virgil.night_summaries
Per-night, per-peak (position, velocity, flux) summaries of a binary's companion, fitted at visibility level with linear motion over the whole night.
Per-night, per-peak (position, velocity, flux) summaries at visibility level.
Provisional
May change in any 0.4.x release without a deprecation cycle.
Each night's companion is fitted peak by peak with linear motion over the whole
night, p(t) = p + v (t - t̄), on the visibilities with the noise model of
the data (gains, closure offsets and spanning gain modes marginalized as in
whitened_residuals). The summary of a
peak is the mean and covariance of (p, v, ln f) at the quoted errors, with
the per-block counts, minimum χ² and Hessians that rescale it to another error
scale in closed form (for data without gains or closure offsets, where a scale
factors out of the covariance). For frames of error σ the velocity variance
is σ² / Σ_j (t_j - t̄)²; in terms of the night position's error
σ_p² = σ² / N it is 12 σ_p² / span² for uniform sampling. A short or
faint night's velocity therefore carries no weight and no threshold is needed.
DELTA = 0.001
module-attribute
NightSummary
dataclass
One night's catalogue of peaks.
Attributes:
| Name | Type | Description |
|---|---|---|
t_ref |
float
|
The night's reference time |
t_ss |
float
|
|
n_frames |
int
|
Number of frames (exposures) |
blocks |
tuple of str
|
The error blocks of |
peaks |
tuple of NightPeak
|
Converged peaks, best first. Empty if none converged. |
loglike_null |
float
|
The night's log likelihood with no companion. |
offsets |
tuple of float, optional
|
The frame times |
weights |
tuple of float, optional
|
The weight of each frame in the |
t_ref
instance-attribute
t_ss
instance-attribute
n_frames
instance-attribute
blocks
instance-attribute
peaks
instance-attribute
loglike_null
instance-attribute
offsets = None
class-attribute
instance-attribute
weights = None
class-attribute
instance-attribute
__init__(t_ref, t_ss, n_frames, blocks, peaks, loglike_null, offsets=None, weights=None)
__len__()
frame_offsets()
The frame offsets t_j - t̄ (days): (n_frames,).
frame_weights()
The frame weights of the (p, v) fit: (n_frames,).
projection()
The (2, n_frames) map from the companion's frame positions to
the night's (p̂, v̂): the weighted least-squares fit of
p + v (t_j - t̄), exact for the positions under these scalar
frame weights. It matches the visibility fit only to first order in
the track's departure from a line: the fit weights each frame by its
own information (rows, errors, baseline orientation), which the
equal default does not. One frame, or a zero t_ss, gives the
instantaneous position and a zero row for the velocity (taken from the
state).
covariance(peak, scale=1.0, jitter=0.0)
peak's covariance at error scale with a jitter (mas),
independent per frame, added in quadrature to every frame's position
error: the night position (at t̄) gains jitter² / n_frames and
the velocity jitter² / t_ss (infinite for one frame).
NightPeak
dataclass
One peak of one night.
Attributes:
| Name | Type | Description |
|---|---|---|
mean |
ndarray
|
|
cov |
ndarray
|
|
loglike |
float
|
Log likelihood of the night at the peak, normalization included. |
n, chi2 |
ndarray
|
Per block ( |
hessian |
ndarray
|
|
rescalable |
bool
|
Whether |
mean
instance-attribute
cov
instance-attribute
loglike
instance-attribute
n
instance-attribute
chi2
instance-attribute
hessian
instance-attribute
rescalable = True
class-attribute
instance-attribute
__init__(mean, cov, loglike, n, chi2, hessian, rescalable=True)
precision(scale=1.0)
Σ⁻¹ with the errors of block g multiplied by scale[g]
(a number or one per block).
covariance(scale=1.0)
Σ at error scale (see precision).
log_scale(scale)
ln L_j(s) - ln L_j(1) for block scales s: -Σ_g [n_g ln
s_g + (χ²_g / 2)(s_g⁻² - 1)] (the mean is held at its s = 1
value).
night_summaries(data, positions, *, amp_range=None)
Each night's per-peak linear-motion fit on the visibilities.
The peaks start from each night's catalogue
(EpochPositions.peaks) at zero velocity
and are refined, in float64, over the whole night with the primary of
positions (no half-night split, so closure offsets and spanning gain
modes are kept). A refinement that does not converge is dropped; peaks
that converge to one point are merged.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
data
|
Epochs or sequence of OIData
|
The nights, in the order of |
required |
positions
|
EpochPositions
|
The nights' peak catalogues and the primary diameter. |
required |
amp_range
|
tuple of float
|
Profile one visibility amplitude per night in this range, as the orbit fit does (default None). |
None
|
Returns:
| Type | Description |
|---|---|
tuple of NightSummary
|
|
keep_peaks(summaries, log_outlier, delta=DELTA)
The peaks the position-level likelihood needs.
A peak is kept iff (1 - ε_lo) L_j ≥ δ ε_lo e^{b_k}: an orbit through it
contributes at best (1 - ε) L_j to the night's mixture factor
(1 - ε) e^{ℓ_k} + ε e^{b_k}, with no Occam factor. ε_lo is the
smallest Gauss–Jacobi node of the EPS_PRIOR integral of ε, with
n = ⌈(K + 1) / 2⌉ nodes for K nights.
Parameters:
| Name | Type | Description | Default |
|---|---|---|---|
summaries
|
sequence of NightSummary
|
From |
required |
log_outlier
|
array - like
|
|
required |
delta
|
float
|
Reporting precision (default |
DELTA
|
Returns:
| Type | Description |
|---|---|
tuple of NightSummary
|
|