[SIESTA-L] Structure does not converge
Dear Herbert, Thanks for your reply. I have added these lines now (some of which I had included already, but Some were not in my input), and that indeed gives the right output. The ANI file can be read by Chemcraft as you said. I also experience some difficulties with gdis showing the structures Properly, as you pointed out in your last post. There appears to be a large number of structure plotted on top of each other. But still, in principle it is working now. Looking at the xyz file with Chemcraft is also very useful at this stage. Best Tobias Dr. Tobias Kraemer Research Associate Institute of Chemical Sciences School of Engineering Physical Sciences Heriot-Watt University Edinburgh EH14 4AS United Kingdom uut.krae...@hw.ac.uk tt+44 (0)131 451 3259 -Original Message- From: siesta-l-requ...@uam.es [mailto:siesta-l-requ...@uam.es] On Behalf Of Herbert Fruchtl Sent: 04 February 2014 15:29 To: siesta-l@uam.es Subject: Re: [SIESTA-L] Structure does not converge Hi Tobias, You need to set the WriteCoorStep value. Here's a section from a random input I once ran: WriteCoorInitaltrue WriteCoorSteptrue WriteForces true WriteCoorXmol true WriteCoorCerius false WriteMDXmol true WriteMDhistory true The first two lines tell it to write the geometry (initial and in every step) into the output file. WriteCoorXmol tells it to write the final geometry into seed.xyz. WriteMDXmol creates seed.ANI, which has the format of an xyz movie (I usually create a link ln -s test.ANI trj.xyz. As I mentioned before, the xyz format doesn't contain cell information, but is more reliable when used with gdis. Anything with MD in it also applies to geometry optimisation (which you can just see as a special case of dynamics). HTH, Herbert On 04/02/14 13:38, Kraemer, Tobias wrote: Dear Herbert, I have installed gdis in the meantime. I was able to open the fdf file without problems, And I can nicely visualise the unit cell that I used as input. However, gdis doesn't find Any coordinates in the SIESTA output (you refer to this as the .sot file). I am not sure I am looking at the right file. My output doesn't contain coordinates which are updated During the run, only the input geometry in the @reading standard input section, later on Only forces are reported. Best Tobias -Original Message- From: siesta-l-requ...@uam.es [mailto:siesta-l-requ...@uam.es] On Behalf Of Herbert Fruchtl Sent: 28 January 2014 11:53 To: siesta-l@uam.es Subject: Re: [SIESTA-L] Structure does not converge Hi Tobias, Try gdis for visualisation. It can shoe you a movie from a .sot (standard SIESTA output), even though it sometimes does it wrong. Haven't found out yet when this happens. You can also create an xyz (not periodic, but readable by most viewing programs) movie with the WriteMDxmol flag. Rename the .ANI file to something ending with .xyz. Which optimizer do you use? I find CG safest. HTH, Herbert On 28/01/14 10:07, Kraemer, Tobias wrote: Dear Abraham, Thanks for your detailed response. I did check the convergence of my calculations, and I should have included this in My previous post. So here is what happens in this calculation: 1)The force oscillates quite significantly within the set 500 optimisation steps, and taken that the MaxForceTol is set to 0.04 eV/Ang, This seems to be far from convergence to me. Here is a snapshot of the last 10 steps. Max0.580458constrained Max0.174727constrained Max0.082654constrained Max0.080264constrained Max0.115938constrained Max0.354063constrained Max0.164293constrained Max0.236174constrained Max0.114993constrained Max0.060479constrained 2)The unit cell seems to undergo some significant change, at least to my eye. outcell: Cell vector modules (Ang) :9.534700 11.749500 15.878700 outcell: Cell vector modules (Ang) :9.675640 11.898404 16.056706 outcell: Cell vector modules (Ang) :9.901185 12.136756 16.341658 [...] outcell: Cell vector modules (Ang) :9.017619 11.059155 15.490068 outcell: Cell vector modules (Ang) :9.018024 11.060081 15.486998 outcell: Cell vector modules (Ang) :9.018673 11.061565 15.482088 outcell: Cell vector modules (Ang) :9.018155 11.060380 15.486008 outcell: Cell vector modules (Ang) :9.018614 11.058782 15.482515 outcell: Cell vector modules (Ang) :9.018461 11.059315 15.483680 outcell: Cell vector modules (Ang) :9.020634 11.060348 15.484791 outcell: Cell vector modules (Ang) :9.024112 11.062001 15.486571 Inspecting the last structure, I can at least say that the structure does
RE: [SIESTA-L] Structure does not converge
Dear Herbert, I have installed gdis in the meantime. I was able to open the fdf file without problems, And I can nicely visualise the unit cell that I used as input. However, gdis doesn't find Any coordinates in the SIESTA output (you refer to this as the .sot file). I am not sure I am looking at the right file. My output doesn't contain coordinates which are updated During the run, only the input geometry in the @reading standard input section, later on Only forces are reported. Best Tobias -Original Message- From: siesta-l-requ...@uam.es [mailto:siesta-l-requ...@uam.es] On Behalf Of Herbert Fruchtl Sent: 28 January 2014 11:53 To: siesta-l@uam.es Subject: Re: [SIESTA-L] Structure does not converge Hi Tobias, Try gdis for visualisation. It can shoe you a movie from a .sot (standard SIESTA output), even though it sometimes does it wrong. Haven't found out yet when this happens. You can also create an xyz (not periodic, but readable by most viewing programs) movie with the WriteMDxmol flag. Rename the .ANI file to something ending with .xyz. Which optimizer do you use? I find CG safest. HTH, Herbert On 28/01/14 10:07, Kraemer, Tobias wrote: Dear Abraham, Thanks for your detailed response. I did check the convergence of my calculations, and I should have included this in My previous post. So here is what happens in this calculation: 1)The force oscillates quite significantly within the set 500 optimisation steps, and taken that the MaxForceTol is set to 0.04 eV/Ang, This seems to be far from convergence to me. Here is a snapshot of the last 10 steps. Max0.580458constrained Max0.174727constrained Max0.082654constrained Max0.080264constrained Max0.115938constrained Max0.354063constrained Max0.164293constrained Max0.236174constrained Max0.114993constrained Max0.060479constrained 2)The unit cell seems to undergo some significant change, at least to my eye. outcell: Cell vector modules (Ang) :9.534700 11.749500 15.878700 outcell: Cell vector modules (Ang) :9.675640 11.898404 16.056706 outcell: Cell vector modules (Ang) :9.901185 12.136756 16.341658 [...] outcell: Cell vector modules (Ang) :9.017619 11.059155 15.490068 outcell: Cell vector modules (Ang) :9.018024 11.060081 15.486998 outcell: Cell vector modules (Ang) :9.018673 11.061565 15.482088 outcell: Cell vector modules (Ang) :9.018155 11.060380 15.486008 outcell: Cell vector modules (Ang) :9.018614 11.058782 15.482515 outcell: Cell vector modules (Ang) :9.018461 11.059315 15.483680 outcell: Cell vector modules (Ang) :9.020634 11.060348 15.484791 outcell: Cell vector modules (Ang) :9.024112 11.062001 15.486571 Inspecting the last structure, I can at least say that the structure does not blow-up or do anything that is unpredicted (I am looking at This from the perspective of someone who has done gas phase DFT / ab initio calculations on molecular/cluster systems, but I guess There is some common ground here). I need to figure out a way though how to visualise single steps of the optimisation itself, Like I would do in one of my more regular calculations. I am sure there must be a way to do this with XCrysden (the program I am using Currently). So following what you have said in your email, I will try your suggestions and play around with the optimisation protocol. However, what still worries me is if this possibly has something to do with the (downloaded) pseudopotentials I have chosen. I feel there are a lot of things that can go wrong here Anyways, I will keep you posted. Thanks again for your comments. Tobias *From:*siesta-l-requ...@uam.es [mailto:siesta-l-requ...@uam.es] *On Behalf Of *Abrah am Hmiel *Sent:* 24 January 2014 20:40 *To:* Siesta, Self-Consistent DFT LCAO program, http://www.uam.es/siesta *Subject:* Re: [SIESTA-L] Structure does not converge Tobias, I'm curious, what do the results of the following commands look like? 1) fgrep constrained 1.out 2) fgrep outcell: Cell vector modules (Ang) 1.out *where 1.out is the name of the output log file You don't have to print them and send them our way, as they'll be very large, but what can you say about them in general? Are the forces fluctuating about a small amount that isn't quite as small as the threshold you've chosen? Are the lattice parameters very different from what you started with? Is everything blowing up? In general, this is what I do. Instead of doing one relaxation with a very large number of CG steps, perform several in steps, making sure the UseSaveData flag is set to true so that you don't have to come up with a new density matrix each time and can use the previous CG trajectory
Re: [SIESTA-L] Structure does not converge
Hi Tobias, You need to set the WriteCoorStep value. Here's a section from a random input I once ran: WriteCoorInitaltrue WriteCoorSteptrue WriteForces true WriteCoorXmol true WriteCoorCerius false WriteMDXmol true WriteMDhistory true The first two lines tell it to write the geometry (initial and in every step) into the output file. WriteCoorXmol tells it to write the final geometry into seed.xyz. WriteMDXmol creates seed.ANI, which has the format of an xyz movie (I usually create a link ln -s test.ANI trj.xyz. As I mentioned before, the xyz format doesn't contain cell information, but is more reliable when used with gdis. Anything with MD in it also applies to geometry optimisation (which you can just see as a special case of dynamics). HTH, Herbert On 04/02/14 13:38, Kraemer, Tobias wrote: Dear Herbert, I have installed gdis in the meantime. I was able to open the fdf file without problems, And I can nicely visualise the unit cell that I used as input. However, gdis doesn't find Any coordinates in the SIESTA output (you refer to this as the .sot file). I am not sure I am looking at the right file. My output doesn't contain coordinates which are updated During the run, only the input geometry in the @reading standard input section, later on Only forces are reported. Best Tobias -Original Message- From: siesta-l-requ...@uam.es [mailto:siesta-l-requ...@uam.es] On Behalf Of Herbert Fruchtl Sent: 28 January 2014 11:53 To: siesta-l@uam.es Subject: Re: [SIESTA-L] Structure does not converge Hi Tobias, Try gdis for visualisation. It can shoe you a movie from a .sot (standard SIESTA output), even though it sometimes does it wrong. Haven't found out yet when this happens. You can also create an xyz (not periodic, but readable by most viewing programs) movie with the WriteMDxmol flag. Rename the .ANI file to something ending with .xyz. Which optimizer do you use? I find CG safest. HTH, Herbert On 28/01/14 10:07, Kraemer, Tobias wrote: Dear Abraham, Thanks for your detailed response. I did check the convergence of my calculations, and I should have included this in My previous post. So here is what happens in this calculation: 1)The force oscillates quite significantly within the set 500 optimisation steps, and taken that the MaxForceTol is set to 0.04 eV/Ang, This seems to be far from convergence to me. Here is a snapshot of the last 10 steps. Max0.580458constrained Max0.174727constrained Max0.082654constrained Max0.080264constrained Max0.115938constrained Max0.354063constrained Max0.164293constrained Max0.236174constrained Max0.114993constrained Max0.060479constrained 2)The unit cell seems to undergo some significant change, at least to my eye. outcell: Cell vector modules (Ang) :9.534700 11.749500 15.878700 outcell: Cell vector modules (Ang) :9.675640 11.898404 16.056706 outcell: Cell vector modules (Ang) :9.901185 12.136756 16.341658 [...] outcell: Cell vector modules (Ang) :9.017619 11.059155 15.490068 outcell: Cell vector modules (Ang) :9.018024 11.060081 15.486998 outcell: Cell vector modules (Ang) :9.018673 11.061565 15.482088 outcell: Cell vector modules (Ang) :9.018155 11.060380 15.486008 outcell: Cell vector modules (Ang) :9.018614 11.058782 15.482515 outcell: Cell vector modules (Ang) :9.018461 11.059315 15.483680 outcell: Cell vector modules (Ang) :9.020634 11.060348 15.484791 outcell: Cell vector modules (Ang) :9.024112 11.062001 15.486571 Inspecting the last structure, I can at least say that the structure does not blow-up or do anything that is unpredicted (I am looking at This from the perspective of someone who has done gas phase DFT / ab initio calculations on molecular/cluster systems, but I guess There is some common ground here). I need to figure out a way though how to visualise single steps of the optimisation itself, Like I would do in one of my more regular calculations. I am sure there must be a way to do this with XCrysden (the program I am using Currently). So following what you have said in your email, I will try your suggestions and play around with the optimisation protocol. However, what still worries me is if this possibly has something to do with the (downloaded) pseudopotentials I have chosen. I feel there are a lot of things that can go wrong here Anyways, I will keep you posted. Thanks again for your comments. Tobias *From:*siesta-l-requ...@uam.es [mailto:siesta-l-requ...@uam.es] *On Behalf Of *Abrah am Hmiel *Sent:* 24 January 2014 20:40 *To:* Siesta, Self-Consistent DFT LCAO program, http://www.uam.es/siesta *Subject:* Re: [SIESTA-L] Structure does not converge Tobias, I'm curious, what
RE: [SIESTA-L] Structure does not converge
Dear Abraham, Thanks for your detailed response. I did check the convergence of my calculations, and I should have included this in My previous post. So here is what happens in this calculation: 1) The force oscillates quite significantly within the set 500 optimisation steps, and taken that the MaxForceTol is set to 0.04 eV/Ang, This seems to be far from convergence to me. Here is a snapshot of the last 10 steps. Max0.580458constrained Max0.174727constrained Max0.082654constrained Max0.080264constrained Max0.115938constrained Max0.354063constrained Max0.164293constrained Max0.236174constrained Max0.114993constrained Max0.060479constrained 2) The unit cell seems to undergo some significant change, at least to my eye. outcell: Cell vector modules (Ang) :9.534700 11.749500 15.878700 outcell: Cell vector modules (Ang) :9.675640 11.898404 16.056706 outcell: Cell vector modules (Ang) :9.901185 12.136756 16.341658 [...] outcell: Cell vector modules (Ang) :9.017619 11.059155 15.490068 outcell: Cell vector modules (Ang) :9.018024 11.060081 15.486998 outcell: Cell vector modules (Ang) :9.018673 11.061565 15.482088 outcell: Cell vector modules (Ang) :9.018155 11.060380 15.486008 outcell: Cell vector modules (Ang) :9.018614 11.058782 15.482515 outcell: Cell vector modules (Ang) :9.018461 11.059315 15.483680 outcell: Cell vector modules (Ang) :9.020634 11.060348 15.484791 outcell: Cell vector modules (Ang) :9.024112 11.062001 15.486571 Inspecting the last structure, I can at least say that the structure does not blow-up or do anything that is unpredicted (I am looking at This from the perspective of someone who has done gas phase DFT / ab initio calculations on molecular/cluster systems, but I guess There is some common ground here). I need to figure out a way though how to visualise single steps of the optimisation itself, Like I would do in one of my more regular calculations. I am sure there must be a way to do this with XCrysden (the program I am using Currently). So following what you have said in your email, I will try your suggestions and play around with the optimisation protocol. However, what still worries me is if this possibly has something to do with the (downloaded) pseudopotentials I have chosen. I feel there are a lot of things that can go wrong here Anyways, I will keep you posted. Thanks again for your comments. Tobias From: siesta-l-requ...@uam.es [mailto:siesta-l-requ...@uam.es] On Behalf Of Abrah am Hmiel Sent: 24 January 2014 20:40 To: Siesta, Self-Consistent DFT LCAO program, http://www.uam.es/siesta Subject: Re: [SIESTA-L] Structure does not converge Tobias, I'm curious, what do the results of the following commands look like? 1) fgrep constrained 1.out 2) fgrep outcell: Cell vector modules (Ang) 1.out *where 1.out is the name of the output log file You don't have to print them and send them our way, as they'll be very large, but what can you say about them in general? Are the forces fluctuating about a small amount that isn't quite as small as the threshold you've chosen? Are the lattice parameters very different from what you started with? Is everything blowing up? In general, this is what I do. Instead of doing one relaxation with a very large number of CG steps, perform several in steps, making sure the UseSaveData flag is set to true so that you don't have to come up with a new density matrix each time and can use the previous CG trajectory information as well: 1) Do one with MD.NumCGSteps ~ 10 and MD.MaxCGDispl ~ 0.15 Ang, to perform a very coarse relaxation (it is unlikely you will meet the threshold, and that's okay. You're not done). Call the output file 1.out or something like that. 2) Do one with MD.NumCGSteps ~ 20 and MD.MaxCGDispl ~ 0.05 Ang, to perform a coarse relaxation. Call the output file 2.out 3) Do one with MD.NumCGSteps ~ 20 and MD.MaxCGDispl ~ 0.01 Ang, to perform a finer relaxation. Call the output file 3.out ... as many successive steps as you need with as much fineness as desired or until you meet your force convergence criteria, at which point you will have a relaxed structure. If 'everything is blowing up' then you may have to take another look at your simulation parameters. Sometimes choosing the wrong Pulay mixing parameters will prevent your simulation from falling into a convergence basin, so another look at that may be helpful. Transition metals sometimes call for using small (~0.01 or smaller) DM.MixingWeight with a kick every several steps (see the manual). Warm regards, -- Abraham Hmiel Katherine Belz Groves Fellow in Nanoscience Xue Group, SUNY College of Nanoscale Science and Engineering http://abehmiel.net/about On Fri, Jan 24, 2014 at 1:09 PM, Kraemer, Tobias t.krae
Re: [SIESTA-L] Structure does not converge
Hi Tobias, Try gdis for visualisation. It can shoe you a movie from a .sot (standard SIESTA output), even though it sometimes does it wrong. Haven't found out yet when this happens. You can also create an xyz (not periodic, but readable by most viewing programs) movie with the WriteMDxmol flag. Rename the .ANI file to something ending with .xyz. Which optimizer do you use? I find CG safest. HTH, Herbert On 28/01/14 10:07, Kraemer, Tobias wrote: Dear Abraham, Thanks for your detailed response. I did check the convergence of my calculations, and I should have included this in My previous post. So here is what happens in this calculation: 1)The force oscillates quite significantly within the set 500 optimisation steps, and taken that the MaxForceTol is set to 0.04 eV/Ang, This seems to be far from convergence to me. Here is a snapshot of the last 10 steps. Max0.580458constrained Max0.174727constrained Max0.082654constrained Max0.080264constrained Max0.115938constrained Max0.354063constrained Max0.164293constrained Max0.236174constrained Max0.114993constrained Max0.060479constrained 2)The unit cell seems to undergo some significant change, at least to my eye. outcell: Cell vector modules (Ang) :9.534700 11.749500 15.878700 outcell: Cell vector modules (Ang) :9.675640 11.898404 16.056706 outcell: Cell vector modules (Ang) :9.901185 12.136756 16.341658 [...] outcell: Cell vector modules (Ang) :9.017619 11.059155 15.490068 outcell: Cell vector modules (Ang) :9.018024 11.060081 15.486998 outcell: Cell vector modules (Ang) :9.018673 11.061565 15.482088 outcell: Cell vector modules (Ang) :9.018155 11.060380 15.486008 outcell: Cell vector modules (Ang) :9.018614 11.058782 15.482515 outcell: Cell vector modules (Ang) :9.018461 11.059315 15.483680 outcell: Cell vector modules (Ang) :9.020634 11.060348 15.484791 outcell: Cell vector modules (Ang) :9.024112 11.062001 15.486571 Inspecting the last structure, I can at least say that the structure does not blow-up or do anything that is unpredicted (I am looking at This from the perspective of someone who has done gas phase DFT / ab initio calculations on molecular/cluster systems, but I guess There is some common ground here). I need to figure out a way though how to visualise single steps of the optimisation itself, Like I would do in one of my more regular calculations. I am sure there must be a way to do this with XCrysden (the program I am using Currently). So following what you have said in your email, I will try your suggestions and play around with the optimisation protocol. However, what still worries me is if this possibly has something to do with the (downloaded) pseudopotentials I have chosen. I feel there are a lot of things that can go wrong here Anyways, I will keep you posted. Thanks again for your comments. Tobias *From:*siesta-l-requ...@uam.es [mailto:siesta-l-requ...@uam.es] *On Behalf Of *Abrah am Hmiel *Sent:* 24 January 2014 20:40 *To:* Siesta, Self-Consistent DFT LCAO program, http://www.uam.es/siesta *Subject:* Re: [SIESTA-L] Structure does not converge Tobias, I'm curious, what do the results of the following commands look like? 1) fgrep constrained 1.out 2) fgrep outcell: Cell vector modules (Ang) 1.out *where 1.out is the name of the output log file You don't have to print them and send them our way, as they'll be very large, but what can you say about them in general? Are the forces fluctuating about a small amount that isn't quite as small as the threshold you've chosen? Are the lattice parameters very different from what you started with? Is everything blowing up? In general, this is what I do. Instead of doing one relaxation with a very large number of CG steps, perform several in steps, making sure the UseSaveData flag is set to true so that you don't have to come up with a new density matrix each time and can use the previous CG trajectory information as well: 1) Do one with MD.NumCGSteps ~ 10 and MD.MaxCGDispl ~ 0.15 Ang, to perform a very coarse relaxation (it is unlikely you will meet the threshold, and that's okay. You're not done). Call the output file 1.out or something like that. 2) Do one with MD.NumCGSteps ~ 20 and MD.MaxCGDispl ~ 0.05 Ang, to perform a coarse relaxation. Call the output file 2.out 3) Do one with MD.NumCGSteps ~ 20 and MD.MaxCGDispl ~ 0.01 Ang, to perform a finer relaxation. Call the output file 3.out ... as many successive steps as you need with as much fineness as desired or until you meet your force convergence criteria, at which point you will have a relaxed structure. If 'everything is blowing up' then you may have to take another look at your simulation parameters. Sometimes choosing the wrong Pulay mixing parameters
RE: [SIESTA-L] Structure does not converge
Hi Herbert, Thanks for your suggestions. Will try gdis, sounds like what I am looking for. For the optimizer I did use conjugate gradients indeed. Tobias -Original Message- From: siesta-l-requ...@uam.es [mailto:siesta-l-requ...@uam.es] On Behalf Of Herbert Fruchtl Sent: 28 January 2014 11:53 To: siesta-l@uam.es Subject: Re: [SIESTA-L] Structure does not converge Hi Tobias, Try gdis for visualisation. It can shoe you a movie from a .sot (standard SIESTA output), even though it sometimes does it wrong. Haven't found out yet when this happens. You can also create an xyz (not periodic, but readable by most viewing programs) movie with the WriteMDxmol flag. Rename the .ANI file to something ending with .xyz. Which optimizer do you use? I find CG safest. HTH, Herbert On 28/01/14 10:07, Kraemer, Tobias wrote: Dear Abraham, Thanks for your detailed response. I did check the convergence of my calculations, and I should have included this in My previous post. So here is what happens in this calculation: 1)The force oscillates quite significantly within the set 500 optimisation steps, and taken that the MaxForceTol is set to 0.04 eV/Ang, This seems to be far from convergence to me. Here is a snapshot of the last 10 steps. Max0.580458constrained Max0.174727constrained Max0.082654constrained Max0.080264constrained Max0.115938constrained Max0.354063constrained Max0.164293constrained Max0.236174constrained Max0.114993constrained Max0.060479constrained 2)The unit cell seems to undergo some significant change, at least to my eye. outcell: Cell vector modules (Ang) :9.534700 11.749500 15.878700 outcell: Cell vector modules (Ang) :9.675640 11.898404 16.056706 outcell: Cell vector modules (Ang) :9.901185 12.136756 16.341658 [...] outcell: Cell vector modules (Ang) :9.017619 11.059155 15.490068 outcell: Cell vector modules (Ang) :9.018024 11.060081 15.486998 outcell: Cell vector modules (Ang) :9.018673 11.061565 15.482088 outcell: Cell vector modules (Ang) :9.018155 11.060380 15.486008 outcell: Cell vector modules (Ang) :9.018614 11.058782 15.482515 outcell: Cell vector modules (Ang) :9.018461 11.059315 15.483680 outcell: Cell vector modules (Ang) :9.020634 11.060348 15.484791 outcell: Cell vector modules (Ang) :9.024112 11.062001 15.486571 Inspecting the last structure, I can at least say that the structure does not blow-up or do anything that is unpredicted (I am looking at This from the perspective of someone who has done gas phase DFT / ab initio calculations on molecular/cluster systems, but I guess There is some common ground here). I need to figure out a way though how to visualise single steps of the optimisation itself, Like I would do in one of my more regular calculations. I am sure there must be a way to do this with XCrysden (the program I am using Currently). So following what you have said in your email, I will try your suggestions and play around with the optimisation protocol. However, what still worries me is if this possibly has something to do with the (downloaded) pseudopotentials I have chosen. I feel there are a lot of things that can go wrong here Anyways, I will keep you posted. Thanks again for your comments. Tobias *From:*siesta-l-requ...@uam.es [mailto:siesta-l-requ...@uam.es] *On Behalf Of *Abrah am Hmiel *Sent:* 24 January 2014 20:40 *To:* Siesta, Self-Consistent DFT LCAO program, http://www.uam.es/siesta *Subject:* Re: [SIESTA-L] Structure does not converge Tobias, I'm curious, what do the results of the following commands look like? 1) fgrep constrained 1.out 2) fgrep outcell: Cell vector modules (Ang) 1.out *where 1.out is the name of the output log file You don't have to print them and send them our way, as they'll be very large, but what can you say about them in general? Are the forces fluctuating about a small amount that isn't quite as small as the threshold you've chosen? Are the lattice parameters very different from what you started with? Is everything blowing up? In general, this is what I do. Instead of doing one relaxation with a very large number of CG steps, perform several in steps, making sure the UseSaveData flag is set to true so that you don't have to come up with a new density matrix each time and can use the previous CG trajectory information as well: 1) Do one with MD.NumCGSteps ~ 10 and MD.MaxCGDispl ~ 0.15 Ang, to perform a very coarse relaxation (it is unlikely you will meet the threshold, and that's okay. You're not done). Call the output file 1.out or something like that. 2) Do one with MD.NumCGSteps ~ 20 and MD.MaxCGDispl ~ 0.05 Ang, to perform a coarse relaxation
[SIESTA-L] Structure does not converge
Dear SIESTA users, I am very new to the SIESTA code, so to start with I am familiarising myself with the protocols involved to run calculations with the code. I am trying to reproduce some numbers from a paper, where a crystal structure (triclinic system) of a Ru-allyl complex was relaxed (Organometallics, 2013, 32, 3784). I have set up a SIESTA input for relaxing this structure, using PBE with the default DZP basis sets for all atoms, along with pseudopotentials from the SIESTA database (I am aware that this might not be the best choice, but to start with it might be a fair option). The calculation seems to fail to converge to a minimum structure and stops after the default 500 opt cycles. This makes me suspicious, since my feeling is that the structure should have converged within 500 cycles. The SCF cycles converge fine. I have attached my input file below, the structure is read in from an external xyz file (attached). I have used as many default settings for the moment as possible. I would appreciate if you could share your experience and advice on this subject. Is such an input in principle a good starting point for the task at hand? What could I improve? Would the use of a Monkhorst-Pack grid be a better choice? There are some other Questions regarding dispersion corrections but first I need to play around with what I understand from the manual. Kind regards, thanks for everybody's help. Really trying to learn how to get reliable results. Tobias SystemName [Ru(AcCN)2(Cp*)(eta3-phenylallyl) unit cell SystemLabel ru_allyl NumberOfAtoms 160 NumberOfSpecies 7 WriteMullikenPop1 %block ChemicalSpeciesLabel 1 44 Ru # Species index, atomic number, species label 2 15 P 3 9 F 4 8 O 5 7 N 6 6 C 7 1 H %endblock ChemicalSpeciesLabel MeshCutoff 400.0 Ry NetCharge 0.00 XC.functional GGA XC.authors PBE SpinPolarized.false. DM.NumberPulay4 DM.MixingWeight 0.3 DM.UseSaveDM .true. NeglNonOverlapInt False AtomicCoordinatesFormat NotScaledCartesianAng AtomicCoordinatesAndAtomicSpecies ru_allylinput.xyz LatticeConstant 1.0 Ang %block LatticeParameters 9.5347 11.7495 15.8787 109.395 96.719 100.965 %endblock LatticeParameters WriteCoorXmol True WriteForces .true. WriteMullikenPop 1 SaveDeltaRho .true. MD.TypeOfRun cg # Type of dynamics: Conjugate gradients MD.NumCGsteps 500 # number of CG steps MD.MaxCGDispl 0.15 Ang MD.MaxForceTol 0.04 eV/Ang MD.UseSaveXV yes MD.VariableCell.true. Dr. Tobias Kraemer Research Associate Institute of Chemical Sciences School of Engineering Physical Sciences Heriot-Watt University Edinburgh EH14 4AS United Kingdom * t.krae...@hw.ac.uk * +44 (0)131 451 3259 - Sunday Times Scottish University of the Year 2011-2013 Top in the UK for student experience Fourth university in the UK and top in Scotland (National Student Survey 2012) We invite research leaders and ambitious early career researchers to join us in leading and driving research in key inter-disciplinary themes. Please see www.hw.ac.uk/researchleaders for further information and how to apply. Heriot-Watt University is a Scottish charity registered under charity number SC000278. ru_allylinput.xyz Description: ru_allylinput.xyz ru_allyl.fdf Description: ru_allyl.fdf
Re: [SIESTA-L] Structure does not converge
Tobias, I'm curious, what do the results of the following commands look like? 1) fgrep constrained 1.out 2) fgrep outcell: Cell vector modules (Ang) 1.out *where 1.out is the name of the output log file You don't have to print them and send them our way, as they'll be very large, but what can you say about them in general? Are the forces fluctuating about a small amount that isn't quite as small as the threshold you've chosen? Are the lattice parameters very different from what you started with? Is everything blowing up? In general, this is what I do. Instead of doing one relaxation with a very large number of CG steps, perform several in steps, making sure the UseSaveData flag is set to true so that you don't have to come up with a new density matrix each time and can use the previous CG trajectory information as well: 1) Do one with MD.NumCGSteps ~ 10 and MD.MaxCGDispl ~ 0.15 Ang, to perform a very coarse relaxation (it is unlikely you will meet the threshold, and that's okay. You're not done). Call the output file 1.out or something like that. 2) Do one with MD.NumCGSteps ~ 20 and MD.MaxCGDispl ~ 0.05 Ang, to perform a coarse relaxation. Call the output file 2.out 3) Do one with MD.NumCGSteps ~ 20 and MD.MaxCGDispl ~ 0.01 Ang, to perform a finer relaxation. Call the output file 3.out ... as many successive steps as you need with as much fineness as desired or until you meet your force convergence criteria, at which point you will have a relaxed structure. If 'everything is blowing up' then you may have to take another look at your simulation parameters. Sometimes choosing the wrong Pulay mixing parameters will prevent your simulation from falling into a convergence basin, so another look at that may be helpful. Transition metals sometimes call for using small (~0.01 or smaller) DM.MixingWeight with a kick every several steps (see the manual). Warm regards, -- *Abraham Hmiel* Katherine Belz Groves Fellow in Nanoscience Xue Group, SUNY College of Nanoscale Science and Engineering http://abehmiel.net/about On Fri, Jan 24, 2014 at 1:09 PM, Kraemer, Tobias t.krae...@hw.ac.uk wrote: Dear SIESTA users, I am very new to the SIESTA code, so to start with I am familiarising myself with the protocols involved to run calculations with the code. I am trying to reproduce some numbers from a paper, where a crystal structure (triclinic system) of a Ru-allyl complex was relaxed (Organometallics, 2013, 32, 3784). I have set up a SIESTA input for relaxing this structure, using PBE with the default DZP basis sets for all atoms, along with pseudopotentials from the SIESTA database (I am aware that this might not be the best choice, but to start with it might be a fair option). The calculation seems to fail to converge to a minimum structure and stops after the default 500 opt cycles. This makes me suspicious, since my feeling is that the structure should have converged within 500 cycles. The SCF cycles converge fine. I have attached my input file below, the structure is read in from an external xyz file (attached). I have used as many default settings for the moment as possible. I would appreciate if you could share your experience and advice on this subject. Is such an input in principle a good starting point for the task at hand? What could I improve? Would the use of a Monkhorst-Pack grid be a better choice? There are some other Questions regarding dispersion corrections but first I need to play around with what I understand from the manual. Kind regards, thanks for everybody’s help. Really trying to learn how to get reliable results. Tobias SystemName [Ru(AcCN)2(Cp*)(eta3-phenylallyl) unit cell SystemLabel ru_allyl NumberOfAtoms 160 NumberOfSpecies 7 WriteMullikenPop1 %block ChemicalSpeciesLabel 1 44 Ru # Species index, atomic number, species label 2 15 P 3 9 F 4 8 O 5 7 N 6 6 C 7 1 H %endblock ChemicalSpeciesLabel MeshCutoff 400.0 Ry NetCharge 0.00 XC.functional GGA XC.authors PBE SpinPolarized.false. DM.NumberPulay4 DM.MixingWeight 0.3 DM.UseSaveDM .true. NeglNonOverlapInt False AtomicCoordinatesFormat NotScaledCartesianAng AtomicCoordinatesAndAtomicSpecies ru_allylinput.xyz LatticeConstant 1.0 Ang %block LatticeParameters 9.5347 11.7495 15.8787 109.395 96.719 100.965 %endblock LatticeParameters WriteCoorXmol True WriteForces .true. WriteMullikenPop 1 SaveDeltaRho .true. MD.TypeOfRun cg # Type of dynamics: Conjugate gradients MD.NumCGsteps 500 # number of CG steps MD.MaxCGDispl 0.15 Ang MD.MaxForceTol 0.04 eV/Ang MD.UseSaveXV yes MD.VariableCell.true. Dr. Tobias Kraemer Research Associate Institute of
