From owner-chemistry@ccl.net Sat Jul 13 02:45:01 2024 From: "Marcel Swart marcel.swart=gmail.com" To: CCL Subject: CCL: DFT Popularity Poll 2024 Message-Id: <-55195-240713024414-4245-x/MfsrppDozGCbGMEPqcTQ{}server.ccl.net> X-Original-From: "Marcel Swart" Date: Sat, 13 Jul 2024 02:44:13 -0400 Sent to CCL by: "Marcel Swart" [marcel.swart]^[gmail.com] MN15 has been included 3 times (including last year): https://www.marcelswart.eu/dft-poll/MN15.html but was not popular enough to be retained. And the rules - that state how new functionals to be included can be suggested - can be found here: https://www.marcelswart.eu/dft-poll/rules.html "There is a maximum of 10 additional suggestions for each year, which are added chronologically (after being suggested by mail to M. Swart)." And I would like to quote again John Perdew on what makes a functional popular: (from https://blogs.nature.com/thescepticalchymist/2014/11/five-years-of-polling-the-computational- chemistry-community.html) "There are many reasons why some functionals are received better than others: accuracy, reliability, wide applicability, computational efficiency, well-founded construction, availability in standard codes, reputation of the functional and its authors, historical priority, novelty, and even hype." Speed, robustness for SCF convergence, robustness for geometry optimizations, general applicability are probably four ingredients which play a major role in the perception of popularity. MS On 12 Jul 2024, at 23:22, Robert Topper rqtopper-#-gmail.com wrote: I just filled out the form. I was sort of surprised because I felt that it focused on certain methods and excluded a number of others. Specifically, I don't understand why MN15 wasn't included, and the same for B3LYP-D3 (as opposed to B3LYP-D, which is there). I'm sure there are others missing that I didn't notice. In the future, I suggest that it would be good for the poll to give users the ability to add their own choices to the poll quickly and easily. All best wishes, Robert Q. Topper http://engfac.cooper.edu/topper From owner-chemistry@ccl.net Sat Jul 13 07:05:00 2024 From: "Thomas Manz thomasamanz]![gmail.com" To: CCL Subject: CCL: DFT Popularity Poll 2024 Message-Id: <-55196-240713003211-21454-jHZUpFao18KRbaRMeZltTA{=}server.ccl.net> X-Original-From: Thomas Manz Content-Type: multipart/alternative; boundary="000000000000200f57061d197a67" Date: Fri, 12 Jul 2024 22:31:49 -0600 MIME-Version: 1.0 Sent to CCL by: Thomas Manz [thomasamanz(!)gmail.com] --000000000000200f57061d197a67 Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: quoted-printable Hi all, There are various perspectives which different people hold. My own preference is to put a high value on convergence robustness as well as accuracy. Thus, according to these criteria, I generally prefer to run CCSD calculations as opposed to CCSD(T) calculations, because perturbation theory (e.g., perturbative triples in in CCSD(T) method) is ill-behaved (undefined? or 'blows up") in degenerate systems (e.g., isolated Al atom). For the same reason, I generally prefer not to use so-called 'double-hybrid' functionals, which rely on perturbation theory. Many people will rank 'double-hybrid' functionals above 'hybrid' functionals, and CCSD(T) above CCSD in accuracy, but I would not do that. The "double-hybrid' functionals and CCSD(T) are theoretically well-motivated if and only if the material has a substantial band gap between the occupied and unoccupied electronic energy levels. In other words, they are ill-behaved for many materials such as metals and many good electrical conductors. So, you can see that even when it comes to benchmarking, there can be differences in interpretations as well as notable misunderstandings in the published literature. Some of the published literature refers to CCSD(T) as 'gold standard' which is totally inappropriate for a method that does not apply to gapless materials. Many of the benchmarking studies incorporate errors of this type, which can make them misleading. Sincerely, Tom On Fri, Jul 12, 2024 at 6:03=E2=80=AFPM Robert Topper rqtopper-#-gmail.com = < owner-chemistry,ccl.net> wrote: > I just filled out the form. I was sort of surprised because I felt that i= t > focused on certain methods and excluded a number of others. Specifically,= I > don't understand why MN15 wasn't included, and the same for B3LYP-D3 (as > opposed to B3LYP-D, which is there). I'm sure there are others missing th= at > I didn't notice. > > In the future, I suggest that it would be good for the poll to give users > the ability to add their own choices to the poll quickly and easily. > > All best wishes, > Robert Q. Topper > http://engfac.cooper.edu/topper > > > On Fri, Jul 12, 2024 at 12:57=E2=80=AFPM Lars Goerigk lars.goerigk+*+ > unimelb.edu.au wrote: > >> Hi all, >> >> >> >> I=E2=80=99m joining this discussion a bit late, but to add to what Stefa= n wrote >> regarding the most popular functionals being also those that perform wel= l >> in benchmark and other rigorous studies: >> >> This is already 5 years old but as part of an account on DFT for >> non-expert users and students we analysed the DFT poll results available= at >> that time and compared them with recommendations based on databases such= as >> GMTKN55. One clearly sees a discrepancy there and I agree that it would = be >> interesting to understand why there=E2=80=99s such a seemingly long time= between >> new methods being made available and the user community picking them up. >> Another great example is the dispersion problem and how long it took to >> convince people that you also need to factor in dispersion in >> thermochemistry, kinetics and geometry optimisations to name a few >> examples. In fact, sometimes you get the impression some still haven=E2= =80=99t >> received that message. >> >> >> >> Anyway, coming back to my main point, if anyone is interested in the >> DFT-poll related part, the paper is freely accessible here: >> https://doi.org/10.1071/CH19023 >> >> >> >> Cheers, >> >> Lars >> >> >> >> *Associate Professor Lars Goerigk *(he/him) >> >> Melbourne Centre for Theoretical & Computational Chemistry >> >> School of Chemistry | The University of Melbourne >> >> Victoria 3010 | Australia >> >> *Phone:* +61 3 834 46784 <+61383446784> >> >> *Website: *http://goerigk.chemistry.unimelb.edu.au >> >> *Follow me on X: *https://twitter.com/lgoer_compchem >> >> >> >> >> >> *From: *owner-chemistry+lars.goerigk=3D=3Dunimelb.edu.au~~ccl.net >> on behalf of >> Stefan Grimme grimme-*-thch.uni-bonn.de >> *Date: *Sunday, 9 June 2024 at 6:49 pm >> *To: *Lars Goerigk >> *Subject: *CCL: DFT Popularity Poll 2024 >> >> >> Sent to CCL by: "Stefan Grimme" [grimme:-:thch.uni-bonn.de] >> > I encourage the CCL community to resist the siren song of this poll. >> The entire point of science is that we have to give evidence to support= a >> claim. A popularity poll yields no legitimate scientific information. If= a >> paper came to me and justified its choice in KS-DFT functional based on >> popularity, I would reject it and tell the authors to read a review pape= r. >> >> I agree in part, but I think that this viewpoint is too extreme. Science >> is made by humans and not (yet) by machines and is >> therefore dependent on our experiences, goals, etc., i.e., on >> non-scientifically justifiable influences. The author of a >> benchmark paper on the performance of density functionals is also subjec= t >> to these influences and therefore a review paper >> is also biased like a poularity poll (of course less in the best case). >> One can see this survey as a valid question in the social sciences. Are >> the most popular density functionals also those >> that perform well in benchmarks? The answer tells us also something abou= t >> our scientific and publication system. The >> question of whether there is a "wisdom of the crowd" is certainly >> relevant and, in my opinion, important for the progress >> of science. For example why does it usually take a long time for new >> theories or methods to establish themselves on the "market"? >> >> But in any case, hard, reproducible scientific facts should be the first >> choice for making decisions about choosing a density functional >> or any other theoretical approximation, I agree with that without a doub= t >> and we have given criteria, examples and decision trees for this >> problem in DFT in our Angew. paper recently (DOI: 10.1002/anie.202205735= ). >> >> Best >> Stefan >> >> >> >> -=3D This is automatically added to each message by the mailing script = =3D- >> >> https://url.au.m.mimecastprotect.com/s/ix7pCgZolKFlOD5lpfNFF-6?domain=3D= ccl.net >> >> >> https://url.au.m.mimecastprotect.com/s/ix7pCgZolKFlOD5lpfNFF-6?domain=3D= ccl.net >> >> >> https://url.au.m.mimecastprotect.com/s/TWI_CjZroMFjKV1jPsRqLS1?domain=3D= ccl.net >> >> >> Before posting, check wait time at: >> https://url.au.m.mimecastprotect.com/s/8M9wCk8vpKsn8P3n1IQJ8LX?domain=3D= ccl.net >> >> >> Job: >> https://url.au.m.mimecastprotect.com/s/mUcbClxwqLco0jqopHqMXAH?domain=3D= ccl.net >> >> Conferences: >> https://url.au.m.mimecastprotect.com/s/uiX7CmOxr6s5B9L5DHQCsT9?domain=3D= server.ccl.net >> >> >> Search Messages: >> https://url.au.m.mimecastprotect.com/s/VpT0CnxyvXcGAkpGjtEZ8AB?domain=3D= ccl.net >> >> >> https://url.au.m.mimecastprotect.com/s/wWR4CoVzwKhXjQpXZIW36cV?domain=3D= ccl.net >> >> >> RTFI: >> https://url.au.m.mimecastprotect.com/s/-yZuCp8AxKsz17EzKTk_dz0?domain=3D= ccl.net >> >> >> --000000000000200f57061d197a67 Content-Type: text/html; charset="UTF-8" Content-Transfer-Encoding: quoted-printable
Hi all,

There are various perspectives = which different people hold.

My own preference is = to put a high value on convergence robustness as well as accuracy.
Thus, according to these criteria, I generally prefer to run CCSD calcula= tions as opposed to CCSD(T) calculations, because perturbation theory (e.g.= , perturbative triples in in CCSD(T) method) is ill-behaved (undefined? or = 'blows up") in degenerate systems (e.g., isolated=C2=A0Al atom).
For the same reason, I generally prefer not to use so-called '= double-hybrid' functionals, which rely on perturbation theory.

Many people will rank 'double-hybrid' functionals = above 'hybrid' functionals, and CCSD(T) above CCSD in accuracy, but= I would not do that.
The "double-hybrid' functionals an= d CCSD(T) are theoretically well-motivated if and only if the material has = a substantial band gap between the occupied=C2=A0and unoccupied electronic = energy levels.
In other words, they are ill-behaved for many mate= rials such as metals and many good electrical conductors.

So, you can see that even when it comes to benchmarking, there can = be differences in interpretations as well as notable misunderstandings in t= he published literature.
Some of the published literature refers = to CCSD(T) as 'gold standard' which is totally inappropriate for a = method that does not apply to gapless materials.

M= any of the benchmarking studies incorporate errors of this type, which can = make them misleading.

Sincerely,

Tom

On Fri, Jul 12, 2024 at 6:03=E2=80=AFPM Robert Topper rqto= pper-#-gmail.com <owner-chemistry,ccl.net> wrote:
I just filled= out the form. I was sort of surprised because I felt that it focused on ce= rtain methods and excluded=C2=A0a number of others. Specifically, I don'= ;t understand why MN15 wasn't included, and the same for B3LYP-D3 (as o= pposed to B3LYP-D, which is there). I'm sure there are others missing t= hat I didn't notice.=C2=A0

In the future,=C2=A0I=C2= =A0suggest that it would be good for the poll to give users the ability to = add their own choices to the poll quickly and easily.

Al= l best wishes,

On Fri, Jul 12, 2024 at 12:57=E2=80=AFPM Lars Goerigk lars= .goerigk+*+unimelb.edu.= au <o= wner-chemistry+*+ccl.net> wrote:

Hi all,

=C2=A0

I=E2=80=99m joining this discussion a bit late= , but to add to what Stefan wrote regarding the most popular functionals be= ing also those that perform well in benchmark and other rigorous studies:

This is already 5 years old but as part of an = account on DFT for non-expert users and students we analysed the DFT poll r= esults available at that time and compared them with recommendations based on databases such as GMTKN55. One clearly sees a discrepancy there and I a= gree that it would be interesting to understand why there=E2=80=99s such a = seemingly long time between new methods being made available and the user c= ommunity picking them up. Another great example is the dispersion problem and how long it took to convince people = that you also need to factor in dispersion in thermochemistry, kinetics and= geometry optimisations to name a few examples. In fact, sometimes you get = the impression some still haven=E2=80=99t received that message.

=C2=A0

Anyway, coming back to my main point, if anyon= e is interested in the DFT-poll related part, the paper is freely accessibl= e here: https://doi.org/10.1071/CH19023=

=C2=A0

Cheers,

Lars

=C2=A0

Associate Professor Lars Goerigk=C2=A0(he/him)

Melbourne Centre for Theoretical & Computational Chemistry=

School of Chemistry | The University=C2=A0of Melbourne<= /span>

Victoria 3010 | Australia

Phone:=C2=A0+61 3 834 46784

Website:=C2=A0http://goerigk.chemistry.unimelb.edu.au<= /a>

Follow me on X:=C2=A0https://= twitter.com/lgoer_compchem

=C2=A0

=C2=A0

From: owner-chemistry+lars.goerigk=3D=3Dunimelb.edu.au~~ccl.net <owner-chemistry+lars.goerigk=3D=3Dunimelb.edu.au~~ccl.net> on behalf of Stefan Grimme grimme-*-thch.uni-bonn.de <= owner-chemistry~~ccl.net&g= t;
Date: Sunday, 9 June 2024 at 6:49 pm
To: Lars Goerigk <lars.goerigk~~unimelb.edu.au>
Subject: CCL: DFT Popularity Poll 2024


Sent to CCL by: "Stefan=C2=A0 Grimme" [grimme:-:thch.uni-bonn.de]
> I encourage the CCL community to resist the siren song of this poll. T= he=C2=A0 entire point of science is that we have to give evidence to suppor= t a claim. A popularity poll yields no legitimate scientific information. I= f a paper came to me and justified its choice in KS-DFT functional based on popularity, I would reject it and tell the a= uthors to read a review paper.

I agree in part, but I think that this viewpoint is too extreme. Science is= made by humans and not (yet) by machines and is=C2=A0
therefore dependent on our experiences, goals, etc., i.e., on non-scientifi= cally justifiable influences. The author of a
benchmark paper on the performance of density functionals is also subject t= o these influences and therefore a review paper
is also biased like a poularity poll (of course less in the best case).
One can see this survey as a valid question in the social sciences. Are the= most popular density functionals also those
that perform well in benchmarks? The answer tells us also something about o= ur scientific and publication system. The
question of whether there is a "wisdom of the crowd" is certainly= relevant and, in my opinion, important for the progress
of science. For example why does it usually take a long time for new theori= es or methods to establish themselves on the "market"?

But in any case, hard, reproducible scientific facts should be the first ch= oice for making decisions about choosing a density functional
or any other theoretical approximation, I agree with that without a doubt a= nd we have given criteria, examples and decision trees for this
problem in DFT in our Angew. paper recently (DOI: 10.1002/anie.202205735).<= br>
Best
Stefan



-=3D This is automatically added to each message by the mailing script =3D-= =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 https://url.au.m.mimecastprotect.com/s/ix7pCgZolKFlOD5lpfNFF-6?domain=3Dccl= .net
=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 https://url.au.m.mimecastprotect.com/s/ix7pCgZolKFlOD5lpfNFF-6?domain=3Dccl= .net
=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 https://url.au.m.mimecastprotect.com/s/TWI_CjZroMFjKV1jPsRqLS1?domain=3Dccl= .net

Before posting, check wait time at: https://url.au.m.mimecastprotect.com/s/8M9wCk8vpKsn8P3n1IQJ8LX?domain=3Dccl= .net

Job: https://url.au.m.mimecastprotect.com/s/mUcbClxwqLco0jqopHqMXAH?domain=3Dccl= .net
Conferences: https://url.au.m.mimecastprotect.com/s/uiX7CmOxr6s5B9L5DHQCsT9?domain=3Dser= ver.ccl.net

Search Messages: https://url.au.m.mimecastprotect.com/s/VpT0CnxyvXcGAkpGjtEZ8AB?domain=3Dccl= .net
=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 https://url.au.m.mimecastprotect.com/s/wWR4CoVzwKhXjQpXZIW36cV?domain=3Dccl= .net

RTFI: https://url.au.m.mimecastprotect.com/s/-yZuCp8AxKsz17EzKTk_dz0?domain=3Dccl= .net

--000000000000200f57061d197a67--