Structure optimization - Revision history

Csheldon at 13:51, 24 March 2026

2026-03-24T13:51:34Z

← Older revision		Revision as of 13:51, 24 March 2026
Line 8:		Line 8:
	(ii) You may need to consider a diverse set of starting structures, if you are not sure about the most reasonable one.		(ii) You may need to consider a diverse set of starting structures, if you are not sure about the most reasonable one.

	There is a lecture on ~~[https~~:~~//youtu.be/gwA0KtAcmH4~~ structure optimization] available on our YouTube channel.		There is a lecture on {{Video\|structure_opt:sudarshan:2025\|structure optimization}} available on our YouTube channel.

	== Common considerations ==		== Common considerations ==

Singraber at 08:05, 24 October 2025

2025-10-24T08:05:58Z

Csheldon: /* Overview */

2025-09-22T10:24:55Z

Overview

← Older revision		Revision as of 10:24, 22 September 2025
Line 7:		Line 7:
	(i) You need to make sure that the starting structure is close enough to a minimum for the optimizers to work.		(i) You need to make sure that the starting structure is close enough to a minimum for the optimizers to work.
	(ii) You may need to consider a diverse set of starting structures, if you are not sure about the most reasonable one.		(ii) You may need to consider a diverse set of starting structures, if you are not sure about the most reasonable one.

			There is a lecture on [https://youtu.be/gwA0KtAcmH4 structure optimization] available on our YouTube channel.

	== Common considerations ==		== Common considerations ==

Wolloch: /* Common considerations */

2025-04-17T12:06:37Z

Common considerations

← Older revision		Revision as of 12:06, 17 April 2025
Line 39:		Line 39:
	The quantity <tt>trialstep</tt> is the size of the current trialstep.		The quantity <tt>trialstep</tt> is the size of the current trialstep.
	Multiply the current value of {{TAG\|POTIM}} with this number to obtain the optimal step size.		Multiply the current value of {{TAG\|POTIM}} with this number to obtain the optimal step size.
			{{NB\|mind\|If you use the default value of {{TAGO\|ISYM\|2}}, lower symmetry structures will be inaccessible to the relaxation algorithms. It is usually preferable to break symmetries intentionally by modifying the starting structure instead of turning symmetry operations off via {{TAGO\|ISYM\|0}}.}}

	== RMM-DIIS ==		== RMM-DIIS ==

Schlipf: /* RMM-DIIS */

2024-10-21T11:39:28Z

RMM-DIIS

← Older revision		Revision as of 11:39, 21 October 2024
Line 43:		Line 43:

	For {{TAGO\|IBRION\|1}}, VASP uses a RMM-DIIS algorithm to relax the ions into their instantaneous groundstate.		For {{TAGO\|IBRION\|1}}, VASP uses a RMM-DIIS algorithm to relax the ions into their instantaneous groundstate.
	RMM-DIIS stands for ''residual-minimization method, direct inversion of the iterative subspace''.{{cite\|pulay:cpl:1980}}		RMM-DIIS stands for ''residual-minimization method, direct inversion in the iterative subspace''.{{cite\|pulay:cpl:1980}}
	The forces and the stress tensor determine the search directions for finding the equilibrium positions; the total energy is not taken into account.		The forces and the stress tensor determine the search directions for finding the equilibrium positions; the total energy is not taken into account.
	This algorithm is very fast and efficient close to a local minimum, but fails badly if the initial positions are a bad guess (use [[#Conjugate gradient\|conjugate gradient]] instead).		This algorithm is very fast and efficient close to a local minimum, but fails badly if the initial positions are a bad guess (use [[#Conjugate gradient\|conjugate gradient]] instead).

Schlipf: /* References */

2024-10-18T11:15:38Z

References

← Older revision		Revision as of 11:15, 18 October 2024
Line 190:		Line 190:
	<references/>		<references/>

	[[Category:Ionic minimization]][[Category:Howto]]		[[Category:Ionic minimization]][[Category:Howto]][[Category:Forces]]

Schlipf at 11:13, 18 October 2024

2024-10-18T11:13:39Z

← Older revision		Revision as of 11:13, 18 October 2024
Line 180:		Line 180:
	In both cases ''g(F)'' and ''g(S)'' are the norm of forces and stress.		In both cases ''g(F)'' and ''g(S)'' are the norm of forces and stress.
	The extra line in the latter case is a sanity check on the average velocities and should produce vanishingly small numbers.		The extra line in the latter case is a sanity check on the average velocities and should produce vanishingly small numbers.

	~~== References ==~~
	~~<references/>~~

	== Related tags and articles ==		== Related tags and articles ==
Line 189:		Line 186:
	{{TAG\|NSW}},		{{TAG\|NSW}},
	{{TAG\|SMASS}}		{{TAG\|SMASS}}

			== References ==
			<references/>

	[[Category:Ionic minimization]][[Category:Howto]]		[[Category:Ionic minimization]][[Category:Howto]]

Schlipf: /* References */

2024-10-18T11:13:05Z

References

← Older revision		Revision as of 11:13, 18 October 2024
Line 183:		Line 183:
	== References ==		== References ==
	<references/>		<references/>

			== Related tags and articles ==
			{{TAG\|IBRION}},
			{{TAG\|POTIM}},
			{{TAG\|NSW}},
			{{TAG\|SMASS}}

	[[Category:Ionic minimization]][[Category:Howto]]		[[Category:Ionic minimization]][[Category:Howto]]

Schlipf at 11:11, 18 October 2024

2024-10-18T11:11:59Z

← Older revision		Revision as of 11:11, 18 October 2024
Line 183:		Line 183:
	== References ==		== References ==
	<references/>		<references/>

			[[Category:Ionic minimization]][[Category:Howto]]

Schlipf: Schlipf moved page VASP Wiki:Structure optimization to Structure optimization

2024-10-18T11:06:01Z

Schlipf moved page VASP Wiki:Structure optimization to Structure optimization

← Older revision		Revision as of 11:06, 18 October 2024
(No difference)

← Older revision		Revision as of 08:05, 24 October 2025
Line 22:		Line 22:
	In Figure 1, we show some heuristic rules that may help to decide with this selection.		In Figure 1, we show some heuristic rules that may help to decide with this selection.
	These guidelines are a compromise between speed and robustness of the algorithms.		These guidelines are a compromise between speed and robustness of the algorithms.
	The [[#RMM-DIIS\|RMM-DIIS algorithm]] ({{~~TAGO~~\|IBRION\|1}}) is very efficient because it uses the history of many steps to obtain the best next guess.		The [[#RMM-DIIS\|RMM-DIIS algorithm]] ({{TAG\|IBRION\|1}}) is very efficient because it uses the history of many steps to obtain the best next guess.
	However, if your structure is still far from the minimum, it is not wise too include the information from these points far from the minimum and the algorithm struggles.		However, if your structure is still far from the minimum, it is not wise too include the information from these points far from the minimum and the algorithm struggles.
	The [[#Conjugate gradient\|conjugate-gradient (CG) algorithm]] ({{~~TAGO~~\|IBRION\|2}}) chooses a search direction conjugate to previous ones.		The [[#Conjugate gradient\|conjugate-gradient (CG) algorithm]] ({{TAG\|IBRION\|2}}) chooses a search direction conjugate to previous ones.
	Then it selects an optimal step size along this search direction.		Then it selects an optimal step size along this search direction.
	This algorithm is a good default choice because of its robustness.		This algorithm is a good default choice because of its robustness.
	Even simpler are [[#Damped molecular dynamics\|damped molcular dynamics]] ({{~~TAGO~~\|IBRION\|3}}).		Even simpler are [[#Damped molecular dynamics\|damped molcular dynamics]] ({{TAG\|IBRION\|3}}).
	This algorithm propagates through time using the forces and friction on the velocities.		This algorithm propagates through time using the forces and friction on the velocities.

Line 33:		Line 33:
	This step size has no direct physical meaning but scales the forces internally before calling the minimization routine.		This step size has no direct physical meaning but scales the forces internally before calling the minimization routine.
	For many systems, the optimal {{TAG\|POTIM}} is around 0.5.		For many systems, the optimal {{TAG\|POTIM}} is around 0.5.
	Since the RMM-DIIS algorithm and the damped molecular dynamics are sensitive this parameter, use the conjugate gradient algorithm ({{~~TAGO~~\|IBRION}}=2), if you are not sure how large the optimal {{TAG\|POTIM}} is.		Since the RMM-DIIS algorithm and the damped molecular dynamics are sensitive this parameter, use the conjugate gradient algorithm ({{TAG\|IBRION}}=2), if you are not sure how large the optimal {{TAG\|POTIM}} is.
	In this case, the {{FILE\|OUTCAR}} file and <tt>stdout</tt> will contain a line indicating a reliable {{TAG\|POTIM}}.		In this case, the {{FILE\|OUTCAR}} file and <tt>stdout</tt> will contain a line indicating a reliable {{TAG\|POTIM}}.
	For {{~~TAGO~~\|IBRION\|2}}, the following lines will be written to <tt>stdout</tt> after each corrector step (usually each odd step):		For {{TAG\|IBRION\|2}}, the following lines will be written to <tt>stdout</tt> after each corrector step (usually each odd step):

	trial: gam= 0.20820 g(F)= 0.494E+00 g(S)= 0.000E+00 ort = 0.728E-03 (trialstep = 0.881E+00)		trial: gam= 0.20820 g(F)= 0.494E+00 g(S)= 0.000E+00 ort = 0.728E-03 (trialstep = 0.881E+00)
Line 41:		Line 41:
	The quantity <tt>trialstep</tt> is the size of the current trialstep.		The quantity <tt>trialstep</tt> is the size of the current trialstep.
	Multiply the current value of {{TAG\|POTIM}} with this number to obtain the optimal step size.		Multiply the current value of {{TAG\|POTIM}} with this number to obtain the optimal step size.
	{{NB\|mind\|If you use the default value of {{~~TAGO~~\|ISYM\|2}}, lower symmetry structures will be inaccessible to the relaxation algorithms. It is usually preferable to break symmetries intentionally by modifying the starting structure instead of turning symmetry operations off via {{~~TAGO~~\|ISYM\|0}}.}}		{{NB\|mind\|If you use the default value of {{TAG\|ISYM\|2}}, lower symmetry structures will be inaccessible to the relaxation algorithms. It is usually preferable to break symmetries intentionally by modifying the starting structure instead of turning symmetry operations off via {{TAG\|ISYM\|0}}.}}

	== RMM-DIIS ==		== RMM-DIIS ==

	For {{~~TAGO~~\|IBRION\|1}}, VASP uses a RMM-DIIS algorithm to relax the ions into their instantaneous groundstate.		For {{TAG\|IBRION\|1}}, VASP uses a RMM-DIIS algorithm to relax the ions into their instantaneous groundstate.
	RMM-DIIS stands for ''residual-minimization method, direct inversion in the iterative subspace''.{{cite\|pulay:cpl:1980}}		RMM-DIIS stands for ''residual-minimization method, direct inversion in the iterative subspace''.{{cite\|pulay:cpl:1980}}
	The forces and the stress tensor determine the search directions for finding the equilibrium positions; the total energy is not taken into account.		The forces and the stress tensor determine the search directions for finding the equilibrium positions; the total energy is not taken into account.
Line 53:		Line 53:
	The approximation of the Hessian matrix requires very accurate forces, otherwise the algorithm will fail to converge.		The approximation of the Hessian matrix requires very accurate forces, otherwise the algorithm will fail to converge.
	Enforcing a minimum of electronic steps between each ionic step ({{TAG\|NELMIN}}) converges forces efficiently.		Enforcing a minimum of electronic steps between each ionic step ({{TAG\|NELMIN}}) converges forces efficiently.
	For simple bulk materials {{~~TAGO~~\|NELMIN\|4}} is usually adequate, whereas complex surfaces, where the charge density converges very slowly, might require {{~~TAGO~~\|NELMIN\|8}}.		For simple bulk materials {{TAG\|NELMIN\|4}} is usually adequate, whereas complex surfaces, where the charge density converges very slowly, might require {{TAG\|NELMIN\|8}}.

	On startup, the initial Hessian matrix is diagonal and equal to {{TAG\|POTIM}}.		On startup, the initial Hessian matrix is diagonal and equal to {{TAG\|POTIM}}.
Line 63:		Line 63:

	A reasonable {{TAG\|POTIM}} can speed up calculations significantly.		A reasonable {{TAG\|POTIM}} can speed up calculations significantly.
	We recommend to find an optimal {{TAG\|POTIM}} using {{~~TAGO~~\|IBRION\|2}} or performing a few test calculations (see above).		We recommend to find an optimal {{TAG\|POTIM}} using {{TAG\|IBRION\|2}} or performing a few test calculations (see above).

	=== Understanding the output ===		=== Understanding the output ===
Line 87:		Line 87:
	== Conjugate gradient ==		== Conjugate gradient ==

	In the conjugate-gradient algorithm {{~~TAGO~~\|IBRION\|2}}, we optimize the structure along a search direction.		In the conjugate-gradient algorithm {{TAG\|IBRION\|2}}, we optimize the structure along a search direction.
	The initial search direction is given by forces and stress;		The initial search direction is given by forces and stress;
	in subsequent steps, we require that the search is conjugate (perpendicular) to the previous direction.		in subsequent steps, we require that the search is conjugate (perpendicular) to the previous direction.
Line 143:		Line 143:

	== Damped molecular dynamics ==		== Damped molecular dynamics ==
	With {{~~TAGO~~\|IBRION\|3}} and {{TAG\|SMASS}}, VASP will execute a damped second order equation of motion for the degrees of freedom		With {{TAG\|IBRION\|3}} and {{TAG\|SMASS}}, VASP will execute a damped second order equation of motion for the degrees of freedom
	:<math>		:<math>
	{\ddot {\vec x}} = -2 \alpha {\vec F} - \mu {\dot {\vec x}},		{\ddot {\vec x}} = -2 \alpha {\vec F} - \mu {\dot {\vec x}},
Line 166:		Line 166:
	If {{TAG\|POTIM}} and {{TAG\|SMASS}} are chosen correctly, the damped molecular dynamics mode may outperform the conjugate gradient method by a factor of two.		If {{TAG\|POTIM}} and {{TAG\|SMASS}} are chosen correctly, the damped molecular dynamics mode may outperform the conjugate gradient method by a factor of two.

	If {{~~TAGO~~\|SMASS\|0\|op=<}} a velocity quench algorithm is used.		If {{TAG\|SMASS\|0\|op=<}} a velocity quench algorithm is used.
	:<math>		:<math>
	\vec v_{\rm quench} = \rm{max}((\vec v + \alpha \vec F) \cdot \vec F, 0) \frac{\vec F}{F^2} + \alpha \vec F		\vec v_{\rm quench} = \rm{max}((\vec v + \alpha \vec F) \cdot \vec F, 0) \frac{\vec F}{F^2} + \alpha \vec F
	</math>		</math>
	You can see that velocities are projected onto the forces and only remain if that projection is a positive number.		You can see that velocities are projected onto the forces and only remain if that projection is a positive number.
	{{NB\|mind\|For {{~~TAGO~~\|IBRION\|3}}, a reasonable time step {{TAG\|POTIM}} ''must'' be supplied. Too large time steps will result in divergence, too small ones will slow down the convergence. A good choice is usually twice the ''smallest'' step size you would observe with the conjugate gradient algorithm.}}		{{NB\|mind\|For {{TAG\|IBRION\|3}}, a reasonable time step {{TAG\|POTIM}} ''must'' be supplied. Too large time steps will result in divergence, too small ones will slow down the convergence. A good choice is usually twice the ''smallest'' step size you would observe with the conjugate gradient algorithm.}}

	=== Understanding the output ===		=== Understanding the output ===