Quasi-particles of a 2D system: Difference between revisions

From The Yambo Project
Jump to navigation Jump to search
 
(51 intermediate revisions by the same user not shown)
Line 6: Line 6:


In the end, you will obtain a quasiparticle band structure based on the simulations, the first step towards the reproduction of an ARPES spectrum.  
In the end, you will obtain a quasiparticle band structure based on the simulations, the first step towards the reproduction of an ARPES spectrum.  
Beware: we won’t use fully converged parameters, so the final result should not be considered very accurate.
Beware: we won’t use fully converged parameters, so the final result should not be considered accurate.
 
==Step 1: Initialization==
 
The first step would consist as usual in performing a non-self-consistent DFT calculation of MoS<sub>2</sub>  (using for example Quantum ESPRESSO). Here, you can directly download the QE files from [https://media.yambo-code.eu/educational/tutorials/files/MoS2_HPC_tutorial.tar.gz].
Once completed the extraction, move inside the <code> 00_QE-DFT</code> folder and then convert QE data into a convenient format for Yambo. Just as a reminder, to convert, run the <code> p2y</code> executable to generate the uninitialised <code> SAVE</code>.
 
cd 00_QE-DFT/mos2.save
p2y


==Step 1:  Speeding up the self-energy convergence in the k-grid ==


In this section, you will learn to use two algorithms present in Yambo that lead to a speed up of the self-energy convergence with respect to the k-grid.


Now, you need to run the initialization step. Every Yambo run **must** start with this step. Just type
To appreciate the impact of these algorithms, let us first perform a GW computation for the monolayer of MoS<sub>2</sub>.


yambo
Enter the <code> 01_GW_first_run</code> folder and generate the input file:


  cd ../02_GW_convergence
  yambo -F gw_ppa.in -p p


==Step 2:  Speeding up the self-energy convergence ==
Modify the input file as follows:
 
In this section, you will learn to use two algorithms present in Yambo that lead to a speed up of the self-energy convergence with respect to the k-grid.
 
To appreciate the impact of these algorithms, let us first perform a GW computation for the monolayer of MoS<sub>2</sub>. Here, consider the number of G vectors and the number of bands as being already converged.  ( To work in a 2D system in Yambo, you should normally make use of Cutoff potential ?)


  gw0                              # [R] GW approximation
  gw0                              # [R] GW approximation
Line 38: Line 28:
  DIP_Threads=0                    # [OPENMP/X] Number of threads for dipoles
  DIP_Threads=0                    # [OPENMP/X] Number of threads for dipoles
  SE_Threads=0                    # [OPENMP/GW] Number of threads for self-energy
  SE_Threads=0                    # [OPENMP/GW] Number of threads for self-energy
rim_cut                          # [R] Coulomb potential
CUTGeo= "slab z"                  # [CUT] Coulomb Cutoff geometry: box/cylinder/sphere/ws/slab X/Y/Z/XY..
% CUTBox
  0.000000 | 0.000000 | 0.000000 |        # [CUT] [au] Box sides
%
CUTRadius= 0.000000              # [CUT] [au] Sphere/Cylinder radius
CUTCylLen= 0.000000              # [CUT] [au] Cylinder length
CUTwsGvec= 0.700000              # [CUT] WS cutoff: number of G to be modified
  EXXRLvcs=  37965          RL    # [XX] Exchange    RL components
  EXXRLvcs=  37965          RL    # [XX] Exchange    RL components
  VXCRLvcs=  37965          RL    # [XC] XCpotential RL components
  VXCRLvcs=  37965          RL    # [XC] XCpotential RL components
Line 67: Line 49:
  %
  %


You should obtain a GW gap of
Here, consider the number of G vectors and the number of bands as already being converged. So, compute the gap at GW level:
 
yambo -F gw_ppa.in -J 80b_10Ry
 
Once terminated the computation, you can now inspect the output file <code> o-80b_10Ry.qp</code>.
 
You should have obtained a GW gap of 2.483 eV.


==RIM==
==RIM==
[[File:Circle box.gif|thumb]]
[[File:Circle box.gif|thumb]]


To understand how a speed up may be achieved, consider once again the expression of the exchange part of the self-energy
To understand how we can improve the calculation in the k-grid to achieve a speed-up, consider briefly again the expression of the exchange part of the self-energy


[[File:Sx.png|none|x50px|caption]]
[[File:Sx.png|none|x50px|caption]]


The integration around q=0 here is problematic for the presence of a 1/q from the Coulomb potential that diverges, while all the other terms are slowly varying in q. Usually in Yambo this integration is performed analytically in a small sphere around q=0.
You can notice that the integration around q=0 here is problematic for the presence of a 1/q from the Coulomb potential that diverges, while all the other terms are slowly varying in q. Usually in Yambo this integration is performed analytically in a small sphere around q=0.


However, in this way the code lost part of the integral out of the circle. This usually  
However, in this way the code lost part of the integral out of the circle. This usually  
is not problematic because for a large number of q and k point the missing term goes to zero.
is not problematic because for a large number of q and k point the missing term goes to zero.
However in system that requires few k-points or even only the gamma one, it is possible
However in system that requires few k-points or even only the gamma one, it is possible
to perform a better integration of this term by adding the flag <code>-r</code> to generate the input:
to perform a better integration of this term by adding the flag <code>-r</code>. So as
 
yambo -F gw_ppa.in -p p -r
 
In this manner, you generate the input:


  HF_and_locXC                # [R XX] Hartree-Fock Self-energy and Vxc
  HF_and_locXC                # [R XX] Hartree-Fock Self-energy and Vxc
Line 133: Line 125:
  XTermKind= "none"                # [X] X terminator ("none","BG" Bruneval-Gonze)
  XTermKind= "none"                # [X] X terminator ("none","BG" Bruneval-Gonze)
  % GbndRnge
  % GbndRnge
     1 | 20 |                        # [GW] G[W] bands range
     1 | 80 |                        # [GW] G[W] bands range
  %
  %
  GTermKind= "none"                # [GW] GW terminator ("none","BG" Bruneval-Gonze,"BRS" Berger-Reining-Sottile)
  GTermKind= "BG"                # [GW] GW terminator ("none","BG" Bruneval-Gonze,"BRS" Berger-Reining-Sottile)
  DysSolver= "n"                  # [GW] Dyson Equation solver ("n","s","g")
  DysSolver= "n"                  # [GW] Dyson Equation solver ("n","s","g")
  %QPkrange                        # [GW] QP generalized Kpoint/Band indices
  %QPkrange                        # [GW] QP generalized Kpoint/Band indices
  7|7|13|14|
  7|7|13|14|
  %
  %
yambo -F gw_ppa.in -J 80b_10Ry_rim
You can now inspect the output file <code>o-80b_10Ry_rim.qp</code>. The GW gap should have increased to 4.109 eV.


==RIM-W==
==RIM-W==
In 2D systems, the improvement granted by the better integration around q=0 of the Coulomb potential via the RIM approach does not work well for the correlation part of self-energy
[[File:Sigma_c.png|none|x50px|caption]]


[[File:Rimw.png|thumb| Convergence of the quasiparticle band gap of MoS2
[[File:Rimw.png|thumb| Convergence of the quasiparticle band gap of MoS2
Line 148: Line 148:
an horizontal dashed line. Orange lines: RIM-W method.  The grey shaded regions show
an horizontal dashed line. Orange lines: RIM-W method.  The grey shaded regions show
the converge tolerance (±50 meV) centered at the centered at the extrapolated values.]]
the converge tolerance (±50 meV) centered at the centered at the extrapolated values.]]
In 2D systems, the improvement granted by the better integration around q=0 of the Coulomb potential via the RIM approach does not work well for the correlation part of self-energy
[[File:Sigma_c.png|none|x50px|caption]]


as the dielectric function <math>\epsilon_{00}</math> goes as q around q=0, matching the 1/q behavior of the Coulomb potential.  
as the dielectric function <math>\epsilon_{00}</math> goes as q around q=0, matching the 1/q behavior of the Coulomb potential.  


To solve this issue, it has been developed a method specific for 2D systems that performs the integration of <math>\epsilon_{0 0}(\mathbf{q}) \ v(\mathbf{q})</math> rather than only of <math>v(\mathbf{q})</math> when computing the correlation part of self-energy.


Details of this method can be found in the paper [https://www.nature.com/articles/s41524-023-00989-7| Efficient GW calculations in two-dimensional materials through a stochastic integration of the screened potential]. You can activate the RIM-W algorithm adding a further flag to your input:


To solve this issue, it has been developed a method specific for 2D systems that performs the integration of <math>\epsilon_{0 0}(\mathbf{q}) \ v(\mathbf{q})</math> rather than only of <math>v(\mathbf{q})</math> when computing the correlation part of self-energy.
yambo -F gw_ppa.in -p p -r -rw
In your input three further lines should have appeared:


Details of this method can be found in the paper [https://www.nature.com/articles/s41524-023-00989-7| Efficient GW calculations in two-dimensional materials through a stochastic integration of the screened potential]. You can activate the RIM-W algorithm adding two variables to your input:
  ...
  ...
  rim_cut                          # [R] Coulomb potential
  rim_cut                          # [R] Coulomb potential
Line 167: Line 165:
  <span style="color:red  >RIM_W</span>
  <span style="color:red  >RIM_W</span>
  <span style="color:red  >RandGvecW= 15              RL</span>
  <span style="color:red  >RandGvecW= 15              RL</span>
<span style="color:red  >rimw_type="semiconductor"</span>
  ...
  ...


The variable <code>RandGvecW</code> defines the number of G vectors to integrate W numerically. <code>RandGvecW</code> '''must''' be always smaller than the size of the <math>\chi</math> matrix.
The variable <code>RandGvecW</code> defines the number of G vectors to integrate W numerically. <code>RandGvecW</code> '''must''' be always smaller than the size of the <math>\chi</math> matrix.


Now repeat the computation of before, using the RIM-W algorithm. How much do you get for the band gap ?


==Step 3: GW parallel strategies==
Now repeat the computation again using the RIM-W algorithm.


===MPI parallelization===
yambo -F gw_ppa.in -J 80b_10Ry_rimw


For this section, let us enter the <code> 03_GW_parallel</code> directory. If you were in the <code> 02_GW_convegence</code> folder just type
How much has the band gap changed ?


cd ../03_GW_parallel
==Step 2: GW parallel strategies==


and inspect the input </code> gw.in</code>. You will see that we set low values for most of the convergence parameters except bands:
For this part of the tutorial, we will see how to use a parallel strategy on Yambo. As a test calculation, we compute the full band structure on a larger number of bands with respect the previous calculation.


vim gw.in
Differently to the approach used up to now, we will not work interactively but rather we will submit a job script as is commonly done when using Yambo on clusters. So, now exit the interactive mode and in the login node access the folder:


  FFTGvecs= 20      Ry    # [FFT] Plane-waves
  cd $CINECA_SCRATCH/YAMBO_TUTORIALS/MoS2_2Dquasiparticle_tutorial/02_GW_parallel
EXXRLvcs= 2        Ry    # [XX] Exchange RL components
VXCRLvcs= 2        Ry    # [XC] XCpotential RL components
% BndsRnXp
    1 |  300 |              # [Xp] Polarization function bands
%
NGsBlkXp= 1            Ry    # [Xp] Response block size
% GbndRnge
    1 |  300 |              # [GW] G[W] bands range
%
%QPkrange                    # [GW] QP generalized Kpoint/Band indices
  1| 19| 23| 30|
%


Note that we also added the quantity <code> FFTGvecs</code> to reduce the size of the Fourier transforms (the default corresponds to Quantum ESPRESSO <code> ecutwfc</code>, being 60 Ry in this case).
Here, inspect the input <code>gw.in</code>. You should see we compute a much larger number of quasi-particles:


In addition, we have deleted all the parallel parameters since we will be setting them via the submission script.
...
 
%QPkrange                       # [GW] QP generalized Kpoint/Band indices
Actually we are now dealing with a heavier system than before: as you can see from the <code> QPkrange</code> values, we have switched to a 12x12x1 '''k'''-point grid - having 19 points in the irreducible Brillouin zone - and turned spin-orbit coupling on in the DFT calculation - now the top valence band is number 26 instead of 13 because the bands are spin-polarized.
1|7|13|17|
 
%
The inspected file should look like [this one ????](https://hackmd.io/@palful/HkU0xblHj#GW-input-file-for-parallel-section).
...


For this part of the tutorial, we will be using the <code> slurm</code> submission script [`job_parallel.sh`](https://hackmd.io/@palful/HkU0xblHj#Slurm-submission-script-on-VEGA), which is available in the calculation directory.  
For this part of the tutorial, we will be using the slurm submission script job_parallel.sh, which is available in the calculation directory.
If you inspect it, you will see that the script adds additional variables to the yambo input file.
If you inspect it, you will see that the script adds additional variables to the yambo input file.
These variables control the parallel execution of the code:
These variables control the parallel execution of the code:
Line 222: Line 208:
  SE_Threads=  0            # [OPENMP/GW] Number of threads for self-energy
  SE_Threads=  0            # [OPENMP/GW] Number of threads for self-energy


The keyword <code> DIP</code> refers to the calculations of the screening matrix elements (also called "dipoles") needed for the screening function, <code> X</code> is the screening function (it stands for <math>\chi</math> since it is a response function), <code> SE</code> the self-energy.
The keyword <code>DIP</code> refers to the calculations of the screening matrix elements (also called “dipoles”) needed for the screening function, <code>X</code> is the screening function (it stands for χ since it is a response function), <code>SE</code> the self-energy.
These three sections of the code can be parallelised independently.
These three sections of the code can be parallelised independently.


:::info
The [PARALLEL] variables refer to MPI tasks, instead the threads are used for [OPENMP] parallelisation.
- In this subsection we are mainly concerned with the **[PARALLEL]** variables which refer to MPI _tasks_.
- We will see later an example of **[OPENMP]** parallelisation (i.e., adding _threads_).
:::


 
We start by calculating the QP corrections using only the MPI tasks and a single openMP thread. Therefore, create the submission script as:
We start by calculating the QP corrections using 16 MPI tasks and a single openMP thread. Therefore, edit the submission script as:


  #!/bin/bash
  #!/bin/bash
  #SBATCH --nodes=1
  #SBATCH --nodes=1
  #SBATCH --ntasks-per-node=16
  #SBATCH --ntasks-per-node=2
  #SBATCH --cpus-per-task=1
  #SBATCH --cpus-per-task=1
 
#SBATCH --partition=m100_sys_test
#SBATCH --qos=qos_test
#SBATCH --reservation=s_tra_yambo
#SBATCH --time=0:30:00
#SBATCH --account=tra23_Yambo
#SBATCH --mem=230000MB
#SBATCH --job-name=rutile
#SBATCH --error=err.job-%j
#SBATCH --output=out.job-%j
# load yambo and dependencies
module purge
module load hpc-sdk/2022--binary
module load spectrum_mpi/10.4.0--binary
export PATH=/m100_work/tra23_Yambo/softwares/YAMBO/5.2-cpu/bin:$PATH
export OMP_NUM_THREADS=${SLURM_CPUS_PER_TASK}
# info
ncpu=${SLURM_NTASKS}
nthreads=${OMP_NUM_THREADS}
label=MPI${ncpu}_OMP${nthreads}
jdir=run_${label}
cdir=run_${label}.out
# Update input file
filein0=gw.in        # Original file
filein=gw_${label}.in # New file
cp -f $filein0 $filein
cat >> $filein << EOF
DIP_CPU= "1 $ncpu 1"          # [PARALLEL] CPUs for each role
DIP_ROLEs= "k c v"            # [PARALLEL] CPUs roles (k,c,v)
DIP_Threads=  0              # [OPENMP/X] Number of threads for dipoles
X_and_IO_CPU= "1 1 1 $ncpu 1" # [PARALLEL] CPUs for each role
X_and_IO_ROLEs= "q g k c v"  # [PARALLEL] CPUs roles (q,g,k,c,v)
X_and_IO_nCPU_LinAlg_INV=1    # [PARALLEL] CPUs for Linear Algebra (if -1 it is automatically set)
X_Threads=  0                # [OPENMP/X] Number of threads for response functions
SE_CPU= "1 $ncpu 1"          # [PARALLEL] CPUs for each role
SE_ROLEs= "q qp b"            # [PARALLEL] CPUs roles (q,qp,b)
SE_Threads=  0                # [OPENMP/GW] Number of threads for self-energy
EOF
# run yambo
mpirun -n $ncpu yambo -F $filein -J ${jdir} -C $cdir
and submit the job
and submit the job
  sbatch job_parallel.sh
  sbatch job_parallel.sh


This will create a new input file and run it. The calculation databases and the human-readable files will be put in separate directories. Check the location of the report <code>r-*</code> file and the log <code>l-*</code> files, and inspect them while the calculation runs (it should take a couple of minutes).


This will create a new input file and run it. The calculation databases and the human-readable files will be put in separate directories. Check the location of the report <code>r-*</code> file and the log <code>l-*</code> files, and inspect them while the calculation runs.
For simplicity you could just type
For simplicity you could just type


  tail -f run_MPI16_OMP1.out/LOG/l-*_CPU_1
  tail -f run_MPI8_OMP1.out/LOG/l-*_CPU_1


to monitor the progress in the master thread (<code>Ctrl+C</code> to exit).
to monitor the progress in the master thread (<code>Ctrl+C</code> to exit). As you can see, the run takes some time, even though we are using minimal parameters.
As you can see, the run takes some time, even though we are using minimal parameters.


Meanwhile, we can run other jobs increasing the parallelisation. Let's employ 128 CPUs (i.e., half a CPU node on Vega). We'll use 128 MPI tasks, still with a single openMP thread. To this end modify the <code>job_parallel.sh</code> script changing
On a cluster like m100, to activate the full potential of the machine, it is useful to activate OpenMP threads by modifying cpus-per-task in the submission file. The product of the number of OpenMP and MPI tasks is equal to the total number of CPUs.  


#!/bin/bash
Therefore, we can distribute 8 cpus with 4 MPI tasks and then use 2 OpenMP threads:
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=128
#SBATCH --cpus-per-task=1
 
This time the code should be much faster. Once the run is over, try to run the simulation also on 32 and 64 MPI tasks. Each time, change <code>ntasks-per-node</code> to this number. Finally, you can try to produce a scaling plot.
 
The timings are contained in the <code>r-*</code> report files. You can already have a look at them typing
 
grep Time-Profile run_MPI*/r-*
 
 
The python script [<code>parse_ytiming.py</code>](https://hackmd.io/@palful/HkU0xblHj#parse_ytiming.py) is useful for the post-processing of report files. You can already find it in the directory, together with the reports for the longer calculations with 1, 2, 4 and 8 MPI tasks which have been provided.
 
If you didn't do so already, load the python module ????
```
module purge
module load matplotlib/3.2.1-foss-2020a-Python-3.8.2
```
 
Then, after your jobs have finished, run the script as
```
python parse_ytiming.py run_MPI
```
to look for a report file in each <code>run_MPI*.out</code> folder. Make sure you have only one report file per folder.
You can also play with the script to make it print detailed timing information, however you should already see that it produced a png plot showing times-to-completion on y axis against number of MPI tasks on the x axis.
 
[[File:Time parall.png|center]]
 
 
What can we learn from this plot? In particular, try to answer the following questions:
* Up to which number of MPI tasks our system efficiently scales? At which point adding more threads starts to become a waste of resources?
* How could we change our parallelisation strategy to improve the scaling even more? 
 
:::info
Keep in mind that the MPI scaling we are seeing here is not the true yambo scaling, but depends on the small size of our tutorial system. In a realistic calculation for a large-sized system, '''yambo has been shown to scale well up to tens of thousands of MPI tasks!'''
:::
 
===Hybrid parallelization strategy===
 
It turns out that for this system it's not a good idea to use more than 32 MPI tasks. But we still have 128 CPUs available for this run. How to optimize the calaculation even more? We can try using OpenMP threads.
 
This is turned on by modifying <code>cpus-per-task</code> in the submission file. The product of the number of OpenMP threads and MPI tasks is equal to the total number of CPUs.
Therefore, we can distribute 128 cpus with 32 MPI tasks (efficient MPI scaling, see above) and then use 4 OpenMP threads:  


  #!/bin/bash
  #!/bin/bash
  #SBATCH --nodes=1
  #SBATCH --nodes=1
  #SBATCH --ntasks-per-node=32
  #SBATCH --ntasks-per-node=4
  #SBATCH --cpus-per-task=4
  #SBATCH --cpus-per-task=2


 
Actually, we don’t need to change the related openMP variables for the yambo input, since the value 0 means “use the value of OMP_NUM_THREADS” and we have now set this environment variable to 2 via our script.
Actually, we don't need to change the related openMP variables for the yambo input, since the value <code>0</code> means "use the value of <code>OMP_NUM_THREADS</code>" and we have now set this environment variable to <code>4</code> via our script.
Otherwise, any positive number can directly specify the number of threads to be used in each section of the code.
Otherwise, any positive number can directly specify the number of threads to be used in each section of the code.


Line 313: Line 296:
  SE_Threads=  0      # [OPENMP/GW] Number of threads for self-energy
  SE_Threads=  0      # [OPENMP/GW] Number of threads for self-energy


You can try to run this calculation and check if it is faster than the pure 128 MPI one from before.  
You can try to run this calculation and check if it is faster than the pure MPI one from before. In general, you expect to a massive gain with OpenMP if you are already close to an efficient MPI scaling
You should find a massive gain: in our case, the pure <code>MPI128_OMP1</code> took 01m-02s (to be changed), while the hybrid <code>MPI32_OMP4</code> took (to be changed) just 37s, a 40% speedup.  
 
You can also try any other thread combinations including the pure OpenMP scaling, and compare the timings.
 
In real-life calculations running on a large number of cores, it may be a good idea to adopt a hybrid approach. The most efficient scaling can depend both on your system and on the HPC facility you’re running on.
 
In general, OpenMP can help lower memory requirements within a node. You can try to increase the OpenMP share of threads if getting Out Of Memory errors.


You can also try any other thread combinations including the pure openMP scaling, and compare the timings.
==Step 3: Running on GPU==


:::info
For this part of the tutorial, we will repeat the calculation of before, making use of gpus.  
* In real-life calculations running on ''n''<sub>''cores''</sub> > 100, as we have seen, it may be a good idea to adopt a hybrid approach.
* The most efficient scaling can depend both on your system and on the HPC facility you're running on. For a full CPU node on Vega (256 cores), using a large-scale system, we have found that 16 tasks times 16 threads gives the best performance.
* OpenMP can help lower memory requirements within a node. You can try to increase the OpenMP share of threads if getting Out Of Memory errors.
:::


As mentioned above, we can conclude that the "best" parallelisation strategy is in general both system-dependent and cluster-dependent. If working on massive systems, it may be well worth your while to do preliminary parallel tests in order to work out the most efficient way to run these jobs.
So now move to:


===[OPTIONAL] Comparing different parallelisation schemes ===
02_GW_gpu


Up to now we always parallelised over a single parameter, i.e. <code>c</code> or <code>qp</code>. However, Yambo allows for tuning the parallelisation scheme over several parameters broadly corresponding to "loops" (i.e., summations or discrete integrations) in the code.
Here we are using yet another capability of yambo: running on GPUs instead of CPUs. This usually leads to extreme speedups in the calculations. Fortunately, the m100 cluster also has some GPU nodes! So now let us have a look at the submission script:


To this end you can open again the <code>run_parallel.sh</code> script and modify the section where the yambo input variables are set.
vim gpu_job.sh
 
#!/bin/bash
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=2
#SBATCH --cpus-per-task=1
#SBATCH --gres=gpu:4


For example, <code>X_CPU</code> sets how the MPI Tasks are distributed in the calculation of the response function. The possibilities are shown in the <code>X_ROLEs</code>. The same holds for <code>SE_CPU</code> and <code>SE_ROLEs</code> which control how MPI Tasks are distributed in the calculation of the self-energy.
each GPU node contains four GPUs (<code>--gres=gpu:4</code>). In yambo, each GPU corresponds to a single MPI task, therefore we have still the total number of tasks equal to 8. OpenMP threading is allowed - but not too much, otherwise we lose efficiency. Here in fact, we are not using OpenMP so </code>export OMP_NUM_THREADS=1</code>.  


X_CPU= "1 1 1 $ncpu 1"      # [PARALLEL] CPUs for each role
In addition, you will see that in order to run on GPUs we are now using a different executable than before, obtained with a GPU-specific compilation of the code.
X_ROLEs= "q g k c v"        # [PARALLEL] CPUs roles (q,g,k,c,v)
X_nCPU_LinAlg_INV= $ncpu    # [PARALLEL] CPUs for Linear Algebra
X_Threads=  0              # [OPENMP/X] Number of threads for response functions
DIP_Threads=  0            # [OPENMP/X] Number of threads for dipoles
SE_CPU= "1 $ncpu 1"        # [PARALLEL] CPUs for each role
SE_ROLEs= "q qp b"          # [PARALLEL] CPUs roles (q,qp,b)
SE_Threads=  0              # [OPENMP/GW] Number of threads for self-energy


You can try different parallelization schemes and check the performances of Yambo. In doing so, you should also change the jobname <code>label=MPI${ncpu}_OMP${nthreads}</code> in the <code>run_parallel.sh</code> script to avoid confusion with previous calculations.
module purge
module load hpc-sdk/2022--binary spectrum_mpi/10.4.0--binary
export PATH=/m100_work/tra23_Yambo/softwares/YAMBO/5.2-gpu/bin:$PATH


You may then check how speed, memory and load balance between the CPUs are affected. You could modify the script [`parse_ytiming.py`](https://hackmd.io/@palful/HkU0xblHj#parse_ytiming.py) to parse the new data, read and distinguish between more file names, new parallelisation options, etc.
In general, the gpu compilation might be different on your machine, or the executable may be the same with no need to load an additional modules. In conclusion, we are running over 4 GPU cards, distributed with 2 MPI tasks, and using 1 thread.


:::info
The calculation should faster, about 2 minutes instead of 2 min and 30s with a purely CPU calculation. The gain can become even greater in larger systems. You can have a look at the results collected in folder <code>MPI2_OMP1</code>. The quasiparticle corrections are stored in human-readable form in the file <code>MPI2_OMP1.out/o-GW_bnds.qp</code>, and in netCDF format in the quasiparticle database <code>MPI2_OMP1/ndb.QP</code>.
*The product of the numbers entering each variable (i.e. <code>X_CPU</code> and <code>SE_CPU</code>) times the number of threads should always match the total number of cores (unless you want to overload the cores taking advantage of multi-threads).
*Using the <code>X_Threads</code> and <code>SE_Threads</code> variables you can think about setting different hybrid schemes between the screening and the self-energy runlevels.
*Memory scales better if you parallelize on bands (<code>c v b</code>).
*Parallelization on '''k'''-points performs similarly to parallelization on bands, but requires more memory.
*Parallelization on '''q'''-points requires much less communication between the MPI tasks. It may be useful if you run on more than one node and the inter-node connection is slow.
:::

Latest revision as of 15:47, 23 May 2023

MoS2 monolayer (top and side views). Gray: Mo atoms, yellow: S atoms.

In this tutorial you will compute the quasiparticle corrections to the band structure of a free-standing single layer of MoS2. Aim of the tutorial is to learn how to efficiently run a GW simulation in a 2D material based on:

  • Acceleration techniques of GW, some of which specific for 2D systems
  • Parallelization techniques

In the end, you will obtain a quasiparticle band structure based on the simulations, the first step towards the reproduction of an ARPES spectrum. Beware: we won’t use fully converged parameters, so the final result should not be considered accurate.

Step 1: Speeding up the self-energy convergence in the k-grid

In this section, you will learn to use two algorithms present in Yambo that lead to a speed up of the self-energy convergence with respect to the k-grid.

To appreciate the impact of these algorithms, let us first perform a GW computation for the monolayer of MoS2.

Enter the 01_GW_first_run folder and generate the input file:

yambo -F gw_ppa.in -p p

Modify the input file as follows:

gw0                              # [R] GW approximation
ppa                              # [R][Xp] Plasmon Pole Approximation for the Screened Interaction
dyson                            # [R] Dyson Equation solver
HF_and_locXC                     # [R] Hartree-Fock
em1d                             # [R][X] Dynamically Screened Interaction
X_Threads=0                      # [OPENMP/X] Number of threads for response functions
DIP_Threads=0                    # [OPENMP/X] Number of threads for dipoles
SE_Threads=0                     # [OPENMP/GW] Number of threads for self-energy
EXXRLvcs=  37965           RL    # [XX] Exchange    RL components
VXCRLvcs=  37965           RL    # [XC] XCpotential RL components
Chimod= "HARTREE"                # [X] IP/Hartree/ALDA/LRC/PF/BSfxc
% BndsRnXp
   1 | 80 |                       # [Xp] Polarization function bands
%
NGsBlkXp= 10                Ry    # [Xp] Response block size
% LongDrXp
1.000000 | 1.000000 | 1.000000 |        # [Xp] [cc] Electric Field
%
PPAPntXp= 27.21138         eV    # [Xp] PPA imaginary energy
XTermKind= "none"                # [X] X terminator ("none","BG" Bruneval-Gonze)
% GbndRnge
   1 | 80 |                         # [GW] G[W] bands range
%
GTermKind= "BG"                  # [GW] GW terminator ("none","BG" Bruneval-Gonze,"BRS" Berger-Reining-Sottile)
DysSolver= "n"                   # [GW] Dyson Equation solver ("n","s","g")
%QPkrange                        # [GW] QP generalized Kpoint/Band indices
7|7|13|14|
%

Here, consider the number of G vectors and the number of bands as already being converged. So, compute the gap at GW level:

yambo -F gw_ppa.in -J 80b_10Ry

Once terminated the computation, you can now inspect the output file o-80b_10Ry.qp.

You should have obtained a GW gap of 2.483 eV.

RIM

Circle box.gif

To understand how we can improve the calculation in the k-grid to achieve a speed-up, consider briefly again the expression of the exchange part of the self-energy

caption

You can notice that the integration around q=0 here is problematic for the presence of a 1/q from the Coulomb potential that diverges, while all the other terms are slowly varying in q. Usually in Yambo this integration is performed analytically in a small sphere around q=0.

However, in this way the code lost part of the integral out of the circle. This usually is not problematic because for a large number of q and k point the missing term goes to zero. However in system that requires few k-points or even only the gamma one, it is possible to perform a better integration of this term by adding the flag -r. So as

yambo -F gw_ppa.in -p p -r

In this manner, you generate the input:

HF_and_locXC                 # [R XX] Hartree-Fock Self-energy and Vxc
ppa                          # [R Xp] Plasmon Pole Approximation
gw0                          # [R GW] G0W0 Quasiparticle energy levels
em1d                         # [R Xd] Dynamical Inverse Dielectric Matrix
rim_cut                      # [R] Coulomb potential
RandQpts=  1000000           # [RIM] Number of random q-points in the BZ
RandGvec=  100           RL  # [RIM] Coulomb interaction RS components
CUTGeo= "slab z"             # [CUT] Coulomb Cutoff geometry: box/cylinder/sphere/ws/slab X/Y/Z/XY..
...


In this input, RandGvec is the number of component of the Coulomb potential we want integrate numerically and RandQpts is the number of random points used to perform the integral by Monte Carlo. The [CUT] keyword refers to the truncation of the Coulomb interaction to avoid spurious interaction between periodically repeated copies of the simulation supercell along the z-direction (that in this case is the non periodic direction). Keep in mind that the vacuum space between two copies of the system should be converged: here we are using 20 bohr but a value of 40 bohr would be more realistic.

If you turn on this integration you will get a slightly different band gap, but in the limit of large k points the final results will be the same of the standard method. However this correction is important for systems that converge with few k-points or with gamma only and it is applied both at the exchange and correlation part of the self-energy.

Now make a GW computation using the RIM method

gw0                              # [R] GW approximation
ppa                              # [R][Xp] Plasmon Pole Approximation for the Screened Interaction
dyson                            # [R] Dyson Equation solver
HF_and_locXC                     # [R] Hartree-Fock
em1d                             # [R][X] Dynamically Screened Interaction
X_Threads=0                      # [OPENMP/X] Number of threads for response functions
DIP_Threads=0                    # [OPENMP/X] Number of threads for dipoles
SE_Threads=0                     # [OPENMP/GW] Number of threads for self-energy
rim_cut                          # [R] Coulomb potential
RandQpts=1000000                 # [RIM] Number of random q-points in the BZ
RandGvec= 100              RL    # [RIM] Coulomb interaction RS components
CUTGeo= "slab z"                   # [CUT] Coulomb Cutoff geometry: box/cylinder/sphere/ws/slab X/Y/Z/XY..
% CUTBox
 0.000000 | 0.000000 | 0.000000 |        # [CUT] [au] Box sides
%
CUTRadius= 0.000000              # [CUT] [au] Sphere/Cylinder radius
CUTCylLen= 0.000000              # [CUT] [au] Cylinder length
CUTwsGvec= 0.700000              # [CUT] WS cutoff: number of G to be modified
EXXRLvcs=  37965           RL    # [XX] Exchange    RL components
VXCRLvcs=  37965           RL    # [XC] XCpotential RL components
Chimod= "HARTREE"                # [X] IP/Hartree/ALDA/LRC/PF/BSfxc
% BndsRnXp
   1 | 80 |                       # [Xp] Polarization function bands
%
NGsBlkXp= 10                Ry    # [Xp] Response block size
% LongDrXp
1.000000 | 1.000000 | 1.000000 |        # [Xp] [cc] Electric Field
%
PPAPntXp= 27.21138         eV    # [Xp] PPA imaginary energy
XTermKind= "none"                # [X] X terminator ("none","BG" Bruneval-Gonze)
% GbndRnge
   1 | 80 |                         # [GW] G[W] bands range
%
GTermKind= "BG"                # [GW] GW terminator ("none","BG" Bruneval-Gonze,"BRS" Berger-Reining-Sottile)
DysSolver= "n"                   # [GW] Dyson Equation solver ("n","s","g")
%QPkrange                        # [GW] QP generalized Kpoint/Band indices
7|7|13|14|
%
yambo -F gw_ppa.in -J 80b_10Ry_rim

You can now inspect the output file o-80b_10Ry_rim.qp. The GW gap should have increased to 4.109 eV.

RIM-W

In 2D systems, the improvement granted by the better integration around q=0 of the Coulomb potential via the RIM approach does not work well for the correlation part of self-energy

caption
Convergence of the quasiparticle band gap of MoS2 with respect to the number of sampling points of the BZ. Blue lines: standard integration methods. The extrapolated values are indicated with an horizontal dashed line. Orange lines: RIM-W method. The grey shaded regions show the converge tolerance (±50 meV) centered at the centered at the extrapolated values.

as the dielectric function [math]\displaystyle{ \epsilon_{00} }[/math] goes as q around q=0, matching the 1/q behavior of the Coulomb potential.

To solve this issue, it has been developed a method specific for 2D systems that performs the integration of [math]\displaystyle{ \epsilon_{0 0}(\mathbf{q}) \ v(\mathbf{q}) }[/math] rather than only of [math]\displaystyle{ v(\mathbf{q}) }[/math] when computing the correlation part of self-energy.

Details of this method can be found in the paper Efficient GW calculations in two-dimensional materials through a stochastic integration of the screened potential. You can activate the RIM-W algorithm adding a further flag to your input:

yambo -F gw_ppa.in -p p -r -rw

In your input three further lines should have appeared:

...
rim_cut                           # [R] Coulomb potential
RandQpts=1000000                  # [RIM] Number of random q-points in the BZ
RandGvec= 100              RL     # [RIM] Coulomb interaction RS components
RIM_W
RandGvecW= 15              RL
rimw_type="semiconductor"
...

The variable RandGvecW defines the number of G vectors to integrate W numerically. RandGvecW must be always smaller than the size of the [math]\displaystyle{ \chi }[/math] matrix.


Now repeat the computation again using the RIM-W algorithm.

yambo -F gw_ppa.in -J 80b_10Ry_rimw

How much has the band gap changed ?

Step 2: GW parallel strategies

For this part of the tutorial, we will see how to use a parallel strategy on Yambo. As a test calculation, we compute the full band structure on a larger number of bands with respect the previous calculation.

Differently to the approach used up to now, we will not work interactively but rather we will submit a job script as is commonly done when using Yambo on clusters. So, now exit the interactive mode and in the login node access the folder:

cd $CINECA_SCRATCH/YAMBO_TUTORIALS/MoS2_2Dquasiparticle_tutorial/02_GW_parallel

Here, inspect the input gw.in. You should see we compute a much larger number of quasi-particles:

...
%QPkrange                        # [GW] QP generalized Kpoint/Band indices
1|7|13|17|
%
...

For this part of the tutorial, we will be using the slurm submission script job_parallel.sh, which is available in the calculation directory. If you inspect it, you will see that the script adds additional variables to the yambo input file. These variables control the parallel execution of the code:

DIP_CPU= "1 $ncpu 1"       # [PARALLEL] CPUs for each role
DIP_ROLEs= "k c v"         # [PARALLEL] CPUs roles (k,c,v)
DIP_Threads=  0            # [OPENMP/X] Number of threads for dipoles
X_and_IO_CPU= "1 1 1 $ncpu 1"     # [PARALLEL] CPUs for each role
X_and_IO_ROLEs= "q g k c v"       # [PARALLEL] CPUs roles (q,g,k,c,v)
X_and_IO_nCPU_LinAlg_INV= 1   # [PARALLEL] CPUs for Linear Algebra
X_Threads=  0              # [OPENMP/X] Number of threads for response functions
SE_CPU= " 1 $ncpu 1"       # [PARALLEL] CPUs for each role
SE_ROLEs= "q qp b"         # [PARALLEL] CPUs roles (q,qp,b)
SE_Threads=  0             # [OPENMP/GW] Number of threads for self-energy

The keyword DIP refers to the calculations of the screening matrix elements (also called “dipoles”) needed for the screening function, X is the screening function (it stands for χ since it is a response function), SE the self-energy. These three sections of the code can be parallelised independently.

The [PARALLEL] variables refer to MPI tasks, instead the threads are used for [OPENMP] parallelisation.

We start by calculating the QP corrections using only the MPI tasks and a single openMP thread. Therefore, create the submission script as:

#!/bin/bash
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=2
#SBATCH --cpus-per-task=1
#SBATCH --partition=m100_sys_test
#SBATCH --qos=qos_test
#SBATCH --reservation=s_tra_yambo
#SBATCH --time=0:30:00
#SBATCH --account=tra23_Yambo
#SBATCH --mem=230000MB
#SBATCH --job-name=rutile
#SBATCH --error=err.job-%j
#SBATCH --output=out.job-%j

# load yambo and dependencies
module purge
module load hpc-sdk/2022--binary
module load spectrum_mpi/10.4.0--binary
export PATH=/m100_work/tra23_Yambo/softwares/YAMBO/5.2-cpu/bin:$PATH

export OMP_NUM_THREADS=${SLURM_CPUS_PER_TASK}

# info
ncpu=${SLURM_NTASKS}
nthreads=${OMP_NUM_THREADS}

label=MPI${ncpu}_OMP${nthreads}
jdir=run_${label}
cdir=run_${label}.out

# Update input file
filein0=gw.in         # Original file
filein=gw_${label}.in # New file

cp -f $filein0 $filein
cat >> $filein << EOF

DIP_CPU= "1 $ncpu 1"          # [PARALLEL] CPUs for each role
DIP_ROLEs= "k c v"            # [PARALLEL] CPUs roles (k,c,v)
DIP_Threads=  0               # [OPENMP/X] Number of threads for dipoles
X_and_IO_CPU= "1 1 1 $ncpu 1" # [PARALLEL] CPUs for each role
X_and_IO_ROLEs= "q g k c v"   # [PARALLEL] CPUs roles (q,g,k,c,v)
X_and_IO_nCPU_LinAlg_INV=1    # [PARALLEL] CPUs for Linear Algebra (if -1 it is automatically set)
X_Threads=  0                 # [OPENMP/X] Number of threads for response functions
SE_CPU= "1 $ncpu 1"           # [PARALLEL] CPUs for each role
SE_ROLEs= "q qp b"            # [PARALLEL] CPUs roles (q,qp,b)
SE_Threads=  0                # [OPENMP/GW] Number of threads for self-energy

EOF

# run yambo
mpirun -n $ncpu yambo -F $filein -J ${jdir} -C $cdir

and submit the job

sbatch job_parallel.sh

This will create a new input file and run it. The calculation databases and the human-readable files will be put in separate directories. Check the location of the report r-* file and the log l-* files, and inspect them while the calculation runs (it should take a couple of minutes).

For simplicity you could just type

tail -f run_MPI8_OMP1.out/LOG/l-*_CPU_1

to monitor the progress in the master thread (Ctrl+C to exit). As you can see, the run takes some time, even though we are using minimal parameters.

On a cluster like m100, to activate the full potential of the machine, it is useful to activate OpenMP threads by modifying cpus-per-task in the submission file. The product of the number of OpenMP and MPI tasks is equal to the total number of CPUs.

Therefore, we can distribute 8 cpus with 4 MPI tasks and then use 2 OpenMP threads:

#!/bin/bash
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=4
#SBATCH --cpus-per-task=2

Actually, we don’t need to change the related openMP variables for the yambo input, since the value 0 means “use the value of OMP_NUM_THREADS” and we have now set this environment variable to 2 via our script. Otherwise, any positive number can directly specify the number of threads to be used in each section of the code.

DIP_Threads=  0     # [OPENMP/X] Number of threads for dipoles
...
X_Threads=  0       # [OPENMP/X] Number of threads for response functions
...
SE_Threads=  0      # [OPENMP/GW] Number of threads for self-energy

You can try to run this calculation and check if it is faster than the pure MPI one from before. In general, you expect to a massive gain with OpenMP if you are already close to an efficient MPI scaling

You can also try any other thread combinations including the pure OpenMP scaling, and compare the timings.

In real-life calculations running on a large number of cores, it may be a good idea to adopt a hybrid approach. The most efficient scaling can depend both on your system and on the HPC facility you’re running on.

In general, OpenMP can help lower memory requirements within a node. You can try to increase the OpenMP share of threads if getting Out Of Memory errors.

Step 3: Running on GPU

For this part of the tutorial, we will repeat the calculation of before, making use of gpus.

So now move to:

02_GW_gpu

Here we are using yet another capability of yambo: running on GPUs instead of CPUs. This usually leads to extreme speedups in the calculations. Fortunately, the m100 cluster also has some GPU nodes! So now let us have a look at the submission script:

vim gpu_job.sh
#!/bin/bash
#SBATCH --nodes=1
#SBATCH --ntasks-per-node=2
#SBATCH --cpus-per-task=1
#SBATCH --gres=gpu:4

each GPU node contains four GPUs (--gres=gpu:4). In yambo, each GPU corresponds to a single MPI task, therefore we have still the total number of tasks equal to 8. OpenMP threading is allowed - but not too much, otherwise we lose efficiency. Here in fact, we are not using OpenMP so export OMP_NUM_THREADS=1.

In addition, you will see that in order to run on GPUs we are now using a different executable than before, obtained with a GPU-specific compilation of the code.

module purge
module load hpc-sdk/2022--binary spectrum_mpi/10.4.0--binary
export PATH=/m100_work/tra23_Yambo/softwares/YAMBO/5.2-gpu/bin:$PATH

In general, the gpu compilation might be different on your machine, or the executable may be the same with no need to load an additional modules. In conclusion, we are running over 4 GPU cards, distributed with 2 MPI tasks, and using 1 thread.

The calculation should faster, about 2 minutes instead of 2 min and 30s with a purely CPU calculation. The gain can become even greater in larger systems. You can have a look at the results collected in folder MPI2_OMP1. The quasiparticle corrections are stored in human-readable form in the file MPI2_OMP1.out/o-GW_bnds.qp, and in netCDF format in the quasiparticle database MPI2_OMP1/ndb.QP.