VASP: TiO₂ surface relaxation

This tutorial will teach you how to:

  • construct a slab for rutile TiO2(001)

  • set up a geometry optimization job with constraints

  • set up a DFT+U VASP job to be run via the AMS driver


Step 1: Check the VASP installation


VASP is not distributed together with the Amsterdam Modeling Suite, but needs to be obtained and installed separately.

Verify that you have access to a working installation of VASP 5, on either your local or a remote machine. If you want to run on a remote machine or computer cluster, check that you have set up a working AMSjobs queue for that system.

Check that you can run VASP, for example using one of the following commands:

mpirun -np 4 vasp #parallelize over 4 cores

If you do not know the proper command to launch VASP, ask your system administrator.

Step 2: Locate the POTCAR library

VASP requires the use of pseudopotentials or PAW potentials for each element. These are distributed with VASP in files called POTCAR or POTCAR.Z.

In this tutorial, you will use the Projector Augmented Wave (PAW) potentials for Ti and O constructed for the PBE density functional. You should have obtained those PAW potentials with VASP.

On your local machine (the machine running AMSinput), locate the needed POTCAR files. If you do not have them, you can download them from the VASP website.

For example, if the needed POTCAR or POTCAR.Z files are at


then the path /some/path/PAW_PBE/ would be the POTCAR Library that you need to specify in one of the following steps.

Step 3: Set up the system - a TiO2(001) slab

1. Start AMSinput
2. Switch to VASP: ADFPanel VASPPanel
3. In the menu bar, select Edit → Crystal → Tetragonal → Rutile
4. Click OK to get the experimental rutile TiO2 crystal structure unit cell (a = 4.59 Å, c = 2.96 Å).
5. Click Yes when asked whether to map the atoms to the unit cell.


In production calculations, one would normally perform a lattice optimization of the bulk structure before creating a slab. In this tutorial, we will just use the experimental lattice parameters.

Now create a 2-layer (001) slab (because the crystal unit cell contains 2 layers of atoms at different positions along c, the resulting slab will actually be 4 atomic layers thick).

6. Select Edit → Crystal → Generate slab
7. Enter 0 0 1 as the Miller indices.
8. Set the number of layers to 2.
9. Click “Generate slab”.

This should create a 2-layer slab in the 3D area on the left. Select View → View Direction → Along X from the menu bar, or rotate the system so that the z-axis (shown as a blue line) is roughly displayed vertically, giving you a side view of the slab. The slab is 2c = 5.92 Å thick.


  • It may happen that bonds are drawn between adjacent Ti ions. These bonds do not affect the VASP calculation and can safely be ignored. If you prefer, you can remove them by clicking on them and pressing Backspace.

  • In production calculations, one would normally use a much thicker slab.

All VASP calculations are performed under 3D periodic boundary conditions. Therefore, slabs are necessarily separated by a vacuum gap in the surface normal direction.

10. Click on the Periodic View Tool PeriodicViewTool to visualize a few periodically repeated images of the central cell.

Note that the vacuum gap is very large. Because VASP uses a planewave basis set, a larger vacuum gap will increase the computational cost. Therefore, it is a good idea to use a smaller vacuum gap, but the vacuum gap should not be so small that one side of the slab interacts with the other side through the vacuum gap. Here, we will set a rather small vacuum gap of about 8 Å.


In production calculations, the vacuum gap should be systematically varied for the particular system at hand, and be large enough such that any calculated quantity (surface energy, adsorption energy, work function, etc.) is converged.

11. In the panel bar, select Model → Lattice.
12. Set the components of the c lattice vector to 0.0 0.0 14.0. (Note: this c lattice vector now refers to the slab system including the vacuum gap, and no longer to the original unit cell of the crystal).
13. Click on one of the Ti atoms in the center of the slab to select it. Then, in the menu bar select Edit → Crystal → Set (0.5, 0.5, 0.5). This moves the slab to the center of the vacuum gap, which typically makes visualization easier.
14. Click again on the Periodic View Tool PeriodicViewTool so that only a single periodic image is displayed.
15. In the menu bar, select Edit → Crystal → Map atoms to (0..1). This places all of the atoms inside the displayed unit cell.

Step 4: Set the VASP settings

1. In the panel bar, select Main.
2. In Command to execute VASP enter the command to execute VASP (from Step 1).
3. In POTCAR library enter the path to the PBE PAW POTCAR library (from Step 2). This path will be remembered the next time you start AMSinput.


In the panel bar, select Details → Pseudopotentials to see and/or change exactly which POTCAR files that will be used in the calculation.

4. In the main panel, set the Type of k-point grid to Monkhorst-Pack (default).
5. Set the k-point grid dimensions to 3 3 1.
6. Set the XC Functional to PBE.
7. Set the Planewave energy cutoff to 400 eV (default).
8. Set the Type of smearing to Gaussian (default).
9. Set the Smearing width to 0.1 eV.


In production calculations, the k-point grid dimensions and planewave energy cutoff need to be converged by means of convergence tests.

10. In the panel bar, select Model → Hubbard U.
11. Check the Use DFT+U checkbox.
12. Set Hubbard U-type to 2 (default, the method by Dudarev et al.)
13. Set LMAXMIX to 4.
14. For Ti, set U to 3.0 eV, and l to d (default).
15. For O, set l to Disable.

The main VASP input file, INCAR, allows the user to set an enormous number of settings and have great control over the resulting calculation. AMSinput does not have dedicated panels for all of the settings. Instead, it is possible for you to set arbitrary settings by simply typing them in. In this example, you will change the self-consistent-field algorithm.

16. In the panel bar, select Details → Expert VASP.
17. Type ALGO = Fast in the Additional INCAR options text box.


On the Details → Expert VASP panel, consider checking the Only preprocess box. When you submit your calculation, VASP will then not be executed, but all of the input files that VASP would have seen will be left on disk for you to inspect. This way, you can double-check that the input to VASP is correct before actually running the calculation.

Step 5: Set the AMS settings

1. In the main panel, select Task → Geometry Optimization.
2. Click the MoreBtn button, or in the menu bar select Details → Geometry Optimization.
3. Inspect, but do not change, the convergence criteria.
4. Click the MoreBtn next to Constraints, or in the menu bar select Model → Geometry Constraints and PES Scan.
5. In the molecule drawing area, select the six atoms (2 Ti and 4 O) in the center of the slab, by first clicking on one of them, and subsequently clicking the others while holding Shift.
6. Click the AddButton button next to selected atoms (fix positions).
7. In the menu bar, select File → Save, and save your job with the name VASP_PBE_U_TiO2_slab.ams.


When running VASP via the AMS driver, the geometry optimization settings and constraints are set for the AMS driver and not for VASP. This means that VASP’s internal geometry convergence criteria, like the EDIFFG setting in INCAR, should not be set. You also do not need to enable VASP’s “Selective optimization” to keep some atoms fixed. Instead, all geometry optimization settings are handled by the AMS driver.

Step 6: Run your job

1. Open AMSjobs: SCM → Jobs.
2. Select your job in the list. In the menu bar, select Queue → Interactive if you want to run the job on your local machine, or select the queue for the remote machine that you want to run on.
3. In the menu bar, select Job → Run.

After the calculation has finished, visualize the geometry optimization in AMSmovie.

4. Select your job in AMSjobs, and open AMSmovie: SCM → Movie.
5. In the menu bar, select View → Molecule → Balls
6. Use the horizontal scrollbar at the bottom of the window to see the steps of the geometry optimization.

Did the top-layer Ti atoms relax “in” towards the bulk or “out” towards the vacuum?