Note
Click here to download the full example code
Graphene hv scan¶
Simple workflow for analyzing a photon energy scan data of graphene as simulated from a third nearest neighbor tight binding model. The same workflow can be applied to any photon energy scan.
Import the “fundamental” python libraries for a generic data analysis:
import numpy as np
import matplotlib.pyplot as plt
Instead of loading the file as for example:
# from navarp.utils import navfile
# file_name = r"nxarpes_simulated_cone.nxs"
# entry = navfile.load(file_name)
Here we build the simulated graphene signal with a dedicated function defined just for this purpose:
from navarp.extras.simulation import get_tbgraphene_hv
entry = get_tbgraphene_hv(
scans=np.arange(90, 150, 2),
angles=np.linspace(-7, 7, 300),
ebins=np.linspace(-3.3, 0.4, 450),
tht_an=-18,
)
Plot a single analyzer image at scan = 90¶
First I have to extract the isoscan from the entry, so I use the isoscan method of entry:
iso0 = entry.isoscan(scan=90)
Then to plot it using the ‘show’ method of the extracted iso0:
iso0.show(yname='ekin')

Out:
<matplotlib.collections.QuadMesh object at 0x7f068ddae0a0>
Or by string concatenation, directly as:
entry.isoscan(scan=90).show(yname='ekin')

Out:
<matplotlib.collections.QuadMesh object at 0x7f068dd871c0>
Fermi level determination¶
The initial guess for the binding energy is: ebins = ekins - (hv - work_fun). However, the better way is to proper set the Fermi level first and then derives everything form it. In this case the Fermi level kinetic energy is changing along the scan since it is a photon energy scan. So to set the Fermi level I have to give an array of values corresponding to each photon energy. By definition I can give:
efermis = entry.hv - entry.analyzer.work_fun
entry.set_efermi(efermis)
Or I can use a method for its detection, but in this case, it is important to give a proper energy range for each photon energy. For example for each photon a good range is within 0.4 eV around the photon energy minus the analyzer work function:
energy_range = (
(entry.hv[:, None] - entry.analyzer.work_fun) +
np.array([-0.4, 0.4])[None, :])
entry.autoset_efermi(energy_range=energy_range)
Out:
scan(eV) efermi(eV) FWHM(meV) new hv(eV)
90.0000 85.4005 58.3 90.0005
92.0000 87.3999 59.3 91.9999
94.0000 89.4006 57.7 94.0006
96.0000 91.4002 59.8 96.0002
98.0000 93.4002 59.3 98.0002
100.0000 95.3996 59.7 99.9996
102.0000 97.4008 56.7 102.0008
104.0000 99.4007 59.6 104.0007
106.0000 101.4008 57.9 106.0008
108.0000 103.4002 58.9 108.0002
110.0000 105.4004 58.1 110.0004
112.0000 107.3996 60.4 111.9996
114.0000 109.3998 59.9 113.9998
116.0000 111.4001 59.0 116.0001
118.0000 113.3999 58.9 117.9999
120.0000 115.4006 57.8 120.0006
122.0000 117.4002 59.9 122.0002
124.0000 119.4000 59.3 124.0000
126.0000 121.3996 60.1 125.9996
128.0000 123.4001 59.4 128.0001
130.0000 125.4006 58.4 130.0006
132.0000 127.4008 58.3 132.0008
134.0000 129.3997 60.9 133.9997
136.0000 131.4005 58.6 136.0005
138.0000 133.4003 58.9 138.0003
140.0000 135.4004 58.5 140.0004
142.0000 137.4004 58.6 142.0004
144.0000 139.4005 57.9 144.0005
146.0000 141.4001 59.1 146.0001
148.0000 143.4006 58.1 148.0006
In both cases the binding energy and the photon energy will be updated consistently. Note that the work function depends on the beamline or laboratory. If not specified is 4.5 eV.
To check the Fermi level detection I can have a look on each photon energy. Here I show only the first 10 photon energies:
for scan_i in range(10):
print("hv = {} eV, E_F = {:.0f} eV, Res = {:.0f} meV".format(
entry.hv[scan_i],
entry.efermi[scan_i],
entry.efermi_fwhm[scan_i]*1000
))
entry.plt_efermi_fit(scan_i=scan_i)
Out:
hv = 90.00049742385123 eV, E_F = 85 eV, Res = 58 meV
hv = 91.99990331519464 eV, E_F = 87 eV, Res = 59 meV
hv = 94.00057487742069 eV, E_F = 89 eV, Res = 58 meV
hv = 96.0002003239522 eV, E_F = 91 eV, Res = 60 meV
hv = 98.00016513582878 eV, E_F = 93 eV, Res = 59 meV
hv = 99.99956872726189 eV, E_F = 95 eV, Res = 60 meV
hv = 102.00081477901384 eV, E_F = 97 eV, Res = 57 meV
hv = 104.00067479598724 eV, E_F = 99 eV, Res = 60 meV
hv = 106.0008095560272 eV, E_F = 101 eV, Res = 58 meV
hv = 108.00022009795744 eV, E_F = 103 eV, Res = 59 meV
Plot a single analyzer image at scan = 110 with the Fermi level aligned¶
entry.isoscan(scan=110).show(yname='eef')

Out:
<matplotlib.collections.QuadMesh object at 0x7f068dd48760>
Plotting iso-energetic cut at ekin = efermi¶
entry.isoenergy(0).show()

Out:
<matplotlib.collections.QuadMesh object at 0x7f068de92a30>
Plotting in the reciprocal space (k-space)¶
I have to define first the reference point to be used for the transformation. Meaning a point in the angular space which I know it correspond to a particular point in the k-space. In this case the graphene Dirac-point is for hv = 120 is at ekin = 114.3 eV and tht_p = -0.6 (see the figure below), which in the k-space has to correspond to kx = 1.7.
hv_p = 120
entry.isoscan(scan=hv_p, dscan=0).show(yname='ekin', cmap='cividis')
tht_p = -0.6
e_kin_p = 114.3
plt.axvline(tht_p, color='w')
plt.axhline(e_kin_p, color='w')
entry.set_kspace(
tht_p=tht_p,
k_along_slit_p=1.7,
scan_p=0,
ks_p=0,
e_kin_p=e_kin_p,
inn_pot=14,
p_hv=True,
hv_p=hv_p,
)

Out:
tht_an = -18.040
scan_type = hv
inn_pot = 14.000
phi_an = 0.000
k_perp_slit_for_kz = 0.000
kspace transformation ready
Once it is set, all the isoscan or iscoenergy extracted from the entry will now get their proper k-space scales:
entry.isoscan(120).show()

Out:
<matplotlib.collections.QuadMesh object at 0x7f068dfcfa90>
sphinx_gallery_thumbnail_number = 17
entry.isoenergy(0).show(cmap='cividis')

Out:
<matplotlib.collections.QuadMesh object at 0x7f068dfd6d30>
I can also place together in a single figure different images:
fig, axs = plt.subplots(1, 2)
entry.isoscan(120).show(ax=axs[0])
entry.isoenergy(-0.9).show(ax=axs[1])
plt.tight_layout()

Many other options:¶
fig, axs = plt.subplots(2, 2)
scan = 110
dscan = 0
ebin = -0.9
debin = 0.01
entry.isoscan(scan, dscan).show(ax=axs[0][0], xname='tht', yname='ekin')
entry.isoscan(scan, dscan).show(ax=axs[0][1], cmap='binary')
axs[0][1].axhline(ebin-debin)
axs[0][1].axhline(ebin+debin)
entry.isoenergy(ebin, debin).show(
ax=axs[1][0], xname='tht', yname='phi', cmap='cividis')
entry.isoenergy(ebin, debin).show(
ax=axs[1][1], cmap='magma', cmapscale='log')
axs[1][0].axhline(scan, color='w', ls='--')
axs[0][1].axvline(1.7, color='r', ls='--')
axs[1][1].axvline(1.7, color='r', ls='--')
x_note = 0.05
y_note = 0.98
for ax in axs[0][:]:
ax.annotate(
"$scan \: = \: {} eV$".format(scan, dscan),
(x_note, y_note),
xycoords='axes fraction',
size=8, rotation=0, ha="left", va="top",
bbox=dict(
boxstyle="round", fc='w', alpha=0.65, edgecolor='None', pad=0.05
)
)
for ax in axs[1][:]:
ax.annotate(
"$E-E_F \: = \: {} \pm {} \; eV$".format(ebin, debin),
(x_note, y_note),
xycoords='axes fraction',
size=8, rotation=0, ha="left", va="top",
bbox=dict(
boxstyle="round", fc='w', alpha=0.65, edgecolor='None', pad=0.05
)
)
plt.tight_layout()

Total running time of the script: ( 0 minutes 5.766 seconds)