Phononlimited mobility in graphene using the Boltzmann transport equation¶
Version: Q2019.12SP1
In this tutorial you will learn how to calculate the phononlimited mobility in graphene. The mobility will be calculated using the Boltzmann transport equation (BTE) with the electronic structure, phonons and electronphonon coupling calculated using density functional theory (DFT).
The mobility \(\mu\) will be calculated using two different methods to calculate the relaxation times \(\tau\) entering the BTE:
 Full angular (k,q)dependence: in this method, the full dependency of \(\tau\) on the electron and phonon wave vectors \(\mathbf{k}\) and \(\mathbf{q}\) is taken into account, so that \(\tau = \tau(\mathbf{k},\mathbf{q})\). In the following, we will refer to this as the (k,q)dependent method.
 Isotropic scattering rate: in this method, only the energy dependence of \(\tau\) is considered, so that \(\tau = \tau(E)\), and \(\tau(E)\) is assumed to vary isotropically in momentumspace. In the following, we will refer to this as the Edependent method.
As it will be shown, the two methods give essentially the same results, but the second method is considerably faster than the first one.
In the theory section, you can read about the theoretical background. A more comprehensive description can also be found in the paper [GMSB16].
Geometry and electronic structure of graphene¶
In the Builder, add a graphene configuration by clicking and searching for ‘Graphene‘ in the structure database.
Next, you have to increase the vacuum gap above and below the graphene. Click on
and set the lattice parameter along the Cdirection to C = 20 Å.Center the configuration by clicking on button to send the structure to the Script Generator. In the main panel of the Script Generator, set the Results file to
Graphene_relax.hdf5
,
To perform an accurate calculation of the relaxation times \(\tau\) and the mobility \(\mu\), the first important step is to optimize the lattice parameters A and B and calculate the electronic band structure. In order to do this, add an LCAOCalculator block and in the Main section set the following parameters:

Set the Exchange correlation to LDA
Set the Pseudopotential to FHI
Set the Occupation method to MethfesselPaxton
Set the Density mesh cutoff to 90 Ha
Set the kpoint sampling to:
 \(\mathrm{k_A}\) = 33
 \(\mathrm{k_B}\) = 33
 \(\mathrm{k_C}\) = 1
On the Iteration Control tab, Set the Tolerance to 0.000001, i.e. 10^{}^{6}

 Set the Force tolerance to 0.001 eV/Å
 Untick Fix lattice vectors in the \(x\) and \(y\) directions.

 Set the number of Points per segment to 100
 Set the Brillouin zone route to G, K, M, G
Now, send the script to the Job manager, save it as Graphene_relax.py
, and click on the
button to run the calculation.
When the calculation is done, click on the Bandstructure object contained
in the file Graphene_relax.hdf5
on the LabFloor and use the Bandstructure Analyzer to visualize the
band structure.
By placing the mouse cursor on top of a band, information about the band is shown. You see that the valence band (highlighted in yellow in the figure above) is number 3 and the conduction band is number 4. These are the two electronic bands relevant for the calculation of the mobility, and in the following we will concentrate on these two bands.
Phonons in Graphene¶
The next step is to calculate the dynamical matrix of graphene. In order to test the quality of the result you will also calculate the phonon band structure, which is based on the calculated dynamical matrix.
Open the Script generator , and create a new script:
 Add an block and select BulkConfiguration_1 in
Graphene_relax.hdf5
.  Add a block and modify the following settings:
 Set Repetitions to Custom
 Set the Number of repetitions to:
 \(\mathrm{n_A}\) = 11
 \(\mathrm{n_B}\) = 11
 \(\mathrm{n_C}\) = 1
Note
In previous versions of QuantumATK it was necessary to manually scale down the number of kpoints when doing a DynamicalMatrix or HamiltonianDerivatives calculation to take into account the repeated cell. From QuantumATK2019.03 and later versions the kpoints are automatically scaled to maintain the same accuracy as the unit cell, and we can simply reuse the calculator settings.
Tip
Due to the large dimensions of the \(11 \times 11\) graphene super cell (242 atoms), the calculation of the DynamicalMatrix object is rather time consuming. However, the calculation can be parallelized over the atomic displacements. Since there are two atoms in the graphene unit cell, and each is displaced in the \(x\), \(y\) and \(z\) directions, there are in total 6 calculations to be performed. Maximum efficiency is obtained when the number of calculations times the value of the parameter Processes per displacement matches the total number of cores used for the calculations. In the present case, the calculation takes about 30 minutes if Processes per displacement = 4 and the calculation is run on 24 cores.
Add an block and set the following parameters:
 Set the number of Points per segment to 100
 Set the Brillouin zone route to G, M, K, G
Finally, in the main panel of the Script Generator set the Default output file to
Graphene_dynmat.hdf5
, send the script to the
Job manager, save it as Graphene_dynmat.py
and click on the button to run the calculation.
Warning
Remember to make the file Graphene_relax.hdf5
available for the script if you are running on a cluster. In the Job Manager you can use the I/O tab to transfer additional files together with a script. In this case, you may also need to modify the path in the script before submitting to the cluster, to make sure it points at the correct location of the file.
When the calculation is done, go back to the in the LabFloor, and inspect the
PhononBandstructure object contained in the file Graphene_dynmat.hdf5
using Compare data or
PhononBandstructure Analyzer. The calculated phonon band structure should match with that shown in the figure
below.
Mobility of graphene¶
In the following, the procedure to calculate the electron mobility in graphene will be described. Provided that one has already calculated the electronic structure and dynamical matrix of the system, the following three steps are necessary to evaluate the mobility:
 Calculation of the Hamiltonian derivatives
 Calculation of the Electronphonon couplings
 Calculation of the Mobility
The two methods to calculate \(\mu\) described above differ in the way in which steps 2 and 3 are carried out. These two steps will therefore be described separately for each method. The procedure for the (k, q)dependent method is outlined in sections 1, 2A and 3A, whereas the procedure for the energydependent method is outlined in in sections 1, 2B and 3B
1. Hamiltonian derivatives¶
In order to calculate the electronphonon coupling matrix, it is necessary to calculate the derivative of the Hamiltonian \(\partial \hat{H}/\partial R_{i,\alpha}\) with respect to the coordinate of the \(i\)th atom along the Cartesian direction \(\alpha\).
Open the Script generator , and modify the script as follows:
Add an block and select BulkConfiguration_1 in
Graphene_relax.hdf5
.Add a block and set the Number of repetitions to:
 \(\mathrm{n_A}\) = 11
 \(\mathrm{n_B}\) = 11
 \(\mathrm{n_C}\) = 1
Warning
It is important that the number of repetitions matches the calculation of the dynamical matrix.
In the main panel of the Script Generator, set the Results file to
Graphene_dHdR.hdf5
, send the script to the Job manager, save it as
Graphene_dHdR.py
and click on the button to run the calculation.
Note
As for the DynamicalMatrix, the calculation of the HamiltonianDerivatives is rather time consuming, but can be parallelized over the atomic displacements. In the present case, the calculation takes around 1 hour if it is run on 24 cores with Processes per displacement = 4 .
2A. ElectronPhonon couplings: (k,q)dependent method¶
In order to calculate the lifetimes \(\tau(\mathbf{k},\mathbf{q})\) and the mobility \(\mu\), we need to calculate the electronphonon coupling matrix on a fine grid of \(\mathbf{k}\) and \(\mathbf{q}\)points.
Open the Script generator , and modify the script as follows:
You will notice that two additional blocks have been also added:
In the present case, both the dynamical matrix and the Hamiltonian derivatives have been calculated already and can be
reused. As they are study objects, this is automatically detected by QuantumATK if the provided filename and other input parameters are the same. Open each of them and change the repetitions to 11x11x1 and the filenames to Graphene_dynmat.hdf5
and Graphene_dHdR.hdf5
, respectively.
Now set the parameters in the block as shown below. We will sample kpoints in a small area around the \(\mathrm{K}\)point at [1/3, 1/3, 0], and qpoints in a small area around the \(\mathrm{\Gamma}\)point at [0,0,0].
Tip
If you are unsure of the coordinates of a particular symmetry point you may use the builtin functionality of the BravaisLattice
class, f.ex. like this:
k0=bulk_configuration.bravaisLattice().symmetryPoints()['K']
)
Warning
For production calculations, it is strongly recommended to always converge the sampling resolution in q and kspace. The above settings are a result of such a study, with more information shown in the appendix: Convergence of q and kpoint sampling
In the main panel of the Script Generator, set the Results file to
Graphene_eph.hdf5
. Send the script to the Job manager, save it as
Graphene_eph.py
and click on the button to run the calculation.
Warning
For graphene, we need quite dense samplings in \(\mathbf{k}\) and \(\mathbf{q}\)space, and therefore we have chosen to sample very finely a small area around the relevant highsymmetry points. This still includes the full angular dependence, but neglects intervalley scattering. We therefore recommend to sample the entire Brillouin zone if this could be an important effect.
Note
Sampling the \(\mathbf{k}\) and \(\mathbf{q}\)space using dense meshes means that the calculation might become very time consuming. Contemporary versions of QuantumATK automatically detect the relevant regions of \(\mathbf{k}\) and \(\mathbf{q}\)space, and do not compute matrix elements which will not contribute, considerably decreasing the computational time and memory requirements. The two parameters energy_tolerance and initial_state_energy_range govern the filters which reduce the number of calculated coupling elements. More information about these two parameters can be found on the manual page: ElectronPhononCoupling
.
As QuantumATK parallelizes over \(\mathbf{k}\) and \(\mathbf{q}\)points, a high number of MPI processes can be used if sufficient memory is available. In the present case, the calculation took approximately 10 hours on a 16core node on a cluster.
3A. Mobility: (k,q)dependent method¶
Now open a new Script Generator window and add an block and then a block. In the block, select the file Graphene_relax.hdf5
and load BulkConfiguration_1 included in the file. Remove the , and blocks, and replace them with a block. This allows us to easily add arbitrary code to any script without editing the full script manually. Open the block, and write the following:
electron_phonon_coupling = nlread('Graphene_eph.hdf5', ElectronPhononCoupling)[0]
Set the parameters for the block as shown below:
 Leave the Method at Full angular (k,q)dependence
 Set the Fermi shift to 0.13 eV
 Untick Calculate Hall coefficients
Note
The Fermi shift of 0.13 eV corresponds to a carrier concentration of \(n = 10^{12} \mathrm{cm^{2}}\)
In the main panel of the Script Generator, set the Results file to
Graphene_mufull.hdf5
, send the script to the Job manager, save it as
Graphene_mufull.py
and click on the button to run the calculation. It should take no more than a few minutes on a laptop.
Note
You will get a notification that the Mobility block is invalid. You can ignore this warning, as NanoLab does not parse the Code snippet block and is thus unaware of the ElectronPhononCoupling
object that will be loaded there.
Once the calculation is done, select the file Graphene_mufull.hdf5
, and on the LabFloor, select the
Mobility object, and click on Text Representation. You should now see the following:
++
 Mobility Report 
  
 Input parameters: 
 Temperature = 300.00 K 
 Fermi level shift = 0.13 eV 
 Energy broadening = 0.0030 eV 
 qgrid refinement = 1 
 
++
 Trace of linear responce tensors: 
++
 
 Electrons: 
 
 Mobility = 2.34e+05 cm^2/(V*s) 
 Conductivity = 4.19e+01 S/m 
 Seebeck coefficient = 1.94e05 V/K 
 Thermal conductivity = 2.43e04 W/(m*K) 
 Carrier density (2D, xy) = 2.24e+06 cm^2 
 
 
 Holes: 
 
 Mobility = 8.66e04 cm^2/(V*s) 
 Conductivity = 1.19e01 S/m 
 Seebeck coefficient = 1.14e03 V/K 
 Thermal conductivity = 1.18e05 W/(m*K) 
 Carrier density (2D, xy) = 1.72e+12 cm^2 
 
++
The calculated electron mobility, highlighted in yellow, is \(2.34 \cdot 10^{5}\ \mathrm{cm^{2} V^{1} s^{1}}\), and agrees well with previously reported data at room temperature and \(n = 10^{12} \mathrm{cm^{2}}\) [KTJ12].
Note
Note that the carrier density listed here is significantly lower than \(n = 10^{12} \mathrm{cm^{2}}\). The carrier density is calculated independently from the mobility, and converges much more slowly than the mobility itself. In this tutorial, the focus is on achieving a converged value for the mobility  converging also the carrier density would require inclusion of more \(\mathbf{k}\)points and/or a larger region in \(\mathbf{k}\)space in the ElectronPhononCoupling calculation.
Alternatively, you can calculate the carrier density with a finer sampling directly from the DensityOfStates object: carrier_density_test.py
.
Tip
Starting from version Q2019.12, QuantumATK NanoLab includes a Mobility Analyzer. This will be presented in a future update of this tutorial.
2B. ElectronPhonon couplings: energydependent method¶
We will now calculate the electronphonon couplings to be used for the calculation of the energydependent relaxation times \(\tau(E)\). We will assume that the relaxation times vary isotropically in \(\mathbf{k}\)space. This means that it will be sufficient to evaluate the electronphonon coupling matrix for a number of \(\mathbf{k}\)points along a line through one of the Dirac (K) points, thereby reducing significantly the computational workload of the calculation.
Open the Script generator , and modify the script as follows:
You will notice that two additional blocks have been also added:
In the present case, both the dynamical matrix and the Hamiltonian derivatives have been already calculated and can be
reused. As they are study objects, this is automatically detected by QuantumATK if the provided filename and other input parameters are the same. Open each of them and change the repetitions to 11x11x1 and the filenames to Graphene_dynmat.hdf5
and Graphene_dHdR.hdf5
, respectively.
In the block, modify the parameters as follows:
In the kpoint sampling, set the Grid type to Regular kpoint grid, and the sampling parameters as shown in the figure below.
Note
As you can see from the figure, the \(\mathbf{k}\)space is sampled only along a line!
In the qpoint sampling, set the Grid type to Regular qpoint grid, and the sampling parameters as shown in the figure below.
Set the Energy tolerance to 0.01 (eV) and the Initial state energy range to go from 0.5 to 0.5.
In the main panel of the Script Generator, set the Results file to
Graphene_ephline.hdf5
, send the script to the Job manager, save it as Graphene_ephline.py
and click on the button to run the calculation.
Note
Since here we sample the \(\mathbf{k}\)space only along a line, the calculation will in general be faster than the one described in section 2A. In the present case, the calculation took approximately 50 minutes on a 16core node on a cluster, or about 1/10th of the full calculation in section 2A.
3B. Electron mobility: energydependent method¶
We will now use a twostep procedure to evaluate the roomtemperature mobility \(\mu\) based on the energydependent relaxation times \(\tau(E)\):
 In the first step, the \(\mathbf{k}\) and \(\mathbf{q}\)dependent relaxation times \(\tau(\mathbf{k},\mathbf{q})\) are evaluated along the line outward from the Kpoint.
 In the second step, the values of \(\tau(\mathbf{k},\mathbf{q})\) are averaged in \(\mathbf{k}\)space to obtain the values of \(\tau(E)\), which are then used to calculate the mobility \(\mu\).
Now open a new Script Generator window and add an block and then a block. In the block, select the file Graphene_relax.hdf5
and load BulkConfiguration_1 included in the file. Remove the and objects and replace the with a block. Open the block, and write the following:
electron_phonon_coupling = nlread('Graphene_ephline.hdf5', ElectronPhononCoupling)[0]
Finally, set the parameters in the block as shown below:

 Set the Method to Full angular (k,q)dependence
 Set the Fermi shift to 0.13 eV
 Untick Calculate Hall coefficients
Note
Note that we choose a \(\mathbf{k}\)line direction somewhat arbitrarily here. The point is that a low number of \(\mathbf{k}\)points can be used to generate an energy dependent rate. An alternative to a line could be to coarsely sample the entire BZ to capture the anisotropy in \(\mathbf{k}\)space, but with fewer points than needed in the full (\(\mathbf{k}\), \(\mathbf{q}\))dependent method.
In the main panel of the Script Generator, set the Results file to
Graphene_mulinefull.hdf5
, send the script to the Job manager,save it as
Graphene_mulinefull.py
and click on the button to run the calculation. You will again see a warning that the script might be invalid, and you should ignore it this time as well.
Once the calculation is done, you are ready to calculate the mobility using the isotropic scattering rate method.
In this case, you will reuse the bulk configuration and the full angular (\(\mathbf{k}\), \(\mathbf{q}\)) dependent mobility from the
Graphene_relax.hdf5
and Graphene_mulinefull.hdf5
files.
Go back to the Script generator, and open the block before the Mobility icon . Change the contents to:
mobility_full = nlread('Graphene_mulinefull.hdf5', Mobility)[0]
Then modify the block as follows:

 Set the Method to Isotropic scattering rate
 Set the Fermi shift to 0.13 eV
 Untick Calculate Hall coefficients
 Set the following values of the Energy Range:
 \(\mathrm{E_0}\) = 0.24 eV
 \(\mathrm{E_0}\) = 0.24 eV
 Points = 100
 Set the kpoint Sampling to \(99 \times 99 \times 1\)
In the main panel of the Script Generator, set the Results file to
Graphene_mulineiso.hdf5
. Send the script to the Editor, change mobility_object=None,
to mobility_object=mobility_full,
and remove this line:
inverse_relaxation_time=numpy.linspace(0, 1e+12, 100)*Second**1,
Send it to the Job manager, save it as Graphene_mulineiso.py
and click on
the button to run the calculation. This should take no more than a minute on a modern laptop.
Once the calculation is done, open the file Graphene_mulineiso.hdf5
, select the
Mobility object in the LabFloor and click on Text Representation.
++
 Mobility Report 
  
 Input parameters: 
 Temperature = 300.00 K 
 Fermi level shift = 0.13 eV 
 Energy broadening = 0.0030 eV 
 qgrid refinement = 1 
 
++
 Trace of linear responce tensors: 
++
 
 Electrons: 
 
 Mobility = 2.53e+05 cm^2/(V*s) 
 Conductivity = 1.73e+07 S/m 
 Seebeck coefficient = 3.00e05 V/K 
 Thermal conductivity = 9.31e+01 W/(m*K) 
 Carrier density (2D, xy) = 8.55e+11 cm^2 
 
 
 Holes: 
 
 Mobility = 1.89e+06 cm^2/(V*s) 
 Conductivity = 5.72e+04 S/m 
 Seebeck coefficient = 1.14e03 V/K 
 Thermal conductivity = 5.74e+00 W/(m*K) 
 Carrier density (2D, xy) = 3.78e+08 cm^2 
 
++
The calculated value for the electron mobility, highlighted in yellow above, is \(2.53 \cdot 10^{5}\ \mathrm{cm^{2} V^{1} s^{1}}\), in good agreement with the value obtained by using the full (\(\mathbf{k}\), \(\mathbf{q}\)) dependent method.
Temperature dependence of the mobility: (k,q)dependent method vs. energydependent method¶
A more stringent test for the reliability of the energydependent method is to calculate the temperature dependence of the mobility in an energy range up to room temperature.
 In the (k,q)dependent method, this can be done by simply modifying the value of the target temperature in the
Mobility analysis object.
temperature_dependence_full_mobility.py
 In the energydependent method, the twostep procedure must be repeated for each temperature, and the target
temperature has to be set both when calculating the values of the \(\mathbf{k}\) and
\(\mathbf{q}\)dependent relaxation times \(\tau(\mathbf{k},\mathbf{q})\) along the line, and when
calculating the energydependent relaxation times \(\tau(E)\).
temperature_dependence_isotropic_mobility.py
As it can be seen from the figure below, where the \(T\)dependency of \(\mu\) is calculated in the
temperature range \(100 \mathrm{K} \leq T \leq 300 \mathrm{K}\), both methods reproduce well the expected
\(1/T\) behavior of \(\mu\). Script to create the plot: plot_temperature_dependence_mobility.py
Convergence of q and kpoint sampling¶
In order to select the appropriate samplings in \(\mathbf{k}\) and \(\mathbf{q}\)space, we first studied the mobility as a function of the number of \(\mathbf{q}\)points, for a fixed sampling of \(\mathbf{k}\)points:
We see that the mobility is reasonably converged for 50 \(\mathbf{q}\)points and above. This corresponds to a density of at least 200 \(Å\). We then study the mobility as a function of \(\mathbf{k}\)points, and see that it converges very quickly in this case. We chose to use 20 kpoints, corresponding to a density of about 160 \(Å\).
Warning
This convergence study is only valid for this combination of system and computational settings. You should always study convergence for your system of interest and chosen computational model.
Theory section¶
The mobility \(\mu\) is related to the conductivity \(\sigma\) via
where \(n\) is the carrier density, \(e\) is the electronic charge.
In this tutorial, you will calculate the conductivity using the semiclassical Boltzmann transport equation (BTE). In the case of zero magnetic field and a timeindependent electric fields in the steady state limit the BTE simplifies to:
The right hand side (RHS), often called the collision integral, describes different sources of scattering and dissipation that drives the system towards steady state. The left handside is approximated to linear order in the electric field by changing to the equilibrium distribution.
The electronphonon scattering is described by the collision integral. Most commonly this is approximated by a relaxation time:
describing quasielastic scattering on acoustic phonons (relaxation time approximation). We let \(\lambda\) label the phonon modes, \(n\) the electronic bands, \(\mathbf{k}\) the electron momentum. The transition rate from a state \(\mathbf{k}n\rangle\) to \(\mathbf{k}'n'\rangle\) is obtained from Fermi’s golden rule (FGR):
\[\begin{split}P_{\mathbf{k}\mathbf{k'}}^{\lambda nn'} &= \frac{2\pi}{\hbar} g_{\mathbf{k}\mathbf{k'}}^{\lambda n n'}^2 [ n_{\mathbf{q}}^{\lambda} \delta \left(\epsilon_{\mathbf{k}'n'}\epsilon_{\mathbf{k}n}\hbar \omega_{q \lambda} \right) \delta_{\mathbf{k}',\mathbf{k}+\mathbf{q}} \\[.5cm] &+ (n_{\mathbf{q}}^{\lambda}+1) \delta \left( \epsilon_{\mathbf{k}'n'}\epsilon_{\mathbf{k} n}+\hbar \omega_{q \lambda} \right) \delta_{\mathbf{k}',\mathbf{k}\mathbf{q}} ],\end{split}\]
where the first and last terms in the square brackets describes phonon absorption and emission, respectively. The electronphonon coupling matrix elements \(g_{\mathbf{k}\mathbf{k'}}^{\lambda n n'}\) are in ATK calculated using the ElectronPhononCoupling analysis module.
The general electronphonon collision integral is given by
Relaxation time approximation¶
In the relaxationtime approximation (RTA) one assumes a special simplified form of the RHS/collision integral:
where \(\delta f_{\mathbf{k}n} = f_{\mathbf{k}n}f^0_{\mathbf{k}n}\). The linearized BTE becomes:
Hereby we find the solution:
The relaxationtime can be evaluated from the general collision integral, (3). However, the specific form in eqn. (4) only applies in the limit of quasielastic scattering (\(\omega_{q \lambda}\rightarrow 0\)), in which case the collision integral, in eqn. (3), simplifies significantly:
since \(P_{\mathbf{k}\mathbf{k'}}^{nn'} = P_{\mathbf{k}'\mathbf{k}}^{n'n}\) in this limit.
We then obtain the expression for the scattering rate or inverse relaxationtime:
where we applied the detailed balance equation in this limit, \(P_{\mathbf{k}\mathbf{k'}}^{nn'}(f^0_{\mathbf{k}'n'}f^0_{\mathbf{k}n})=0\), and assumed isotropic scattering.
To evaluate the relaxation time we introduce the scattering angle
where \(\mathbf{v}_{\mathbf{k}n}\) is the electron velocity. Following [GMSB16] we obtain
Mobility¶
Once knowing the relaxation times one obtain the mobility as:
where the factor 2 accounts for spin degeneracy.
References¶
[GMSB16]  (1, 2) T. Gunst, T. Markussen, K. Stokbro, and M. Brandbyge. Firstprinciples method for electronphonon coupling and electron mobility: Applications to twodimensional materials. Phys. Rev. B, 93:035414, Jan 2016. doi:10.1103/PhysRevB.93.035414. 
[KTJ12]  Kristen Kaasbjerg, Kristian S Thygesen, and Karsten W Jacobsen. Unraveling the acoustic electronphonon interaction in graphene. Physical Review B, 85(16):165440, 2012. 