Convergence - Science topic

Here are a few things that might help.

For a single atom, you don't need such a large unit cell, something like 10Å×10Å×10Å should be enough. Also, use ibrav=1 for a cubic lattice (you just need a box with the atom in it). As for K-mesh, use 1x1x1 since there is no periodicity to sample. Furthermore, there are four 'starting_magnetization(i)' why? ~keep just one (choose accordingly). Your values for ecutrho and ecutwfc are very high. Consider reducing them to something like 240 Ry and 40 Ry (4×ecutwfc for ecutrho is standard). Finally, ion_dynamics = 'bfgs' is unnecessary for a single atom. Instead, use calculation ='scf' or 'relax'.

These changes will very likely resolve your convergence issue.

Hope that helps,

Rgds

Rajkumar Shufen

Hi! I know that the question is posted a long time ago. I wonder if it worked with you.

The issue you are experiencing—non-convergence and failure of ZHEGV—is likely due to the inclusion of Yb, which involves f-electron states that require careful treatment in VASP. The ZHEGV error indicates an eigenvalue solver problem, which often arises due to numerical instabilities in systems with strongly correlated electrons, like those in lanthanide elements.

Here are several steps to address this issue:

Yb has partially filled f-electrons, which can create strong correlations and cause convergence issues. You may need to adjust the U value:

If LDAUU = 3 (eV) for Yb is too small or inappropriate, the self-consistent field (SCF) calculation can become unstable. Test a range of U values, e.g., 4-7 eV, which is more typical for Yb.

Modify:

LDAUU = 3 0 0 6

Since you are using LDAUTYPE = 2 (Dudarev), ensure that Ueff = U - J is meaningful. You might set:

LDAUJ = 0 0 0 0

Yb's strongly correlated states often cause SCF instabilities. Increase the maximum number of SCF cycles (NELM) to ensure convergence:

NELM = 200

Reduce mixing parameters to stabilize the SCF cycle. Add or adjust:

AMIX = 0.1

BMIX = 0.01

Switch the electronic minimization algorithm (ALGO) to a more stable setting for difficult cases involving f-electrons. Try the conjugate gradient (ALGO = All) or the block Davidson (ALGO = Fast) solver:

ALGO = All

Proper initialization of MAGMOM is crucial for systems with partially filled f-orbitals. Instead of assigning 4.0 for Yb's magnetic moment (which could be unrealistic), try:

MAGMOM = 79*0.6 1*1.0

Start with a smaller or neutral moment for Yb (1.0 or 0.6), as improper initialization can hinder convergence.

High-precision settings may cause aliasing errors, especially with large systems and f-electrons. Temporarily lower PREC to Normal and adjust the plane wave cutoff slightly:

PREC = Normal

ENCUT = 450

Explicitly set the FFT grid dimensions (NGX, NGY, NGZ) to resolve aliasing warnings:

NGX = 82

NGY = 82

NGZ = 58

Since the calculation is already unstable, smaller ionic step sizes might help:

POTIM = 0.3

Verify that the pseudopotential for Yb (PAW_PBE Yb 23Dec2003) is suitable for systems requiring f-electron treatment. Consider using Yb_sv or a GW-ready pseudopotential if available.

If the SCF calculation fails after several iterations, restart from a pre-converged charge density and wavefunction:

ICHARG = 1 # Read from CHGCAR

LWAVE = .TRUE. # Write the wavefunction file

I hope this can help

I am still awaiting a response on ResearchGate and would greatly appreciate any insights you might share.

Best regards, Rekib

Thank you Mohamed for taking the time to comment.

Respectfully yours;

Lucio

Definitely use the number of evaluations if you can't guarantee that the algorithms consume the same number of evaluations per generation/iteration.

Chi Yang Thank you so much for your suggestion.

Can you share the Silvaco code to develop tfet as a biosensor

Anuradha Chhillar Thank you for your answer but can you please explain a little bit about which parameter needs to be changed in the input file and correspondingly which parameters to check the convergence in the output file?

Jyri Kimari Each component of my model was built in material studio and then assembled in lammps

Manuscripts that meet the scope of this special issue can come to consult and enjoy a certain discount.

Hey Rikke,

Sounds like you’re close, but ORCA’s being picky with your transition state search!

So, that “no” in the convergence column for Max gradient and RMS gradient means that ORCA's optimizer hasn’t reached the criteria needed to call the geometry fully converged. The Max gradient and RMS gradient are not quite where ORCA wants them to be to consider the transition state legit. Essentially, it’s telling you, "Hey, I'm not happy with this geometry yet."

I would have the following suggestions to help nudge it to a “yes”:

I hope that this resolves your issue!

Best,

Igor

Enter your question in ChatGPT ;)

Then you will have the following answer:

In COMSOL, convergence issues with nonlinear boundary conditions, like your temperature-dependent heat flux Q(T)=aT3+bT2+cT+100010−5Q(T) = \frac{a T^3 + b T^2 + c T + 1000}{10^{-5}}Q(T)=10−5aT3+bT2+cT+1000​, are common due to the high nonlinearity of the function, especially for high values of TTT. Here are some strategies that might help improve convergence:

1. Scaling the Heat Flux Expression

Since your denominator is on the order of 10−510^{-5}10−5, this makes the flux very large, leading to steep gradients and numerical instability. Try normalizing or scaling the function by a larger constant to reduce the overall magnitude of Q(T)Q(T)Q(T) and see if this helps in achieving convergence. After the problem converges, you can scale the solution back to get the actual heat flux.

2. Starting with Simpler Boundary Conditions (Continuation Method)

Start with a simpler form of your boundary condition and gradually introduce the nonlinear terms. For example:

3. Use a Transient Study with Initial Temperature Guess

In cases with strong nonlinear dependencies on TTT, a transient solution can sometimes help. This is especially true if the system has large gradients or if the nonlinear boundary condition is very sensitive to TTT. Try running a transient study with a small time step and a reasonable initial guess for the temperature distribution.

4. Adding Smoothing Functions

If you observe oscillations or sharp gradients in the boundary heat flux, consider adding a smoothstep function (or another smoothed approximation) to the heat flux expression. Smoothstep functions prevent the solver from having to handle abrupt changes that can disrupt convergence. In COMSOL, you can add smoothing functions like flc2hs(T, smoothing factor) to smooth out temperature gradients.

5. Solver Settings Adjustments

Fine-tuning solver settings can often be essential for these cases. Consider these options:

6. Stabilization Techniques

COMSOL provides stabilization methods like Automatic stabilization or Isotropic diffusion. These techniques add a small amount of artificial diffusion to the problem, helping it stabilize. Be careful to keep this effect minimal, as it can alter the results slightly, but it often helps the solver find a converged solution in nonlinear problems.

Applying these strategies should enhance your chances of achieving convergence with your nonlinear boundary heat flux. Let me know if you need more specific guidance on implementing any of these in COMSOL!

This should provide your enough guidance to tackle your problem ;)

Best,

Igor

Yes, walker positions can converge when using an ensemble sampler, such as the affine-invariant MCMC ensemble sampler (e.g., emcee). In this method, a group of "walkers" (or chains) explore the parameter space simultaneously. As the walkers move, they collectively sample the posterior distribution. Over time, if the sampling is efficient and the model parameters are well-behaved, the walker positions converge towards the true posterior distribution. Convergence is typically assessed through diagnostics like the Gelman-Rubin statistic or by visually inspecting trace plots.

Hello Maithra Nagaraju

You can advance the calculations by adding the keyword 'opt=tight'.

Note that this calculation requires high RAM. If you need a server to perform calculations, you can get help from the MolQube team. Their email address:

If no corrections are made to the Jacobian in each iteration of Newton's method, then the search direction will become no longer the descent direction of the residuals of the equation. While this method of not updating the Jacobian exists, it does not work for my system of nonlinear equations.

Hi,

A sequence ( x_n ) that converges to L is said to have order of convergence q ≥ 1 and rate of convergence μ if

lim n → ∞ |x_{n + 1} − L| / |x_n − L|^q = μ.

Best

You should look for contours where the SIF values become stable ( don't change significantly from one contour to the next). If the SIF values stabilize after a certain contour, the results from these contours are generally considered trustworthy.

Note that: The outermost contours (contour 6) often include elements farther from the crack tip, where the stress field might be less accurate. These can sometimes show significant deviation, as observed with the top contour line in your graph. Therefore, they are typically less reliable. Also the innermost contours (Contour 1 or 2) might be too close to the crack tip and can be affected by the singularity, leading to less reliable results. Any way it's common practice to use results from Contour 3 to Contour 5, where the SIF values have stabilized.

you can know more about how to work with XFEM and fracture mechanics and how to interpret the results from this course case studies

Dear Maia

I am deeply grateful for your help. Additionally, I would like to ask if the surface in my input file corresponds to the (211) plane. Also, what considerations should I keep in mind if I substitute CH₄ onto the surface in different adsorption sites like ontop, bridge, and hollow during a vc-relax calculation and also for NEB calculation.

Ok, I understood. I don't have much experience in doping calculations, but I have some suggestion you might try.

How many atoms you have to substitute? Let's say you wanted to dope other atom in Sb sites by 50%. You have 4 Sb atoms in your cell, then you just need to change 2 of them, and you don't need to make a super cell. You only have to describe all the atoms as if there weren't any symmetry, and put the new atoms where you want.

Thank you Dr. Borys Kapochkin for your answers. These are insightful on the subducting plate behaviour. To be honest I'm surprised myself that the question has garnered so much less attention. I was hoping someone might add something on slab rollback initiation.

try this input and run it different in full path name :

! B3LYP 6-311G NormalPrint Grid3 FinalGrid5 NOSOSCF

%pal nprocs 32 end

%scf

MaxIter 99999

end

%output

print[P_AtCharges_M] 1

print[P_AtCharges_L] 1

end #output

* xyz 0 2

O 6.52685 -0.73958 -0.60612

O 6.00262 -1.67364 1.36686

N 0.02656 1.02768 -0.14143

N -1.96580 -1.10331 0.21717

N -2.14235 2.92651 0.10135

N -4.01394 0.76086 0.44309

C 2.28855 0.42545 -0.42274

C 2.19580 1.88101 -0.46468

C 0.91718 -0.07010 -0.22669

C 3.54463 -0.36382 -0.56170

C 0.78607 2.20981 -0.28425

C 3.32043 2.84128 -0.65602

C 4.48542 -0.06213 0.59139

C 0.28815 3.45265 -0.25755

C -2.92922 -2.05161 0.38431

C -1.13332 3.80888 -0.07277

C 5.72390 -0.88460 0.47560

C -1.61130 5.10437 -0.05281

C -4.22648 -1.77583 0.55990

C -2.97758 5.01700 0.14249

C -3.26625 3.66749 0.23242

C -4.78734 -0.41009 0.59448

C -6.20792 -0.04567 0.78294

C -0.79409 6.34344 -0.21226

C -4.93426 1.82287 0.53553

C -6.28304 1.28839 0.74599

C -4.62309 3.12217 0.44428

C -7.40736 -0.93939 0.98445

C -3.96952 6.10751 0.24266

C -7.53241 2.09796 0.89398

C -3.64660 7.39609 0.15056

C 7.68363 -1.46287 -0.77870

C 0.43867 -1.47026 -0.11628

C -0.83713 -1.81563 0.07726

C 1.16544 -2.77232 -0.18274

C 0.23678 -3.78127 -0.01692

C -1.04183 -3.13945 0.14773

C -2.38149 -3.34113 0.34626

C -3.08246 -4.64855 0.48648

N 0.56372 -5.09275 -0.02589

C 1.86711 -5.45840 -0.20700

C 2.87912 -4.46116 -0.38749

C 2.50060 -3.10065 -0.37134

C 4.31844 -4.78149 -0.59510

O 4.74862 -6.06913 -0.61164

C 6.06101 -6.43909 -0.81471

C 6.43157 -6.29546 -2.28009

O 5.12041 -3.86953 -0.75075

C 2.17416 -6.91743 -0.21020

C -7.08726 -2.42226 0.99213

*

Ricardo Tadeu Maia Thank you a lot. Your suggestions make the simulation converge. I will use this parameter to simulate an heterojunction. Which parameter do you suggest modifying to create a faster simulation?

It seems that your mesh is very big. Reduce the mesh and try it please.

Dear Yoshinori,

I apologize for the delay in responding to your generous comments on my previously published articles. I did not have enough time until now. I answer you now your interesting question.

Saari and Simon's criticisms focus on the adjustment processes known as Walrasian tâtonnement, where 1) there is no exchange, production, or consumption activities during disequilibrium, and 2) prices are adjusted according to the law of supply and demand or Newton's iterative process, explained as if they result from perfectly competitive mechanisms where a Walrasian auctioneer sets the prices. In these cases, the stability results depend on the shape of the excess market demand functions for each commodity. According to the SMD theorem, this presents a significant problem and a huge limitation for such stability results.

In contrast, the Hahn and Fisher adjustment processes are different: 1) they are not Walrasian tâtonnement processes (Hahn's process involves exchanges, and Fisher includes also consumption and production activities), and 2) firms or individuals set prices through more or less monopolistic competitive processes. Most importantly, their stability results do not depend on the shape of the excess market demand functions for each commodity. Therefore, they are not constrained by the SMD theorem.

It is a shame these adjustment processes are not as well known or taught in our economics courses.

Best regards,

Martin

It is generally accepted that the RMSD fluctuation ranging between 1-3 Angstrom is normal and accepted. So, for most of the simulations, this criterion was met.

Building H-atoms in tleap does not have anything to do with it.

Moreover, what atoms were used for RMSD calculation? Ca or backbone or whole heavy atoms?

Hi

I think, You have manually raised the convergence tolerance to ten to the power of minus 6 when it was not needed and the problem probably converged to ten to the power of minus three or four.

best

IoT, AI, and Blockchain technologies, as well as other technologies categorized in the sphere of Industry 4.0/5.0 and ICT are successively changing more and more aspects of business activities conducted by companies, enterprises and commercially functioning financial institutions. In addition, the aforementioned technologies are also revolutionizing the functioning of other types of entities, most notably public institutions. In recent years, the particularly fast-growing technology of artificial intelligence, including mainly generative artificial intelligence, is being implemented into business entities in order to improve a specific type, a specific sphere of business activity. Business entities are rapidly implementing new technologies into their business operations, including web applications, advanced information systems, intelligent chatbots, intelligent agents technology, etc. equipped with generative artificial intelligence technology seeing it as increasing the efficiency of their business operations, streamlining processes, reducing operating costs. Unfortunately, these are not only positive aspects alone, as they may involve job cuts. Besides, the various positive and negative aspects of the development of artificial intelligence technology and its applications are much, much more. I am conducting research on this issue.

I described the key issues of opportunities and threats to the development of artificial intelligence technology in my article below:

Please write what you think in this issue? Do you see rather threats or opportunities associated with the development of artificial intelligence technology?

What is your opinion on this issue?

I invite you to familiarize yourself with the issues described in the article given above and to scientific cooperation in this issue.

I invite you to scientific cooperation in this problematic.

Please write what you think in this problematic?

Those errors are related to the usage of the "restart" option in the opt route. Restart "copy" all the old opt parameters from the chk file, the error is due to a mismatch between the old route and the new one.

To solve those errors just remove the restart option and set geom=check to read the geometry from the chk file but not the opt options.

Also, the optimization in cartesian coordinates it is deprecable as it is really slow compared to general internal coordinates (GIC), then the calcfc keyword slows down the calculation by not bringing enough benefit to justify it, at least if you don't need force constants of the first point.

Other things, is there a real need to set scf convergence values other than the default ones? Why are you not using symmetry constraints? You should check if your system belongs to a point group or not, if the answer is yes you must remove nosymm (especially to obtain reliable frequencies) and symmetrize the molecule prior to the optimization with software like gaussview or avogadro. In addition, for a frequency calculation, it is desirable to set "very tight" as the convergence criterion of the optimization in order to be more sure of converging to a minimum so as not to end up with imaginary frequencies after. Finally, 3000 cycles both for the opt and the scf are overkill, if your structure does not converge after 500 steps you should better check your system and probably try a different level of theory.

You colud try

#p b3lyp/6-311++g(d,

p) scf=(xqc, MaxConventionalCycles=60) geom=check opt=(maxcycles=300, verytight) freq

I realised some loads just don't want to converge no matter how small the initial increment is. This is so strange!

Dear Hesham

Nice question, but maybe for nonlinear time (or response) history analysis you do not need to overcome the convergence failure, at all; in brief the reason is the fact that in these analyses the inaccuracy is under the influence of both time step size and the nonlinearity iterations, and these two sources are mostly independent. Have a look at "Soroushian, A. and Wriggers, P., 2023. Elimination of the stops because of failure of nonlinear solutions in nonlinear seismic time history analysis. Journal of Vibration Engineering & Technologies, 11(6), pp.2831-2849", and if needed be in contact. I need to leave now. Have a very nice day.

Yoon Junho : Hey, have you tried saving an input from the last optimization step and resubmitting the job?

Also, if your initial structure is unsuitable, you have this convergence failure issue. Maybe, you can modify the initial structure and then try it again.

I hope this gives you some hints.

Contact details on this link & key aims

I also met this question, I dont know what to do

Hi Qian Zhao,

In Molpro unrestricted couple-cluster method can be run only with RHF reference wavefunction, UHF won't work. (In Molpro RHF means the restrected-open-shell HF). As a rule of thumb: with larger basis set (avtz, avqz...) use UCCSD(T)-F12b, however for smaller one (avdz) use the UCCSD(T)-F12a method.

I do not understand why you need the keyword for UCCSD(T)-F12: gem_beta=0.9...The standard UCCSD(T)-F12 method works pretty well for most of the cases.

In addition to mentioning multiplicity 3, your basis set for Nickel is incomplete. You used LANL2DZ for Ni but didn't mention the pseudo-potential as well as pseudo=read.

Hamed Ebrahimi thank you, I will try

Aiswarya Chandran Increasing the number of steps probably solve the problem.

Usually, if you increase the K-point mesh the calculation becomes more accurate but becomes more and more expensive. The same is true for cut-off energy. One should test and converge the K-point and cut-off energy with respect to the total energy of the system.

The size of the supercell depends on the user's choice of what kind of cell dimension with which the user wants to start the calculation. In the periodic boundary condition anyway, the periodicity will be maintained according to Bloch's formula. A large supercell with a smaller K-point calculation is equivalent to a small supercell with a large K-point calculation.

There are many possibilities for the errors you have got. I am mentioning a few of them below. Hope it helps.

1) When there are not enough Governing equations to solve for the number of dependent variables you have set.

2) When there are not enough Boundary conditions for the given problem.

3) When some constraint / Physics is missing in the Problem.

4) You might have used an operator that directly implies '0' as the coefficient of dependent variables in one of the Governing equations, thus recheck all the operators and make sure all of them are used correctly.

There are always some things we unconsciously assume ourselves while solving problems and we might have missed giving that information to Simulation software / Comsol. Thus, you might have to look at it again in detail.

If you're encountering convergence issues in the calculation of surface energy for spinel compounds using density functional theory (DFT), here are some parameters and strategies you can adjust to improve convergence:

K-Point Sampling: Ensure that you have a sufficiently dense k-point mesh for Brillouin zone integration, especially for surface calculations. Increasing the k-point sampling density can lead to better convergence.

Plane Wave Cutoff Energy (ECUT): Increase the plane wave cutoff energy to include more basis functions in the expansion of wavefunctions. A higher cutoff energy can improve the accuracy of the calculation and aid convergence.

Electronic Smearing: Implement an appropriate electronic smearing method, such as Fermi-Dirac or Gaussian smearing, to distribute electronic occupations smoothly. This can help reduce numerical noise and improve convergence, particularly for metallic systems.

Pseudopotentials: Ensure that you are using high-quality, well-parametrized pseudopotentials tailored for the elements present in your spinel compounds. Poorly chosen pseudopotentials can lead to convergence issues.

SCF Convergence Criteria: Tighten the convergence criteria for self-consistent field (SCF) iterations, such as the energy change or density change tolerance. This forces the calculation to reach a more precise solution, although it may increase computational cost.

Geometry Optimization: Perform geometry optimization to relax atomic positions and lattice parameters, especially for surface calculations. Relaxed geometries can lead to a more stable electronic structure and improved convergence.

Spin Polarization: Consider the spin polarization effects, particularly if your spinel compounds contain magnetic ions. Properly treating spin polarization can significantly impact convergence behavior.

Symmetry Considerations: Exploit symmetry whenever possible to reduce computational cost and improve convergence. Symmetry operations can simplify the calculation and aid convergence.

DFT Functional: Verify that you are using an appropriate exchange-correlation functional for your system. Some functionals may perform better for specific types of materials or properties.

By adjusting these parameters and employing appropriate computational strategies, you should be able to improve the convergence of your SCF calculations for the surface energy of spinel oxides using DFT.

CFD softwares are based on numerical methods or techniques to predict the fluid behavior for various conditions e.g. LES and RANS turbulence modelling etc. Unlike exact solutions , the numerical methods involve approximations of the governing fluid parameters which cannot be evaluated at once and thus need iterative computational solvers.

During this process several types of errors are introduced while approximating variable property e.g round off errors ( machine precision) , truncation errors depending on the type of numerical scheme used.

However , according to the nature of fluid and it's interaction with surrounding environment , ( in your e.g yplus wall function which is measure of the fluid friction resistance near wall ) the solutions obtained through numerical schemes present a significant source of error which can interpret the fluid behavior in entirely different manner.

Therefore, the solution is often tested by repeating the process using better approximations and schemes with a focus to obtain the exactness of parameter value leading to iterations.

During iteration process , the error can amplify or reduce ( which is indicative of the stability of solution ) depending on boundary conditions used to obtain solution. So, often an error tolerance is introduced as condition in numerical algorithm to make the solution more meaningful and realistic which closely approximates the fluid behavior. In your case wall shear stress is being approximated using wall units in logarithmic boundary layer.

Once that condition is satisfied, the process stops and proceeds further by evaluating the next dependent variable and so on until complete solution is obtained.

This is a convergence issue which can be due to a variety of reasons. My advice is to reformulate your z-matrix (avoid Cartesian coords in possible) so that the starting geometry has the highest possible symmetry, i.e. belongs to a point group with the largest number of symmetry elements (which is probably Th or -worse- D2h). Second, start with a simple basis set (def2svp) and increase its complexity only if needed. Third, try different convergence algorithms (for example a slower quadratic one, qc in Gaussian).

Finally, and maybe most importantly, there is already at least a paper on this topic, dealing exactly [V(OH2)6]3+. Give it a glance:

Also, you can find some experiments regarding your question in the following critical review.

Denis Benasciutti That only applies to steel bars in reinforced concrete; in this type of structure, the shear load at the interface is transmitted through friction, which is somewhat lesser, and through shear studs. Additionally, I need the plates to detach from the concrete when they buckle, so I cannot perform an analysis with rigid node-to-node contact, which forces me to use surface-to-surface or general contact.

The basic idea of ​​this theory is to take the electron density of the ground state ρ (r) as the main variable, and write any other magnitude as a function of it. This theory is based on a variational principle. which requires the total energy to be a single density functional, and that this energy is minimal for the density of the ground state. The best procedure for performing the DFT is Kohn-Sham, they treated the N-body problem using the single-particle Schrõdinger equations called the Kohn-Sham equations. Solving these equations leads normally to the energy E(ρ) and the densityρ (r), of the ground state.The functional E (ρ) contains a non-classical contribution, called the exchange and correlation energy Exc (ρ) and its derivative by relation to ρ (r) which represents the exchange and correlation potential Vxc(ρ). The formalism of DFT is based on the Hohenberg - Kohn theorems. First, Hohenberg and Kohn showed that the total energy of an electron gas in the presence of an external potential, created by the nuclei is a unique functional of electron density ρ (r): E = E[ρ(r) ], Second, they showed that the minimum value of this functional is the exact energy of the ground state and that the density which leads to this energy is the exact density of the ground state. The other properties of the ground state are also functional of this density: E(ρ0) = min E(ρ ). ρ0 : The density of the ground state.

The functional of the total energy of the ground state is written as follows: E[ρ(r)]= F[ρ(r)] + ∫ VXC (r) ρ (r) d3 r. The functional F [ρ (r)] is universal for any system with several electrons. If this functional is known, then it will be relatively easy to use the variational principle. To determine the total energy and electron density of the ground state for a given external potential. Unfortunately, the Hohenberg - Kohn theorem gives no indication of the form of F [ρ (r)].

If an analytic function G is uniformly dense on compact subsets of (-r, r) in the span of its derivatives G^(k) for k > 0, then it is also uniformly dense on C(R), the set of continuous functions on the real line.

To prove this, we can use the Stone-Weierstrass theorem, which states that any continuous function on a compact interval can be uniformly approximated by polynomials. Here's an outline of the proof:

Therefore, we have shown that if an analytic function G is uniformly dense on compact subsets of (-r, r) in the span of its derivatives G^(k) for k > 0, then it is also uniformly dense on C(R), the set of continuous functions on the real line.

This is a known problem in gaussian, that originates from gaussian using two different algorithms for frequency calculations . If the system is too large, gaussian switches to the second algorithm, which is superslow (since it uses disk memory instead of RAM) and often fails.

The solution is to add the following keyword to the input line:

Int=Acc2E=11

which always worked for me (DFT freq for systems of ~100 atoms).

Some people reported also that adding instead:

CPHF(MaxInv=10000)

also solves the problem, though I never used it.

Peter Breuer , 4n+1 and 3n+1 corresponding to n are unique.

The main problem seems not to be due to the algorithm, but to the multireference nature of some transition metal compounds that renders the SCF and the PES rather unreliable. DFT has a lot of problems to describe such systems and that is a physical limitation of the theory.

I'd say that you can't employ DFT on transition metals without some critical thinking, that is a core discussion in the inorganic modelling and you can find references and discussion on it almost everywhere. I'm not sure of what can or can not be done in terms of convergence algorithm, since I believe we should develop more into the multireference methods.

Thank you Ahmad Al Khraisat for your valuable answer.

Hi, this is not a simple question to answer !

But there are some points to focus on to modify the swarm behavior of PSO core algorithm :

- Initialize population using methods to maximize search space coverage (such as latin hypercubes, sequences, ...)

- try different hyper-parameter values (to poderate inertia vs social behavior)

- try different neighborhood topology.

Then there are other evolutions of the core algorithm, but other evolutions as TRIBES, FIPS, ...

you can find more at :

Clerc, M. et J. Kennedy. 2002, "The particle swarm - explosion, stability, and convergence in a multidimensional complex space"

Mendes, R., J. Kennedy et J. Neves. 2004, "The fully informed particle swarm : simpler, maybe better"

Clerc, M. 2003, "TRIBES - Un exemple d’optimisation par essaim particulaire sans paramètre de contrôle"

El Dor, A. 2012, Perfectionnement des algorithmes d’Optimisation par Essaim Particulaire. Application en segmentation d’images et en électronique (study of topologies, in french)

GOOD LUCK !

I have solved this problem. By checking the .dat file, it was found that this problem is due to an error in my script writing material properties while updating the inp file. So after the 7th iteration, I can't extract the strain and Flvel from it.

Hello all,

If hinges really exist, it will be appeared when repeating the analysis with smaller time steps once or by times. For a sample procedure for nonlinear time history analysis, including the repetitions, see the seismic code of New Zealand NZS 1170.5:2004, or some of my recent papers. By the way, non-zero value for the parameter of the HHT is not for this purpose; it is for cancelling extra high frequency oscillation because of discretization in space. OK, have a very nice day, and wonderful future throughout 2024. Bests