mirror of
https://github.com/Richard-Sti/csiborgtools.git
synced 2024-12-22 18:28:02 +00:00
7f58b1f78c
* change to log10 initlogRs * add uncertainty * add check if hessian negative * update TODO * update TODO * output the error too * save e_logRs * add concentration calculation * calcul concentration * missed comma in a list * Update TODO * fix bug * add box units and pretty status * make uncertainty optional * add BIC function * remove BIC again * add new overdensity calculation * change defualt values to max dist and mass * change to r200 * new halo find * speed up the calculation * add commented fucn * update TODO * add check whether too close to outside boundary * Update TODO * extract velocity as well * calculate m200 and m500 * update nb * update TODO
340 KiB
340 KiB
In [1]:
import numpy as np
%matplotlib notebook
import matplotlib.pyplot as plt
try:
import csiborgtools
except ModuleNotFoundError:
import sys
sys.path.append("../")
import csiborgtools
import utils
%load_ext autoreload
%autoreload 2
import joblib
from os.path import join
from glob import glob
In [158]:
Nsim = 9844
simpath = csiborgtools.io.get_sim_path(Nsim)
Nsnap = 1016
outfname = join(utils.dumpdir, "ramses_out_{}_{}.npy".format(str(Nsim).zfill(5), str(Nsnap).zfill(5)))
mmain = csiborgtools.io.read_mmain(Nsim, "/mnt/zfsusers/hdesmond/Mmain")
data = np.load(outfname)
data = csiborgtools.io.merge_mmain_to_clumps(data, mmain)
data = data[(data["npart"] > 100) & np.isfinite(data["m200"])]
boxunits = csiborgtools.units.BoxUnits(Nsnap, simpath)
In [236]:
m200 = boxunits.box2solarmass(data["m200"])
m500 = boxunits.box2solarmass(data["m500"])
plt.figure()
plt.scatter(m200, m500, s=1, rasterized=True)
t = np.linspace(1e11, 1e15)
plt.plot(t,t,c="red", ls="--")
plt.yscale("log")
plt.xscale("log")
plt.xlabel(r"$M_{\rm 200c}$")
plt.ylabel(r"$M_{\rm 500c}$")
plt.savefig("../plots/masses.png", dpi=450)
plt.show()
In [229]:
logm200 = np.log10(boxunits.box2solarmass(data["m200"]))
conc = data["conc"]
N = 10
bins = np.linspace(logm200.min(), logm200.max(), N)
x = [0.5*(bins[i] + bins[i + 1]) for i in range(N-1)]
y = np.full((N - 1, 3), np.nan)
for i in range(N - 1):
mask = (logm200 >= bins[i]) & (logm200 < bins[i + 1]) & np.isfinite(conc)
y[i, :] = np.percentile(conc[mask], [14, 50, 84])
fig, axs = plt.subplots(nrows=2, sharex=True, figsize=(6.4, 6.4 * 1))
fig.subplots_adjust(hspace=0)
axs[0].plot(x, y[:, 1], c="red", marker="o")
axs[0].fill_between(x, y[:, 0], y[:, 2], color="red", alpha=0.25)
axs[1].hist(logm200, bins="auto", log=True)
for b in bins:
for i in range(2):
axs[i].axvline(b, c="orange", lw=0.5)
axs[0].set_ylim(2, 10)
axs[1].set_xlabel(r"$M_{200c}$")
axs[0].set_ylabel(r"$c_{200c}$")
axs[1].set_ylabel(r"Counts")
plt.tight_layout(h_pad=0)
plt.savefig("../plots/mass_concentration.png", dpi=450)
plt.show()