17 KiB

Raw Permalink Blame History

authors
JW, MT

Electron-positron annihilation

Calculation of positron lifetime and momentum distribution in silicon

This tutorial aims at showing how to perform Two-Component Density-Functional Theory (TCDFT) calculations in the PAW framework to obtain the following physical properties:

the positron lifetime in the perfect material,
the lifetime of a positron localized in a vacancy,
the electron-positron momentum distribution.

For the description of the implementation of TCDFT in ABINIT see cite:Wiktor2015.

The user should be familiar with the four basic tutorials of ABINIT and the first PAW tutorial.

This tutorial should take about 2 hours.

[TUTORIAL_README]

Computing the positron lifetime in Si lattice

Before beginning, you might consider to work in a different subdirectory as for the other tutorials. Why not Work_positron?

cd $ABI_TESTS/tutorial/Input
mkdir Work_positron
cd Work_positron
cp ../tpositron_1.abi .

The tutorial begins with a calculation of the positron lifetime in a silicon lattice. In a perfect material the positron is delocalized. We can assume that its density approaches zero and that it cannot affect the electron density. We will perform a calculation in only two steps:

Calculation of the ground-state electron density without the positron.
Calculation of the ground-state positron density in the presence of the electron density from step 1.

The two densities are used to calculate the positron lifetime, which is proportional to the inverse of the overlap of the electron and positron densities. This 2-step calculation, considering the zero-positron density limit, corresponds to the conventional scheme (CONV).

In the tpositron_1.abi file, you will find two datasets.

{% dialog tests/tutorial/Input/tpositron_1.abi %}

The first dataset is a standard ground-state calculation. The second one introduces a positron into the system. You can see that in this case we set:

  positron2  1      # Dataset 2 is a positronic GS calculation
  getden2    1      #   in presence of the previous electronic density

  kptopt2    0      #   Use only k=gamma point

  ixcpositron2 1    # We are using the Boronski and Nieminen parametrization

Here we set positron=1, which corresponds to a positronic ground-state calculation, considering that the electrons are not perturbed by the presence of the positron (zero-positron density limit). The electron density is read from the file resulting from dataset 1. As we consider the positron to be completely delocalized, we only consider the Γ point in the Brillouin zone. The keyword ixcpositron selects the electron-positron correlation functional and enhancement factor. In this calculation we use the functional parametrized by Boronski and Nieminen cite:Boronski1986, using the data provided by Arponen and Pajanne cite:Arponen1979.

We can now run the calculation. In the working directory Work_positron, copy the file tpositron_1.abi.

Then, issue:

abinit tpositron_1.abi >& log

This calculation should only take a few seconds.

You can look at the tpositron_1.abo file. We find the positron lifetime calculated in the RPA limit:

########## Lifetime computation 2

 # Zero-positron density limit of Arponen and Pajanne provided by Boronski & Nieminen
   Ref.: Boronski and R.M. Nieminen, Phys. Rev. B 34, 3820 (1986)
 # Enhancement factor of Boronski & Nieminen IN THE RPA LIMIT
   Ref.: Boronski and R.M. Nieminen, Phys. Rev. B 34, 3820 (1986)

 Positron lifetime                         (ps)   =  2.22879743E+02

The lifetime of 223 ps agrees well with the value of 225 ps calculated with the same number of valence electrons in cite:Wiktor2015 and with the experimental value of about 219 ps cite:Panda1997.

!!! important

If we had not used the "zero positron density limit" approximation
(using, for example, another value of [[ixcpositron]]), we would theoretically
have needed a box of infinite size for the positron to completely delocalise
itself in the crystal. This can be avoided by the use of the [[posocc]] input
parameter. Numerically, it is equivalent to calculate the density of 1
positron in a box of size N and that of x positron in a box of size N * x.
Thus, we can calculate the lifetime of a positron in a primitive cell by
setting [[posocc]] to a small value (0.0001 ...). This value must obviously be tested...

Computing the positron lifetime in a Si monovacancy

We will now perform a positron lifetime calculation for a monovacancy in silicon in the conventional scheme (which we applied to the perfect lattice previously). Note that when the positron localizes inside a vacancy, the zero-positron density limit does not apply anymore. However, in some cases, the conventional scheme proved to yield results in agreement with experiments.

For the purpose of this tutorial, we generate a defect in a cell containing only 16 atoms. This supercell is too small to get converged results, but the calculation is relatively fast.

{% dialog tests/tutorial/Input/tpositron_2.abi %}

You can now, issue:

abinit tpositron_2.abi >& log

Once the calculation is finished, look at the tpositron_2.abo file. Again, we look at the reported lifetime:

########## Lifetime computation 2

 # Zero-positron density limit of Arponen and Pajanne provided by Boronski & Nieminen
   Ref.: Boronski and R.M. Nieminen, Phys. Rev. B 34, 3820 (1986)
 # Enhancement factor of Boronski & Nieminen IN THE RPA LIMIT
   Ref.: Boronski and R.M. Nieminen, Phys. Rev. B 34, 3820 (1986)

 Positron lifetime                         (ps)   =  2.46923233E+02

We observe that when the positron localizes inside the vacancy, its lifetime increases from 223 to 247 ps. This is because now the majority of the positron density is localized in the vacancy region, where the electron density is small. The overlap of the electron and positron densities is reduced, and the lifetime increased.

In the Work_positron directory, you will also find a tpositron_2o_DS2_DEN_POSITRON file, containing the positron density. Visualizing this file (using e.g. cut3d and XcrysDen or VMD) you can see that the positron is localized inside the vacancy. You can see below how the positron (in red, isodensity at 30% of the maximum density) localized the silicon monovacancy looks like:

{width=80%}

Performing a self-consistent electron-positron calculation for a Si vacancy

We will now perform a self-consistent calculation of the positron and electron densities. As this calculation will take a few minutes, you can already issue, using the tpositron_3.abi input file:

abinit tpositron_3.abi >& log

{% dialog tests/tutorial/Input/tpositron_3.abi %}

This calculation is significantly longer than the previous one, because the electron and positron steps will be repeated until the convergence criterion is reached.

In tpositron_3.abi we only have one dataset and we set positron = -10 to perform an automatic calculation of electrons and positron densities. The convergence is controlled by postoldfe = 1d-5. This means that we will repeat the electron and positron steps until the energy difference between them is lower than 1d-5 Ha. This value should always be larger than toldfe. In this calculation we still use ixcpositron = 1, which means that we are using the GGGC scheme (see cite:Gilgien1994 and cite:Wiktor2015

Once the calculation is finished, look at the positron lifetime in tpositron_3.abo.

########## Lifetime computation 2

 # Zero-positron density limit of Arponen and Pajanne provided by Boronski & Nieminen
   Ref.: Boronski and R.M. Nieminen, Phys. Rev. B 34, 3820 (1986)
 # Enhancement factor of Boronski & Nieminen IN THE RPA LIMIT
   Ref.: Boronski and R.M. Nieminen, Phys. Rev. B 34, 3820 (1986)

 Positron lifetime                         (ps)   =  2.55612619E+02

Including the self-consistency increases the positron lifetime, because its localization inside the vacancy becomes stronger when the positron and the electron densities are allowed to relax.

Relaxing the vacancy according to forces due to electrons and the positron

In addition to the self-consistency, the lifetime of a positron inside a vacancy can be strongly affected by the relaxation of the atoms due to the forces coming from both the electrons and the positron. You can already start the relaxation (with the tpositron_4.abi input file) of the vacancy by issuing:

abinit tpositron_4.abi >& log

In this calculation we switched on the atomic relaxation by setting ionmov = 2. We need to calculate forces to be able to move the atoms, so we set optforces = 1. In the provided tpositron_4.abi file, we only perform 4 relaxation steps (ntime = 4) to save time, but more steps would be needed to converge the positron lifetime.

{% dialog tests/tutorial/Input/tpositron_4.abi %}

Look at the positron lifetime in the RPA limit after each ionic step:

 Positron lifetime                         (ps)   =  2.55612619E+02
 Positron lifetime                         (ps)   =  2.56978378E+02
 Positron lifetime                         (ps)   =  2.82166606E+02
 Positron lifetime                         (ps)   =  2.82878399E+02

As the vacancy relaxes outwards, the positron lifetime increases. 4 steps were not enough to relax the defect completely, as the lifetime still changes. Indeed, setting ntime to 10 delivers:

 Positron lifetime                         (ps)   =  2.55612619E+02
 Positron lifetime                         (ps)   =  2.56978379E+02
 Positron lifetime                         (ps)   =  2.82166601E+02
 Positron lifetime                         (ps)   =  2.82878398E+02
 Positron lifetime                         (ps)   =  2.86515373E+02
 Positron lifetime                         (ps)   =  2.86983434E+02
 Positron lifetime                         (ps)   =  2.87266489E+02
 Positron lifetime                         (ps)   =  2.87359897E+02
 Positron lifetime                         (ps)   =  2.87313132E+02

Although the results at ionic steps 3 and 4 differ from each other by less than one percent, they differ by more from the final result. The one percent convergence is only reached at ionic step 5 and after.

Also, remember that the 16-atom supercell is not large enough to get converged results. In Table IV of cite:Wiktor2015 you can see converged results of the positron lifetime of Si monovacancy within various methods.

Computing the electron-positron momentum distribution (Doppler spectrum) of a Si lattice

In the last part of the tutorial we will calculate the electron-positron momentum distribution (Doppler spectrum) of a silicon lattice in the conventional scheme. This type of calculation is much more time and memory consuming than the lifetime calculation, as it is using the electron and positron wavefunctions (not only densities).

You can already issue:

abinit tpositron_5.abi >& log

Now take a look at the input file tpositron_5.abi.

{% dialog tests/tutorial/Input/tpositron_5.abi %}

The momentum distribution calculation is activated by posdoppler = 1. You can also notice that instead of having two datasets as in the first part of this tutorial, we now use the automatic electron-positron loop and set posnstep = 2. This is done because we need to have the full electron and positron wavefunctions in memory, which is only the case when positron <= -10. Additionally, the momentum distribution calculations require using a full k-point grid. In the input file we set:

 kptopt 0          # Option for manual setting of k-points
 istwfk *1         # No time-reversal symmetry optimization
 nkpt 8            # Corresponds to a 2x2x2 grid
 kpt               # K-point coordinates in reciprocal space:
   0   0   0
   0   0   0.5
   0   0.5 0
   0.5 0   0
   0   0.5 0.5
   0.5 0   0.5
   0.5 0.5 0
   0.5 0.5 0.5

This grid is used in both electron and positron calculations, but only the positron wavefunction at the first point is taken in the momentum distribution calculation, so the \Gamma point should always be given first.

In the calculation of the momentum distribution, we need to include both core and valence electrons. The wavefunctions of the core electrons are read from a file (one per atom type), which needs to be provided. This core WF file should be named <psp_file_name>.corewf.xml (where <psp_file_name> is the name of the PAW atomic dataset file, without .xml suffix). Core WF files can be obtained with the atompaw tool (see the tutorial on generating PAW datasets (PAW2) ) by the use of the prtcorewf keyword. You will find the core wavefunction file used in this calculation in $ABI_PSPDIR/Si.LDA-PW-paw.abinit.corewf.

!!! Note If you use a PAW dataset in ABINIT legacy proprietary format (with the .abinit suffix), the core wavefunction file has to be named <psp_file_name>.corewf.abinit. It also can be obtained with the atompaw tool by the use of the prtcorewf keyword.

Once the calculation is complete, you can find a tpositron_5o_DOPPLER file containing the momentum distribution on the FFT grid. You can use the $ABI_HOME/scripts/post_processing/posdopspectra.F90 tool to generate 1D projections (Doppler spectra) in (001), (011) and (111) directions and to calculate the low- and high-momentum contributions to the momentum distribution (so called S and W parameters, see cite:Wiktor2015).

Studying the effect of the PAW dataset completeness

The positron lifetime and momentum distribution calculations within the PAW method are very sensitive to the number of valence electrons in the PAW dataset. It is due to the fact that it is not easy to describe the positron wavefunction, tending to zero at the nucleus, using the electron atomic orbitals. The PAW basis set in this case needs to be more complete than only for describing the electron wavefunctions.

The simplest way to make the PAW dataset more complete is to include semicore electrons. It is also possible to add the partial waves corresponding to the semicore electrons in the basis used only for the positron wave function description, while keeping the initial number of valence electrons (as done in cite:Wiktor2015). However, this second method is less straightforward.

The previous calculations were done with only 4 valence electrons (3s and 3p). We will now see what happens if we include the 2s and 2p states in the PAW dataset. We use the Si_paw_pw_12el.xml PAW dataset which includes 8 additional valence electrons.

!!!Tip To generate the new dataset we use the atompaw tool.
To add semicore states, the input file is modified as follows:

 - Replace `c` by `v` for the selected orbitals in the _electronic configuration_ section
 - Decrease the PAW augmentation radius (because semicore states are more localized)

 Don't forget to add `prtcorewf` keyword to create the core orbital file.

We can now rerun the lifetime calculation with the new atomic dataset:

{% dialog tests/tutorial/Input/tpositron_6.abi %}

abinit tpositron_6.abi >& log

We now find the positron lifetime calculated in the RPA limit:

########## Lifetime computation 2

 # Zero-positron density limit of Arponen and Pajanne provided by Boronski & Nieminen
   Ref.: Boronski and R.M. Nieminen, Phys. Rev. B 34, 3820 (1986)
 # Enhancement factor of Boronski & Nieminen IN THE RPA LIMIT
   Ref.: Boronski and R.M. Nieminen, Phys. Rev. B 34, 3820 (1986)

 Positron lifetime                         (ps)   =  2.11470610E+02

This value is significantly lower than 223 ps achieved with 4 valence electrons in the first step. It is, therefore, very important to always test the PAW dataset completeness for positron calculations.

The PAW dataset completeness is even more important in the Doppler spectra calculations. We will now recalculate the momentum distribution including 12 valence electrons using tpositron_7.abi:

{% dialog tests/tutorial/Input/tpositron_7.abi %}

abinit tpositron_7.abi >& log

Before processing the new tpositron_7o_DOPPLER file, you should copy files rho_001, rho_011, rho_111 from the fifth step to for instance si4el_001, si4el_011 and si4el_111. By plotting the Doppler spectra in the (001) direction calculated with 4 and 12 valence electrons, you should obtain a figure like this:

{width=60%}

The dataset with 4 valence electrons is not complete enough to describe the positron wavefunction around the nucleus. This is reflected in the unphysically high probability at high momenta in the spectrum.

Further explanation of the influence of the PAW dataset on the Doppler spectra can be found in cite:Wiktor2015. In case you need to generate your own dataset for momentum distribution calculations, you can follow the tutorial on generating PAW datasets (PAW2).

17 KiB Raw Permalink Blame History