{red}c_\gamma \xi_\gamma}\]
\[\xi_\text{mixed} \sim {\color{grey}
\sum_i c_{\pi, i} \xi_{\pi,i}} + {\color{red} \sum_j c_{\gamma, j}
\xi_{\gamma,j}}\] Mixed mode frequencies and mixing
coefficients can likewise be found by solving the
Generalised Hermitian Eigenvalue Problem: \[
\begin{bmatrix}
{\color{grey}\mathbf{L}_{\pi\pi}} & \mathbf{L}_{\pi\gamma} \\
\mathbf{L}_{\pi\gamma}^T & {\color{red}\mathbf{L}_{\gamma\gamma}}
\end{bmatrix}
\begin{bmatrix}
{\color{grey}\mathbf{c}_\pi} \\ {\color{red}\mathbf{c}_\gamma}
\end{bmatrix}
= -\omega^2 \begin{bmatrix}
\mathbb{1} & \mathbf{D} \\ \mathbf{D}^T & \mathbb{1}
\end{bmatrix}
\begin{bmatrix}
{\color{grey}\mathbf{c}_\pi} \\ {\color{red}\mathbf{c}_\gamma}
\end{bmatrix}.
\] \[
\small
L_{{\color{grey}i}{\color{red}j}} = \langle {\color{grey}\xi_i},
\hat{\mathcal{L}}{\color{red}\xi_j}\rangle;~~~~D_{{\color{grey}i}{\color{red}j}}
= \langle {\color{grey}\xi_i}, {\color{red}\xi_j}\rangle
\]
(Ong & Basu 2020)
Consider two stars with identical \(M\) and \(R\),
differing only in the near surface layers: \[\hat{\mathcal{L}}_0\xi_{0,n} = -\omega_{0,n}^2
\xi_{0,n};~~~~\hat{\mathcal{L}}\xi_{n} = -\omega_{n}^2 \xi_{n}.\]
We write \[\hat{\mathcal{L}} =
\hat{\mathcal{L}}_0 + \lambda {\hat{\mathcal{V}}},\] so that
\(\lambda\) interpolates linearly
between the two structures.
\[\Large \hat{\mathcal{V}}{\color{red}\xi_\gamma} \to 0\]
\({\color{red}\gamma\text{-modes}}\)
are confined to the stellar interior,
so unaffected by surface
term
How is this compatible with a smooth function of frequency?
Mixing coefficients: customarily defined so that \(\zeta \to 1\) for pure g-modes, and \(\to 0\) for pure p-modes
\[\huge \times\]
Apply a partial correction: \(\delta\nu_\text{surf, mixed} = {\color{red} \delta\nu_\text{surf}} \times {\color{blue}(1 - \zeta)}\)
\(\delta\nu_\text{mixed} = (1 - \zeta)\delta\nu_\text{surf}\)
These two operations do not necessarily commute.
Subgiant: \(\Delta\nu = 67.5\
\mu\text{Hz}\)
Full matrix coupling
\(\lambda\) from \(0 \to 1\)
Subgiant: \(\Delta\nu = 67.5\
\mu\text{Hz}\)
Traditional approximation to
coupling
Subgiant: \(\Delta\nu = 67.5\
\mu\text{Hz}\)
Full vs. Traditional mode coupling
Young red giant: \(\Delta\nu = 17.2\ \mu\text{Hz}\)
Evolved red giant: \(\Delta\nu = 3.9\ \mu\text{Hz}\)
Surface-term corrections
should not reorder mixed modes!
(cf. Ball+ 2018)
traditional approx. OK
perturbative series diverges
???
Ignore mixing altogether? (a la Ball et al. 2018)
\[ \left( \begin{bmatrix} {\color{grey}\mathbf{L}_{\pi\pi}} & \mathbf{L}_{\pi\gamma} \\ \mathbf{L}_{\pi\gamma}^T & {\color{red}\mathbf{L}_{\gamma\gamma}} \end{bmatrix} + \lambda \begin{bmatrix} {\color{blue}\mathbf{V}} & 0 \\ 0 & 0 \end{bmatrix} \right) \begin{bmatrix} {\color{grey}\mathbf{c}_\pi} \\ {\color{red}\mathbf{c}_\gamma} \end{bmatrix} = -\omega^2 \begin{bmatrix} \mathbb{1} & \mathbf{D} \\ \mathbf{D}^T & \mathbb{1} \end{bmatrix} \begin{bmatrix} {\color{grey}\mathbf{c}_\pi} \\ {\color{red}\mathbf{c}_\gamma} \end{bmatrix}. \]
We may now continue our methodological inventory.
Subgiants
\[\tiny{(\text{Various Kepler/K2/TESS}: N = 47;\ l=0, 1, 2)}\]
First-order approximation yields
systematically lower stellar masses!
(population-level systematic error)
(two different
correction techniques)
All the surface corrections yield similar best-fit models;
main discrepancies arise in the wings of the posterior distribution
EPIC212478598
e.g. Measuring \(Y_0\) from the TESS CVZs (TASOC WG2 2.12)
Nonparametric treatments of the surface term may yield
qualitatively different results from parametric
ones,
particularly for measuring \(Y_0\), and
for evolved red giants.
Ong, Basu, McKeever (2021)
Decomposition of wave operator into purely
acoustic/buoyant propagation terms and their remainder
operators
permits closed-form evaluation of coupling matrix
elements.
Ong and Basu (2020)
Traditional surface term corrections handle mode coupling
only to first order, if at all.
More sophisticated techniques are required
as stars get more evolved.
Ong, Basu, Roxburgh (2021)
Inference of stellar masses and compositions depend on
how mode mixing is treated
when correcting for the surface term.
Ong, Basu, Lund, et al. (2021)
We generalise conventional surface corrections to apply to mixed modes, by analytically decoupling mixed modes into p- and g-like components. While the surface term behaves differently on red giants vs. main-sequence stars, it exhibits intermediate behaviour in subgiants. Measurements of stellar masses and compositions depend significantly on how the surface term is treated.
\[\nu_\text{model} \mapsto \nu_\text{corr} = \nu_\text{model} + \delta\nu_{nl,\text{surf}},\] with corrections depending only on the model as \(\delta\nu_{nl,\text{surf}} \sim f(\nu_{nl}; x)\).
Corrections with free parameters \(\theta \in \Theta\) as \(\delta\nu_{nl,\text{surf}} \sim f(\nu_{nl}; x; \theta)\).
\[\scriptsize {\text{e.g. Ball \& Gizon (2014) correction: }\delta\nu_{nl,\text{surf}} \sim \left(a_{-1} \left(\nu_{nl}\right)^{-1} + a_{3} \left(\nu_{nl}\right)^{3}\right) / I_{nl}}\]
Use of transformed variables such that \[O_{nl,\text{surf}} \sim f(\nu_{nl})\]
(where the structure of \(f\) is left underspecified)
Interpolation required to compare
\(f^\text{obs}(\nu_\text{obs})\)
vs. \(f^\text{model}(\nu_\text{obs})\)
(instead of \(f^\text{model}(\nu_\text{model})\)).
\[\begin{aligned} \text{e.g. } r_{02, n} = {\nu_{n,0} - \nu_{n-1,2} \over \nu_{n,1} - \nu_{n-1,1}} \end{aligned}\]
Interpretation as surface-independent functions of frequency:
use these to constrain model directly (Otí Floranes et al. 2005)
\[\begin{aligned} r_{02, n} & \sim \delta_2(\nu_{n,0})\\ r_{01, n} \sim \delta_1(\nu_{n,0})&; ~~~ r_{10, n} \sim \delta_1(\nu_{n,1}) \end{aligned}\]
\[\nu_{nl} \sim \Delta\nu\left(n + {l \over 2} + \epsilon_{nl}\right), \implies \epsilon_{nl} = {\nu_{nl} \over \Delta\nu} - n - {l \over 2} \equiv{\epsilon_l(\nu_{nl})}\] \[ \text{Compute } \mathcal{E}_l(\nu^\text{obs}_{nl}) = \epsilon^\text{obs}_{l}(\nu^\text{obs}_{nl}) - \epsilon^\text{model}_{l}(\nu^\text{obs}_{nl}). \]
All of the \(\mathcal{E}_l\) should collapse to a single function if the structural differences are localised to the surface (Roxburgh 2016).
Bad: scatter around best-fit function
Better: data points collapse to single function
\[ \begin{aligned} {\partial \rho \over \partial t} + \nabla \cdot \left(\rho \mathbf{v}\right) &= 0 \\ \rho \left({\partial \mathbf{v} \over \partial t} + \mathbf{v}\cdot\nabla\mathbf{v}\right) &= -\nabla P - \rho \nabla \Phi\\ \nabla^2 \Phi &= 4 \pi G \rho \end{aligned} \]
\[\rho(t, \mathbf{x}) = \rho_0(r) + \underbrace{\left(\rho'(r) + \vec{\xi}(r) \cdot \nabla\rho_0(r)\right)}_{\mathclap{\delta \rho\text{, the Lagrangian perturbation in $\rho$}}}e^{\pm i \omega t} Y_l^m(\theta,\phi)\]
\[\delta s = 0 \implies \delta P = c_s^2 \delta \rho\]
\[\mathbf{u}(t, r, \theta, \phi) = e^{\pm i \omega t} \left[u_r(r) \mathbf{Y}_l^m(\theta,\phi) + u_h(r) \mathbf{\Psi}_l^m (\theta, \phi) + u_t(r) \mathbf{\Phi}_l^m (\theta, \phi)\right].\]
\[ \Large \begin{aligned} \mathcal{L}\vec{\xi} &= \nabla \left(\rho \vec\xi \cdot \vec{g} + c_s^2 \rho \nabla \cdot \vec\xi\right)\\ &- \vec{g} \nabla \cdot (\rho \vec\xi) - \rho G \nabla\left(\int \mathrm d^3 x' {\nabla' \cdot(\rho \vec\xi) \over |x - x'|}\right)\\ &= -\omega^2 \rho \vec{\xi}. \end{aligned} \]
for \(r/R \sim 1\),
\[ \Large \left(c_s {\mathrm d \over \mathrm d r}\right)^2 \xi_r \sim (\omega_\text{cutoff}^2(r) - \omega^2)\xi_r \]