Uncertainty (Monte Carlo)¶
Exoplanet radii, masses and fluxes come with error bars that are often lopsided, such as \(1.34^{+0.11}_{-0.18}\). A single ESI value hides that. exohab samples every input many times and gives you the resulting ESI distribution.
How to give a measurement¶
| Form | Meaning |
|---|---|
1.34 |
Exact value, no uncertainty |
(1.34, 0.1) |
Symmetric 1-sigma error |
(1.34, 0.11, -0.18) |
Asymmetric errors, upper and lower. The sign of the lower error is ignored, so archive err1/err2 pairs can be used directly |
Asymmetric errors are sampled from the two-piece ("split") normal distribution, the standard reading of a \(x^{+a}_{-b}\) quote.
One planet¶
import exohab as eh
mc = eh.compute_esi_2p_mc(radius=(1.34, 0.11, -0.18), flux=(0.66, 0.23, -0.41),
n_samples=20_000, seed=42, truncate=True)
print(mc.median, mc.ci68, mc.ci95, mc.n_valid)
The result is an MCResult with mean, median, std, ci68, ci95, the
raw samples and n_valid. Call mc.interval(0.9) for other levels.
For the four-parameter ESI, pass mass instead of density and escape
velocity. Both are then derived from the same mass and radius in each draw,
which keeps their physical correlation:
import exohab as eh
mc4 = eh.compute_esi_mc(radius=(0.920, 0.013, -0.012), mass=(0.692, 0.022, -0.022),
surface_temp=(283.3, 2.6), n_samples=20_000, seed=42)
print(mc4.median, mc4.components["interior"].median, mc4.components["surface"].median)
Negative draws: truncate¶
Large error bars can produce physically impossible draws, such as a negative flux.
truncate=False(default): those draws are discarded, andn_validreports how many remain.truncate=True: those draws are redrawn, so alln_samplesare kept.
Both describe the same distribution truncated at zero.
The whole catalogue¶
import exohab as eh
df = eh.load_nasa_exoplanets()
unc = eh.compute_esi_mc_catalog(df, method="2p", n_samples=2000, seed=0)
print(unc.columns.tolist())
The result has one row per planet with esi_2p_mc_median, the 68% and 95%
interval bounds, esi_2p_mc_n_valid and esi_2p_mc_errors_complete.
Missing error bars
When the archive gives no error for a value, it is treated as exact, and
*_errors_complete is False. For those planets the interval is a
lower limit on the true uncertainty.
Sampling tools¶
sample_split_normal and split_normal_cdf are public, if you need the
distribution directly: