Hipparcos Modelling

This tutorial explains how to model Hipparcos IAD data. The first example reproduces the catalog values of position, parallax, and proper motion. The second uses Hipparcos to constrain the mass of a directly imaged planet.

Reproduce Catalog Values

This is the so-called "Nielsen" test from Nielsen et al (2020) and available in Orbitize!.

We start by using a system with a planet with zero mass to fit the straight line motion.

using Octofitter
using Distributions
using CairoMakie

hip_like = Octofitter.HipparcosIADLikelihood(
    hip_id=21547,
    renormalize=true, # default: true
    variables=@variables begin
        # Optional: flux ratio for luminous companions, one entry per companion
        # fluxratio ~ Product([Uniform(0, 1), Uniform(0, 1)])  # uncomment if needed for unresolved companions
    end
)

planet_b = Planet(
    name="b",
    basis=AbsoluteVisual{KepOrbit},
    variables=@variables begin
        mass = 0.
        e = 0.
        ω = 0.
        a = 1.
        i = 0
        Ω = 0.
        tp = 0.
    end
)

sys = System(
    name="c_Eri_straight_line",
    companions=[planet_b],
    likelihoods=[hip_like],
    variables=@variables begin
        M = 1.0 # Host mass not important for this example
        rv = 0.0 # system RV not significant for this example
        plx ~ Uniform(10,100)
        pmra ~ Uniform(-100, 100)
        pmdec ~  Uniform(-100, 100)

        # It is convenient to put a prior of the catalog value +- 10,000 mas on position
        ra_hip_offset_mas ~  Normal(0, 10000)
        dec_hip_offset_mas ~ Normal(0, 10000)
        dec = $hip_like.hip_sol.dedeg + ra_hip_offset_mas/60/60/1000
        ra = $hip_like.hip_sol.radeg + dec_hip_offset_mas/60/60/1000/cosd(dec)

        ref_epoch = Octofitter.hipparcos_catalog_epoch_mjd
    end
)

model = Octofitter.LogDensityModel(sys)
LogDensityModel for System c_Eri_straight_line of dimension 5 and 104 epochs with fields .ℓπcallback and .∇ℓπcallback

Let's initialize the starting point for the chains to reasonable values

initialize!(model, (;
    plx=34.,
    pmra=44.25,
    pmdec=-64.5,
    ra_hip_offset_mas=0.,
    dec_hip_offset_mas=0.,
))

We can now sample from the model using Hamiltonian Monte Carlo. This should only take about 15 seconds.

using Pigeons
chain,pt = octofit_pigeons(model, n_rounds=6)
(chain = MCMC chain (64×21×1 Array{Float64, 3}), pt = PT(checkpoint = false, ...))

Plot the posterior values:

octoplot(model,chain,show_astrom=false,show_astrom_time=false)
Example block output

We now visualize the model fit compared to the Hipparcos catalog values:

using LinearAlgebra, StatsBase
fig = Figure(size=(1080,720))
j = i = 1
for prop in (
    (;chain=:ra, hip=:radeg, hip_err=:e_ra),
    (;chain=:dec, hip=:dedeg, hip_err=:e_de),
    (;chain=:plx, hip=:plx, hip_err=:e_plx),
    (;chain=:pmra, hip=:pm_ra, hip_err=:e_pmra),
    (;chain=:pmdec, hip=:pm_de, hip_err=:e_pmde)
)
    global i, j, ax
    ax = Axis(
        fig[j,i],
        xlabel=string(prop.chain),
    )
    i+=1
    if i > 3
        j+=1
        i = 1
    end
    unc = hip_like.hip_sol[prop.hip_err]
    if prop.chain == :ra
        unc /= 60*60*1000 * cosd(hip_like.hip_sol.dedeg)
    end
    if prop.chain == :dec
        unc /= 60*60*1000
    end
    if prop.hip == :zero
        n = Normal(0, unc)
    else
        mu = hip_like.hip_sol[prop.hip]
        n = Normal(mu, unc)
    end
    n0,n1=quantile.(n,(1e-4, 1-1e-4))
    nxs = range(n0,n1,length=200)
    h = fit(Histogram, chain[prop.chain][:], nbins=55)
    h = normalize(h, mode=:pdf)
    barplot!(ax, (h.edges[1][1:end-1] .+ h.edges[1][2:end])./2, h.weights, gap=0, color=:red, label="posterior")
    lines!(ax, nxs, pdf.(n,nxs), label="Hipparcos Catalog", color=:black, linewidth=2)
end
Legend(fig[i-1,j+1],ax,tellwidth=false)
fig
Example block output

Constrain Planet Mass

We now allow the planet to have a non zero mass and have free orbit. We start by specifying relative astrometry data on the planet, collated by Jason Wang and co. on whereistheplanet.com.

astrom_dat = Table(;
    epoch = [57009.1, 57052.1, 57053.1, 57054.3, 57266.4, 57332.2, 57374.2, 57376.2, 57415.0, 57649.4, 57652.4, 57739.1, 58068.3, 58442.2],
    sep   = [454.24, 451.81, 456.8, 461.5, 455.1, 452.88, 455.91, 455.01, 454.46, 454.81, 451.43, 449.39, 447.54, 434.22],
    σ_sep = [1.88, 2.06, 2.57, 23.9, 2.23, 5.41, 6.23, 3.03, 6.03, 2.02, 2.67, 2.15, 3.02, 2.01],
    pa    = [2.98835, 2.96723, 2.97038, 2.97404, 2.91994, 2.89934, 2.89131, 2.89184, 2.8962, 2.82394, 2.82272, 2.79357, 2.70927, 2.61171],
    σ_pa  = [0.00401426, 0.00453786, 0.00523599, 0.0523599, 0.00453786, 0.00994838, 0.00994838, 0.00750492, 0.00890118, 0.00453786, 0.00541052, 0.00471239, 0.00680678, 0.00401426]
)

astrom_like1 = PlanetRelAstromLikelihood(
    astrom_dat,
    name="VLT/SPHERE",
    variables=@variables begin
        # Fixed values for this example - could be free variables:
        jitter = 0        # mas [could use: jitter ~ Uniform(0, 10)]
        northangle = 0    # radians [could use: northangle ~ Normal(0, deg2rad(1))]
        platescale = 1    # relative [could use: platescale ~ truncated(Normal(1, 0.01), lower=0)]
    end
)
PlanetRelAstromLikelihood Table with 5 columns and 14 rows:
      epoch    sep     σ_sep  pa       σ_pa
    ┌────────────────────────────────────────────
 1  │ 57009.1  454.24  1.88   2.98835  0.00401426
 2  │ 57052.1  451.81  2.06   2.96723  0.00453786
 3  │ 57053.1  456.8   2.57   2.97038  0.00523599
 4  │ 57054.3  461.5   23.9   2.97404  0.0523599
 5  │ 57266.4  455.1   2.23   2.91994  0.00453786
 6  │ 57332.2  452.88  5.41   2.89934  0.00994838
 7  │ 57374.2  455.91  6.23   2.89131  0.00994838
 8  │ 57376.2  455.01  3.03   2.89184  0.00750492
 9  │ 57415.0  454.46  6.03   2.8962   0.00890118
 10 │ 57649.4  454.81  2.02   2.82394  0.00453786
 11 │ 57652.4  451.43  2.67   2.82272  0.00541052
 12 │ 57739.1  449.39  2.15   2.79357  0.00471239
 13 │ 58068.3  447.54  3.02   2.70927  0.00680678
 14 │ 58442.2  434.22  2.01   2.61171  0.00401426

We specify our full model:

planet_b_mass = Planet(
    name="b",
    basis=AbsoluteVisual{KepOrbit},
    likelihoods=[astrom_like1],
    variables=@variables begin
        a ~ truncated(Normal(10,1),lower=0.1)
        e ~ Uniform(0,0.99)
        ω ~ Uniform(0, 2pi)
        i ~ Sine()
        Ω ~ Uniform(0, 2pi)
        θ ~ Uniform(0, 2pi)
        M = system.M
        tp = θ_at_epoch_to_tperi(θ, 58442.2; M, e, a, i, ω, Ω)
        mass = system.M_sec
    end
)

sys_mass = System(
    name="cEri",
    companions=[planet_b_mass],
    likelihoods=[hip_like],
    variables=@variables begin
        M_pri ~ truncated(Normal(1.75,0.05), lower=0.03) # Msol
        M_sec ~ LogUniform(0.1, 100) # MJup
        M = M_pri + M_sec*Octofitter.mjup2msol # Msol

        rv =  12.60e3 # m/s
        plx ~ Uniform(20,40)
        pmra ~ Uniform(-100, 100)
        pmdec ~  Uniform(-100, 100)

        # It is convenient to put a prior of the catalog value +- 1000 mas on position
        ra_hip_offset_mas ~  Normal(0, 1000)
        dec_hip_offset_mas ~ Normal(0, 1000)
        dec = $hip_like.hip_sol.dedeg + ra_hip_offset_mas/60/60/1000
        ra = $hip_like.hip_sol.radeg + dec_hip_offset_mas/60/60/1000/cos(dec)

        ref_epoch = Octofitter.hipparcos_catalog_epoch_mjd
    end
)

model = Octofitter.LogDensityModel(sys_mass)
LogDensityModel for System cEri of dimension 13 and 118 epochs with fields .ℓπcallback and .∇ℓπcallback

Initialize the starting points, and confirm the data are entered correcly:

init_chain = initialize!(model, (;
    plx=34.,
    pmra=44.25,
    pmdec=-64.5,
    ra_hip_offset_mas=0.,
    dec_hip_offset_mas=0.,
))
octoplot(model, init_chain)

Now we sample:

using Pigeons
chain,pt = octofit_pigeons(model, n_rounds=8, explorer=SliceSampler())
chain
Chains MCMC chain (256×28×1 Array{Float64, 3}):

Iterations        = 1:1:256
Number of chains  = 1
Samples per chain = 256
Wall duration     = 533.14 seconds
Compute duration  = 533.14 seconds
parameters        = M_pri, M_sec, plx, pmra, pmdec, ra_hip_offset_mas, dec_hip_offset_mas, M, rv, dec, ra, ref_epoch, b_a, b_e, b_ω, b_i, b_Ω, b_θ, b_M, b_tp, b_mass, b_VLT_SPHERE_jitter, b_VLT_SPHERE_northangle, b_VLT_SPHERE_platescale
internals         = loglike, logpost, logprior, pigeons_logpotential

Summary Statistics
           parameters         mean        std      mcse   ess_bulk   ess_tail  ⋯
               Symbol      Float64    Float64   Float64    Float64    Float64  ⋯

                M_pri       1.7529     0.0517    0.0048   118.0062   132.4204  ⋯
                M_sec      17.2154    20.4594   11.3091     7.0298    23.4905  ⋯
                  plx      33.8934     0.3726    0.0411    79.1186   165.4535  ⋯
                 pmra      44.5345     0.5504    0.2273     6.9421    25.6704  ⋯
                pmdec     -64.2037     0.3449    0.0810    17.6640   239.5973  ⋯
    ra_hip_offset_mas      -3.8325     4.6423    2.5844     6.0804    19.1157  ⋯
   dec_hip_offset_mas      -0.9973     1.2300    0.5470    15.0987    50.9793  ⋯
                    M       1.7693     0.0566    0.0073    61.7633    93.9323  ⋯
                   rv   12600.0000     0.0000       NaN        NaN        NaN  ⋯
                  dec      -2.4734     0.0000    0.0000     6.0804    19.1157  ⋯
                   ra      69.4004     0.0000    0.0000    15.0987    50.9793  ⋯
            ref_epoch   48348.5625     0.0000       NaN        NaN        NaN  ⋯
                  b_a      10.5303     0.4647    0.2114     5.6638    25.9301  ⋯
                  b_e       0.5546     0.0502    0.0195     5.6025    39.2774  ⋯
                  b_ω       2.2904     1.6566    0.1166   171.7519    90.1619  ⋯
                  b_i       2.4113     0.0679    0.0306     5.1883    15.0387  ⋯
                  b_Ω       1.8415     1.9038    0.1338   188.2242    59.0667  ⋯
           ⋮                ⋮           ⋮          ⋮         ⋮          ⋮      ⋱
                                                    2 columns and 7 rows omitted

Quantiles
           parameters         2.5%        25.0%        50.0%        75.0%      ⋯
               Symbol      Float64      Float64      Float64      Float64      ⋯

                M_pri       1.6648       1.7183       1.7488       1.7877      ⋯
                M_sec       0.1228       1.0396       7.0044      29.9654      ⋯
                  plx      33.1879      33.6679      33.8618      34.1182      ⋯
                 pmra      43.6774      44.1216      44.4219      44.8935      ⋯
                pmdec     -64.7686     -64.4205     -64.2351     -64.0106      ⋯
    ra_hip_offset_mas     -14.9089      -6.8394      -1.5791      -0.2546      ⋯
   dec_hip_offset_mas      -3.8326      -1.7941      -0.4815      -0.0760      ⋯
                    M       1.6711       1.7298       1.7649       1.8065      ⋯
                   rv   12600.0000   12600.0000   12600.0000   12600.0000   12 ⋯
                  dec      -2.4734      -2.4734      -2.4734      -2.4734      ⋯
                   ra      69.4004      69.4004      69.4004      69.4004      ⋯
            ref_epoch   48348.5625   48348.5625   48348.5625   48348.5625   48 ⋯
                  b_a       9.8028      10.2072      10.4968      10.7581      ⋯
                  b_e       0.4706       0.5215       0.5570       0.5923      ⋯
                  b_ω       0.9496       1.0979       1.2165       4.1593      ⋯
                  b_i       2.2748       2.3662       2.3977       2.4462      ⋯
                  b_Ω       0.2349       0.3802       0.6155       3.5096      ⋯
           ⋮                ⋮            ⋮            ⋮            ⋮           ⋱
                                                     1 column and 7 rows omitted
octoplot(model, chain, show_mass=true)
Example block output

We see that we constrained both the orbit and the parallax. The mass is not strongly constrained by Hipparcos.