How can we encourage (more) people to use the forum?

LAMMPS
1

Hi Everybody!

I’ve mentioned this before, but we keep having a declining activity in the forum and I am still puzzled over why this is the case. I cannot imagine that people rather suddenly have fewer questions or that they all suddenly read the documentation more carefully or that the documentation has become so much better or are that people are better trained by their advisers.

I also have the impression that there are more frequently no responses on any answers given or no additional information provided on followup questions. I used to get annoyed when people respond with “thank you” without stating whether their problem was solved or not, but now even that is less frequent and I am missing even that.

I also cannot believe that LAMMPS has become less popular. The number of citations on the LAMMPS paper is still staggering and the LAMMPS developers are struggling to keep up with the adaptation and integration of contributed code. … and that is even though there are now multiple projects with popular machine learning models that are maintained outside of the LAMMPS distribution as add-on packages.

So I am asking you, the (remaining?) LAMMPS forum participants to give me some hints to understand what might be going on and what we need to do differently to encourage more participation and what would make the forum more attractive? Or what are the alternative sources of support that people are using that I don’t know about? Or is there something we are missing out on that people need to know so they will use the forum (more)?

I am speculating that it may be a generational thing, since the three most senior LAMMPS developers are either retired or close to retirement age, so may need more “fresh blood” and specifically people exploring more modern ways (which may those be) to communicate. Or is this more of a personality thing and we should identify somebody else to be the primary caretaker of the forum and I should stop participating unless specifically notified?

The forum - and before that the mailing list - has always been some indicator for what people are struggling with and thus helped to identify where we need to improve the code or its tools. As a freely available open source software, there is always the problem of not having much indication other of what is needed or what people are happy with.

LAMMPS compilation failed when I want to use metatensor-lammps
2

This is the old problem of asking people who are still here, why people are not here. :slight_smile:

So as someone who follows the forum and even answers (and asks) the questions now and then, I can only offer my general impressions which may or may not be correct. However, it’s a point for starting a discussion.

First, I suspect that the guidelines post ( Please Read This First: Guidelines and Suggestions for posting LAMMPS questions ) may discourage some people from posting. It is very long and requires some time to read it. Especially for people whose first language is not English, it may be an additional effort, compared to just asking some LLM for an advice (I don’t use them anymore myself, so I have not idea whether they got better at LAMMPS - they were pretty mediocre before).

The guidelines require that the poster include a lot of info and some people may be leery of sharing their input script, either because they fear of someone “stealing” their research or that they will be criticized about its quality. Again, sharing the files with LLM may feel more safe. LLM won’t judge and most people don’t know that everything they share with it may be used in training the next generation (except some cases where the university or employer signed NDA with the LLM company).

Don’t get me wrong, I think the guidelines post was necessary and it did improve the questions that are asked, but it may also discourage some people from posting.

Second, the LLMs. As I said, I don’t use them anymore, so I can only suspect that they got better at answering LAMMPS questions. And even if not, I believe they may still confidently give wrong answers.

I think these may be two most important factors for the decline, but I’d like to be proven wrong as I don’t think there are any ways of fixing these two issues.

1 Like
3

I am speculating that it may be a generational thing, since the three most senior LAMMPS developers are either retired or close to retirement age, so may need more “fresh blood” and specifically people exploring more modern ways (which may those be) to communicate.

Indeed, maybe it is a new generation thing (they dont interact as much). People now a days communicate via Instagram, Snapchat and TikTok dancing. Maybe if we tried that… short videos of us dancing with bunny ear filters with a smile in our faces while holding A4 paper with the solution to the questions written using the least amount of words possible… #validsuggestion

No, but seriously, jokes aside now - I still kind of think it is just a combination of people using AI a lot and not caring so much about the results. The AI might give wrong results but the vast majority of people dont really care and only find out when the supervisors or someone else come check on them, case in which they usually also have the answer for what the problem is.

1 Like
4

I suspect something similar. Based on my observation of many years of participating in mailing lists and forums, this seems to be a case of “damned if you do, damned if you don’t”. From the number of times, people - especially those posting here for the first time - need to be reminded of this post, it looks like it may be deterring the wrong people (e.g. those that are not very confident in what they do, but that could be easily helped) and it gets ignored by those it is primarily aimed at (and thus partially fails its purpose).

This kind of paranoia seems to be particularly present in the field of MD simulations and (possibly even more so) in quantum chemical calculations. I find this behavior rather confusing and inconsistent because when you post something in a public forum, how can you expect any significant assistance without sharing what you have? I know from talking to people working in other fields it is different there. But also, in the age of social media, people share more about anything they do with the general public (and big corporations) than people of my generation ever felt comfortable with. In my experience this paranoia is mostly unfounded: most research projects these days require a substantial amount of studying materials and understanding and evaluating previous research and just copying some input will not help anybody to scoop you.
The probability that somebody is working on exactly the same topic at the same time that would benefit from having those inputs are just astronomically small. But being vague about simulation inputs however is something that has been (too!) common in our field and needs to be eliminated as it leads to non-reproducible results.

I understand that it is difficult for some people to accept criticism, especially if they already had a successful career up to the point of working in simulations. However, this is ignoring that the criticism is not aimed at them personally to give them a deliberate dressing down in public, but to help them improve their work. Avoiding this criticism is not making the work any better and can hurt people more in the long term. Given that we have a very international community, there is probably also the cultural component a factor, in that some cultures see public criticism as a personal attack while others are more relaxed about it. I am old fashioned and I do believe in “tough love” and “learning the hard way”. It has served me well being subject to it despite the occasional frustration. I have learned that having a high tolerance for frustration is a prerequisite for working in science, since you usually struggle with things you don’t understand and once you understand them, things get easy until you reach the next point you don’t understand and you get stuck again. By simple statistical considerations that would mean that you spend most of your time struggling to understand and that is something one has to embrace or look for a different, more predictable line of work.

5

I am now thinking that if LLM based AIs would be good enough to solve research problems of grad students and postdocs (I don’t believe they are, because they are programmed to understand language (and that would include programming languages) and not to understand physics) wouldn’t it be just logical that PIs would rather employ an AI expert and have it herd a pool of AI agents to do their research instead of hiring and training humans. The agents would not want promotions and move on to have a career, and they won’t easily forget what they are trained on.

I am regularly experimenting with AI tools and my experience goes from “pretty decent” to “disastrous”. And it largely depends on what I ask to do and what kind of data is available in the training set and the specific instructions. In the “decent” category are tasks to document code and to execute rather simple tasks where I am just too lazy to look it all up and figure it out myself and mostly all kinds of “busy-work”, i.e. stuff that requires not much thinking, but just following patterns.

In no cases were results without flaws, but the time required to correct them was much less than having to do the work from scratch. Also, often the documentation tasks produced incorrect results because of outdated or incorrect comments or embedded documentation strings in the source code, so in those cases the corrections also needed to be applied to the source and thus made things overall better. Since LAMMPS-GUI uses Qt and because Qt is one of the most popular toolkits, I was successful to let the AI coding bot implement some additions to the GUI (e.g. add two buttons for flipping an image horizontally and vertically).

The bad results happen when I ask for something more complex, for example debugging some unexpected behavior of the LAMMPS code. Many times the “fix” was just to disable or remove the broken functionality. In a few cases, at least that could help identifying an area of the code I had not yet looked at as a possible cause.

We also use GitHub Copilot a lot to reduce the workload for reviewing contributions to LAMMPS. It is pretty good at spotting inconsistencies (especially after we added some specific instructions for the bot to look at), but it regularly makes wrong claims, too. In one case it kept repeatedly telling me that I was using a function incorrectly that I had written myself. I can only assume it was trained on a much older version of the source code where this function did not exist. I have come across multiple cases where it was incorrectly assuming that some pattern would carry on across multiple different models, but could not spot that there were models that could only be used for 2d systems and thus had limited applications and options compared to similar models for 3d systems.

6

My 2 cents join what has been said with regard to LLM use and forum disinvestment by people.

Concerning IA agents and LLM I’m very reluctant to use them, and there is some pushback in french academia toward their general deployment for several reasons, but we know we are few. Students use them a lot, many people use them in the general public, and I don’t know their use abroad in academia but I suspect it is a lot.

I don’t think the sharing of script is a thing. I’ve seen it in MD but it was more about publishing libraries, method and codes, which is still a big problem in the scientific literature, than scripts, especially non-working ones.

Moreover, for a lot of the new generation, the internet is limited to a handful of sites which often are some social media (less and less Facebook, more and more TikTok, sometime Instagram), Google and ChatGPT (which is often used as a search engine nowadays). As I started giving internet website classes here, I can only see how low the bar is concerning the understanding of how internet works for non-CS students, even with regard to basic things such as “what a web page is”. So I don’t really know if they know about forums anymore.

In any case, apart from spreading the word about it offline, I don’t really know how to promote forum activity online. I am trying to develop LAMMPS related activities here, were I finally landed for permanent position, but this is still work in progress. Maybe some of our student will come and ask questions soon. :neutral_face:

7

So it is a good thing we don’t have to use dial-up networking and BBSes anymore :wink:

1 Like
8

Hello, just a Ph.D student from South Korea.

I would like to share a few thoughts regarding your concern.

Recently, LLMs (Large Language Models) have become good enough to generate functional input scripts for simple simulations. While they still struggle with complex setups, their ability to provide instant, “good enough” answers is a major draw. Compared to waiting minutes, hours, or even days for a forum response, many beginners naturally gravitate toward AI for quick troubleshooting.

From my experience talking to undergraduate users(not more than 10, just a few users) in Korea, many find the forum intimidating. The posting guidelines, combined with the language barrier and the generally formal atmosphere, make new users hesitant to post. I also needed quite a bit of courage to write my first post here. While these guidelines are essential for maintaining the quality of discussions and helping developers troubleshoot effectively, they can inadvertently act as a “wall” for beginners.

I do not mean that you should remove the guidelines, but to bridge the gap between guidelines and user accessibility, perhaps the forum could implement an automatic posting template or a generator. When a user starts a new topic, a pre-filled format could prompt them for essential details such as:
LAMMPS Version:

Operating System / Hardware Specs:

Installed Packages (e.g., Kokkos, GPU):
This would guide beginners to follow the rules naturally and help experts provide faster, more accurate answers without repeatedly asking for missing information.

I’ve noticed that although the link to the forum is in the LAMMPS main page, it is easy to overlook. To encourage more traffic and engagement, perhaps a more prominent banner or a one-time pop-up for new visitors could help guide them here instead of relying solely on external search engines or AI.

Interestingly, while searching for help with the Kokkos package, I discovered a very active, independent LAMMPS community in China. It seems there is a huge demand for localized support. Unfortunately, we don’t have a similar dedicated platform in South Korea yet, which makes this global forum even more important for us.

I believe balancing the high standards of this forum with a more “inviting” atmosphere—and perhaps better visibility—will help ensure that the next generation of MD researchers continues to contribute here.

8 Likes
9

Thanks for sharing your thoughts.

Many years back, somebody set up an IRC channel for informal LAMMPS communication, but that faltered quickly due to lack of participation. These days people would probably rather use Slack or Whatsapp or Reddit?. There is a Slack we use during LAMMPS workshops, but we have found that the LAMMPS developers have not enough time available to maintain this as a permanent service. And if you want LAMMPS developers to participate, we would run into the same situation (delays before responses and requests to provide required information), so there is no benefit for us to set this up.

Thus, it would be better if a (group!) of LAMMPS users would self-organize and set up some kind of “LAMMPS user community chat” where LAMMPS users could talk more freely about their worries and concerns, and then the more experienced people in that chat could give advice when a topic needs to be moved up to the forum and how to create a successful post in the forum.

This idea has come up before, but here we are running into a limitation of being part of a larger community (so we don’t have to maintain the forum servers and software) and any template feature in the forum software would have to apply to all categories.

This is one of the problems that I see. We have extremely few people that provide responses. Somehow, everybody seems to be waiting for somebody more senior to provide a better answer than they could provide. I think this is a missed opportunity, because helping others has been a way for me, to learn much more about software (not just LAMMPS) than I could have learned on my own. For as long as you formulate your responses sufficiently careful, i.e. don’t say “you must do xxx”, but rather say “perhaps using yyy could solve your problem” you won’t embarrass yourself, because you were just making a suggestion. This also helps to learn how to effectively construct test cases to test out functionality.

So I think if more people would make an effort to respond, it may also encourage more people to post (the worst questions are the ones you don’t ask).

But then again, there is also the problem that we have (too?) many posts by people that ask questions that they should ask their adviser (but don’t dare to).

2 Likes
10

I like this idea. From what I can tell by poking around, MatSci appears to be based on Discourse which does have a real-time chat feature (https://discourse.org/features). That would reduce the friction between moving from chat to forum post because the account used for both is the same.

1 Like
11

You have to ask the MatSci.org admins about how to access or enable this feature.

It probably helps when I stay away from it, so people don’t have to worry about me correcting them. :water_pistol:

12

I have been thinking about what MD and LAMMPS beginners need most when they first join this community.

In my experience, while the official documentation is excellent, what a beginner truly wants is a practical and diverse set of example input scripts. While LAMMPS already provides a variety of standard tutorials, I believe the forum would become even more attractive to newcomers if we had a dedicated subcategory for sharing specific simulation setups (e.g., “Share Your Input Scripts”).

I am not suggesting that researchers should upload their entire, proprietary project files. Instead, I envision a space where users can share “templates” or “mini-setups” showing how they handled a specific physical process.

For instance, I recently searched for how to perform a uniaxial tensile test. While many use fix deform for bulk materials, others might need to use fix move linear for finite-sized or non-periodic structures. A simple, rough template like the one below could be incredibly helpful for discussion:

units metal
atom_style atomic
boundary s s s

read_data Data-name.data 

region fixed_region block x_lo x_hi y_lo y_hi z_lo z_hi units box
region pulling_region block x_lo_2 x_hi_2 y_lo_2 y_hi_2 z_lo_2 z_hi_2 units box

group fixed_region region fixed_region
group pulling_region region pulling_region 
group stress_region subtract all fixed_region pulling_region

# ... (rest of the setup)

variable v_pull equal 1

fix 1 fixed_region move linear 0 0 0 
fix 2 pulling_region move linear 0 -${v_pull} 0 
fix 3 rest nvt temp 300 300 0.05

run 400000

By having such a subcategory, users could discuss whether a specific approach (like the one above) is physically reasonable for their specific case.

For this to be successful, it would be great if current forum users could voluntarily upload one or two examples they have developed—regardless of how simple the simulation might seem.

Each post would include a clear disclaimer that these are illustrative templates and that users must independently verify the physical parameters for their own specific research environment.

I think this “open-source” spirit of sharing simple templates would make the forum more inviting.

3 Likes
13

I would like share some insights into how LAMMPS users are learning and connecting in China. Over the years, a variety of channels have emerged that may be of interest to the international community.

First, many experienced LAMMPS users in China actively share technical content on WeChat, a multi-purpose messaging app. Similar to how blogs were used in the past, users now publish posts on topics such as creating data file, and other advanced tips—all within the mobile ecosystem. I personally run a WeChat account with over 14,000 followers, where I periodically post tutorials and share practical advice on using LAMMPS.

Second, group chats on instant messaging platforms are widely used for peer support. Experienced users often create public groups where anyone interested in LAMMPS can join and ask questions. Basic inquiries usually receive quick responses, and some groups have reached up to 3,000 members.

Third, video-sharing platforms have become increasingly popular for LAMMPS education. The most active platform in China is Bilibili (similar to YouTube), where users upload tutorial videos ranging from beginner introductions to more specialized topics. I have also contributed by posting introductory LAMMPS tutorials there.

Fourth, there are now several Chinese-language textbooks and handbooks on LAMMPS that are widely used among students. These materials systematically cover installation, usage, and common simulation workflows, helping to lower the barrier for beginners.

Finally, commercial training services have also entered the space. A number of companies in China now offer molecular dynamics simulation training, including short courses and one-on-one mentoring, which further supports skill development in both academic and industrial settings.

I hope this overview is helpful and provides some useful context for the community.

4 Likes
14

Enabled the chat for LAMMPS - if you find that it’s not working out (spammy, too high volume posts, etc.) just let me know and I’ll disable it again

2 Likes
15

Thanks @Aaron_Kaplan. Much appreciated.
I see the “chat” icon next to the “search” icon after reloading the web page.

Could you please do us another favor and - following the suggestion by @FTLMD - also add a new subcategory called “LAMMPS Input Examples” with the description “For sharing and discussing LAMMPS input file examples and templates”?

16

Thanks for sharing this information. It is very instructive.

I am wondering, though, how much of this is driven by the language barrier. It is much easier to learn and talk about what you learn in your native language.

And furthermore, could it be that people are more likely to share information and help others, if it is limited to people from the same country and culture instead of the whole world?

17

Sure, done

1 Like
18

I think there may be a misunderstanding here.
I don’t see a new forum sub-category but rather another chat channel.

19

My bad, the channel should be there now

1 Like
20

@Aaron_Kaplan thanks again. I see it and have adjusted the sticky description post.
It seems still set to be read-only, though, so I could not post a first example to get it started.