2026-06-18

2026-06-18
James Kermode
Approved

Meeting #3 — Thursday 18 June 2026, 14:00–16:00 CEST. Hybrid — Mainz (CECAM workshop) & Microsoft Teams.

1. Attendance

Present (in person, Mainz)

  • Adam Jackson (AJJ)
  • Alin-Marin Elena (AE)
  • Andrew Peterson (AP)
  • Ask Hjorth Larsen (AHL)
  • Elena Gelžinytė (EG)
  • James Kermode (JK) — Chair
  • Yuji Ikeda (YI)

Present (online)

  • Bjørk Hammer (BH)
  • Gábor Csányi (GC)
  • Hanna Türk (HT)
  • Karsten Wedel Jacobsen (KWJ)
  • Kristian Thygesen (KT)
  • Lars Pastewka (LP)
  • Michael Herbst (MH)

Apologies

  • Andrew Rosen (AR)
  • Michele Ceriotti (MC)
  • Paul Erhart (PE)
  • Xing Wang (XW)

2. Opening and Approval of Previous Minutes

The Chair welcomed attendees and noted the meeting was being recorded solely for the purpose of preparing minutes, which would be checked before publication.

The minutes from Meeting #2 (5 February 2026) had already been approved by email and published publicly. No further comments were raised, and they were recorded as accurate.

3. Matters Arising and Review of Actions

The Chair noted that most of the actions from Meeting #2 (A20–A29) correspond to substantive items on the present agenda and would be addressed there. The action list was reviewed briefly:

  • Completed: A20 (member links for website), A21 (Terms of Reference updated — TSC members are full SC members), A22 (UK CIC investigated — Paper 3), A23 (German Verein investigated — Paper 4), A25 (website work advanced — see below), A28 (PySimHub listing — submitted by AJJ; ASE is now listed at pysimhub.io).
  • Not pursued: A24 (Swiss options) — not investigated; to be closed unless a member with relevant expertise volunteers.

The detailed status of each action is recorded in Appendix A of the agenda paper.

4. Terms of Reference (v3.0)

The committee reviewed the revised Terms of Reference (Paper 1), whose only substantive change from v2 was the explicit clarification that Technical Sub-Committee (TSC) members are full members of the Steering Committee (action A21).

Key points:

  • It was asked whether the Terms of Reference would become legally binding once a legal entity is formed. The Chair confirmed they are intended as an informal internal document, to be published on the community web page. They will serve as a helpful first draft for the articles of association of a future legal entity but will not themselves cover all legal requirements; that document would need to be drawn up separately.
  • The provisions discussed at Meeting #2 were confirmed: five-year Steering Committee terms, two-year Chair term, no fixed term for TSC members while they are active maintainers, staggered initial terms (3/4/5 years), and provisions for departures and for actively seeking replacements to maintain roughly the current committee size.

Consensus: The Terms of Reference (v3.0) were formally adopted. They will be published alongside the meeting minutes on the community web page.

5. Membership Update

EG announced that she wished to withdraw her earlier notice to step down and to remain on the committee. She will continue to attend meetings and provide guidance, while taking a less active role in code organisation and contribution owing to her full-time job.

The committee warmly welcomed this and thanked EG for her substantial contributions, in particular the organisation of the CECAM workshop, confirming that attending meetings, engaging in discussion, and reading papers is entirely sufficient and very much appreciated.

Chair’s note (added after the meeting): Xing Wang (XW) has since notified the Chair that he will stand down from the committee, as he is leaving his academic role. The committee thanks him for his contributions. His departure will be taken into account in the planned membership rotation and in maintaining the backlog of candidates.

6. ASE Foundation

The committee continued its discussion of options for establishing a legal entity to receive and hold funds (Paper 2, with the UK CIC in Paper 3 and the German Verein in Paper 4).

NumFOCUS: The Chair had contacted the new NumFOCUS executive director about a week before the meeting to ask whether they expected to reopen calls for new projects. No response was received. Combined with the standing public position that they are not accepting new projects, NumFOCUS is not considered an option for now. The Chair will forward any reply to the committee list if one arrives.

Swiss option: Not investigated (action A24); to be set aside unless a member is able to take it forward.

UK Community Interest Company (Paper 3): Relatively cheap and easy to set up; directors need not be UK nationals or residents (a UK address is required, which could be institutional or a paid mailbox). The principal drawback is that, without separate charity registration, the entity would be liable to 20% corporation tax on income raised but not spent within the same financial year. Charity status is achievable but requires the additional step of registering a company and then converting it to a charity.

German Verein / e.V. (Paper 4, LP): A lean, democratic non-profit structure (as used by sports clubs and the Fraunhofer society), essentially free to set up. It can be tax-exempt, but — because pure open-source projects have faced difficulties in Germany — the purpose must be framed as providing research infrastructure / supporting science and education. Tax-free status is confirmed via pre-clearance with the tax authority before incorporation (effectively a one-step process). Membership is flexible, but the executive board (≈3 people, equivalent to company directors) must be EU residents; some additional complications arise if a majority of members are non-EU, though an “Ausländerverein” variant may relax this. A lawyer would need to review the statutes.

Discussion:

  • Both options can receive EUR, USD, and (the UK now being back in Horizon Europe) European grant funds; EUR transfers are likely marginally easier for a German entity.
  • The German option was favoured as more sustainable long-term, given that the centre of gravity of ASE activity is in continental Europe. No objections were raised.
  • Funding the setup: an initial €1,000–2,000 is needed before a bank account exists, so it would have to be donated personally in the first instance. GC offered to pledge €5,000 per year for the coming years to kick-start fundraising, and to provide some administrative support from MPI-P; a further initial personal donation was also offered. Ongoing costs (preparing and submitting accounts) are expected to be of order €1,000 per year.
  • The committee agreed the effort is only worthwhile if there is a genuine intention to run a meaningful budget; the modest aspiration of funding developer and occasional community events (a few thousand EUR per year), potentially growing to support maintainer time in future, was considered sufficient justification.

Consensus: The committee agreed to proceed with the German Verein as the most promising option, investigated initially by LP and GC, with the aim of having a legal entity in place during the 2026 calendar year. Members with concerns were invited to send them to the Chair privately.

New actions: A31 (LP/GC). Carried-over: A24 (Swiss, optional).

7. CECAM Workshop 2026 — Retrospective and Next Community Event

The organisers reported on the workshop “Integration into Wider Community Projects” (15–19 June 2026, Mainz).

Retrospective:

  • The format mirrored the successful 2025 event (talks followed by hands-on work), with good size and engagement, including a back-to-back session of DFT code-developer presentations.
  • Compared with 2025 (largely discussion, planning, and conceptual work), 2026 saw much more hands-on implementation, reflecting greater maturity of ideas and the availability of agentic AI for rapid prototyping (e.g. three GUI plugins demonstrated after a few hours).
  • The committee thanked all co-organisers — including HT, who did substantial work but could not attend in person — AE and the other organisers, the MPI-P local support staff, and GC for providing the venue. Funding support from CECAM and DPG was gratefully acknowledged.

Next community event:

  • There was broad agreement that the next in-person event should be smaller and more coding-focused (e.g. a developer retreat of ~10 people on a specific topic) rather than repeating the full talks-plus-coding format, supplemented by short online sprints/hackathons in between.
  • A realistic timeline to releasing ASE 4 is around a year, which informs the choice of focus.
  • Possible venues/funding: GC offered to host again at MPI-P and contribute funding; Psi-k funding becomes available again but is unlikely to support a primarily internal developer event; CECAM might host an alternative-format hackathon (to be explored informally); CCP9 is a possibility but its funding beyond March 2027 is uncertain. Summer 2027 was favoured as a target, keeping cadence with online events in between.

Pre-Mainz community survey (Paper 5):

  • The survey (analysed with LLM assistance — to be read with caution) makes for interesting and occasionally challenging reading, with a positive closing endorsement of ASE’s value noted.
  • Most priority concerns were already known and under discussion; there were no major surprises.
  • Documentation was an even higher concern than expected despite the past year’s work — partly historical perception, but real gaps remain (difficulty finding things; increasing LLM-mediated use suggests a need to write documentation for machines as well as humans). A concrete, actionable step is to enable complete API documentation (auto-doc for everything, even just signatures).
  • On breaking changes / regressions: the importance of clearly communicating ahead of the disruptive ASE 4 release was emphasised. The committee noted the recent focus on higher-quality release notes (with bold “breaking changes” callouts), appreciated by users; that ~70% of survey respondents reported no regression problems; and that a recurring workshop theme is the need for cross-ecosystem CI across the top downstream packages before ASE 4 is pushed out widely.

New actions: A34 (JK — funding/venue for next event).

8. Community Website and Ecosystem Listing

The Chair presented, on PE’s behalf, the draft community website (to live at the ase-lib.org front page, with documentation mirrored at docs.ase-lib.org). PE has produced steering committee, events, and ecosystem pages; the landing page is still to be designed (proposed as a simple page with large links to Community / Documentation / etc.).

Website:

  • Members were asked to review and approve the portrait photos PE has sourced for them, or supply a preferred photo.
  • There were no objections to releasing the site publicly. Some cleanup of the existing docs (removing the duplicated ecosystem page, etc.) will be needed. The site is a Markdown repository under the ASE GitLab group, maintainable by several people; meeting minutes can be published there directly. PE has paid the domain cost up front; once the foundation exists it should reimburse the ~€50/year domain fee (no hosting fee — GitLab Pages).

Ecosystem listing (Paper 6):

  • A reverse-dependency scan (via Google BigQuery on PyPI) found ~1,025 packages depending on ASE as a core dependency — far more than the ~40 currently listed (some of which are stale). Around 100 prominent candidates not currently listed were shortlisted.
  • A sustainable listing must be largely automatic. Suggested freshness/relevance indicators included last-commit date (with a “stale” flag after, e.g., ~2 years), supported Python versions, and download counts. A self-nomination web form was favoured, allowing maintainers to introduce themselves and write a (neutral, non-advertising) blurb, subject to a light criterion that the package genuinely uses ASE data classes internally rather than only at the periphery.
  • Categorisation (PE’s draft uses analysis / workflow / machine-learning potentials) and the relationship to a future plugins database were noted as needing further thought; detailed suggestions to be sent to JK or PE. The exercise is also valuable for identifying the major downstream packages needed for ecosystem CI.

Consensus: Release PE’s website publicly on ase-lib.org and add a self-nomination form for the ecosystem page; treat the wider ecosystem analysis as a separate, longer-term project.

New actions: A32 (all — approve portraits / send suggestions), A33 (PE/JK — release site, self-nomination form, docs cleanup).

9. NVIDIA Collaboration — ALCHEMI Toolkit

Those present in Mainz saw a presentation from NVIDIA; committee members had also joined a technical meeting with them a few weeks earlier.

Discussion:

  • NVIDIA’s main interest is running MD with a tightly coupled calculator on the GPU. It was noted that this is essentially the same long-standing problem as coupling an optimizer with a calculator (e.g. optimisation with VASP) — not a new or special problem, and one ASE should solve in general; no immediate solution was proposed.
  • A presentation on the NVIDIA ALCHEMI toolkit (and “toolkit ops”) and a small ASE↔ALCHEMI interface demo were shown. In a longer conversation, NVIDIA indicated that what they most want from ASE is community leadership: a community-backed standard for passing atoms from CPU to GPU, which they believe others would adopt.
  • The committee saw potential merit if such a standard is vendor-neutral and technically sound, but questioned whether defining it is ASE’s role. The consensus technical view was that GPU/ML computation needs a data standard (an “atoms as arrays on a device” object), with the Atoms object retained for structure manipulation and visualisation — consistent with patterns already seen in ML applications.
  • It was noted that many community MLIPs are accessed primarily through ASE, making this a major area where ASE and GPUs meet; soliciting advice from MLIP developers was suggested.
  • Licensing: NVIDIA are reluctant to contribute directly to (L)GPL code. It was clarified that ASE is LGPL and that importing it imposes essentially no obligations on downstream packages except when ASE is shipped inside a proprietary product. The suggested way forward is a separate BSD-licensed ase-gpu-style repository on which NVIDIA and others could collaborate.
  • An NVIDIA event being hosted at MPI-P on 3–4 September 2026 may offer an opportunity for further in-person discussion.

Consensus: No urgent vendor lock-in concern at present. The Chair will convene a small, representative independent advisory group on MLIPs / GPU data standards and report back at the next meeting.

New actions: A35 (JK — convene advisory group).

10. Recurring Technical Updates

Documentation sprints: The sprints have been excellent (thanks to HT); the next is scheduled for 24 August 2026, with details to follow on the mailing list.

Release: The TSC agreed to make an ASE release at the end of the workshop week, partly symbolic, including a release note covering the ongoing _4 work merged into the codebase and what users can expect. The note should also state that maintainers of I/O formats are recommended to continue using the existing (v3) I/O-format plugin API, which the TSC commits to keep working.

ASE 4 roadmap and API design:

  • An online sprint for API design (analogous to the documentation sprints) was strongly encouraged. Defining the API soon is important so that external parties (e.g. NVIDIA) can build against a stable interface; this depends on first planning the migration strategy so it is clear which parts are safe to use when.
  • AE asked for a timeline. The committee agreed an ordered list of steps (the sequence of events that need to happen) would be valuable even without firm dates, and would help others plan support. AE will draft it. A concern was noted that the current contributor base is somewhat skewed toward ML, and that DFT codes must not be left behind.

Plugins: The plugins architecture was demonstrated at the workshop and is progressing (API not yet stable; merging into _4 has begun). It supports decentralisation — internal components (e.g. optional optimisers, thermostats, thermochemistry) could become independently maintained plugins over time — and is a route to giving commercial-code plugins (e.g. VASP) their own maintainers, once a documented calculator-plugin API exists to ask them to migrate to.

TSC membership (AR): It was noted that AR has not been active as a maintainer. The understanding is that he remains a Steering Committee member but not a TSC member; AHL/the Chair to confirm with him.

New actions: A36 (TSC — make release), A37 (AE — draft ASE 4 timeline), A39 (TSC — confirm AR’s status).

11. Agentic AI in Contributions

Revisiting the topic from Meeting #2, the committee discussed how to handle AI-assisted contributions.

Proposed principle: Update the contribution guideline so that AI-generated code is not banned, but every merge request must be the full responsibility of the (human) submitter, must come from a human account, and must be human-reviewable; the maintainers reserve the right to close excessively large merge requests without review.

Discussion:

  • A guiding principle was that “a pull request should contain human thinking”: if anyone could have produced it by prompting an AI with no added human value, it is usually not worth opening. The TSC may legitimately decline contributions the team would not be able to understand and maintain.
  • AI was recognised as genuinely useful, especially for writing tests, but a human must check them (AI sometimes produces only smoke tests), and oversized contributions should be closed.
  • Authorship: concern was raised that AI should not be recorded as a “co-author”, since (per the Vancouver recommendations) it cannot take responsibility; the human must be the responsible author. Against this, the Git co-author trailer was seen by some as a useful, transparent disclosure that AI was used rather than a genuine authorship claim, while it was cautioned that tools such as Zenodo derive metadata from the Git author list, risking misattribution. The agreed pragmatic position was that contributors should disclose the tool used in the commit message (and make their own commits) rather than rely on machine “co-author” trailers; forcing the issue risks people simply not disclosing AI use at all.
  • agents.md: The committee favoured adding an agents.md / skills file (which can itself be largely AI-written), but noted this depends on first updating the developer guide so the two are consistent and machine-readable.

Consensus: Broad agreement on the principle, but no final wording was settled. AE (with the documentation group) will redraft the contribution statement — drawing on the Leiden declaration on AI — update the developer guide, and subsequently add an agents.md. Much of this can be progressed over the remaining workshop days.

New actions: A38 (AE / Docs group — redraft contribution statement, update dev guide, add agents.md).

12. Any Other Business

Committee membership and renewal:

  • A workshop participant — a long-standing ASE contributor associated with a licence-encumbered electronic-structure code — expressed interest in joining the committee. No commitment was made, but the individual was considered a strong future candidate when the committee is next refreshed. The committee preferred to keep a backlog of candidates for the planned rotation rather than expand immediately; joining via the TSC was noted as a possible route given the candidate’s maintainer background. Representation from a licence-encumbered code was seen as positive, provided it does not create imbalance among codes.
  • The importance of equality, diversity and inclusion in future appointments was raised and endorsed, while cautioning against over-burdening members of under-represented groups.

13. Date of Next Meeting

The next meeting will be held online on Tuesday 1 December 2026, 14:00–16:00 CET (13:00–15:00 UK), to be confirmed by email.

Summary of New Actions

ID Owner Description
A30 James Kermode (JK) Forward any response from NumFOCUS to the committee list; continue to monitor for a future open call
A31 Lars Pastewka (LP) / Gábor Csányi (GC) Progress the German Verein (e.V.): consult a lawyer on statutes and costs; work towards incorporation during 2026 (GC pledge of €5,000/yr noted)
A32 All members Review and approve (or supply a replacement for) your portrait/entry on the draft community website; send ecosystem-listing suggestions to JK or PE. Update: the community website is now live at ase-lib.org (with thanks to Paul Erhart); members are asked to check their entries there.
A33 Paul Erhart (PE) / James Kermode (JK) Release the community website publicly at ase-lib.org with an ecosystem self-nomination form; tidy the docs to remove duplication. Done: the community website is now live at ase-lib.org, with thanks to Paul Erhart.
A34 James Kermode (JK) Investigate funding and venue for a smaller, coding-focused community/developer event in ~12 months (explore CECAM alternative-format and CCP9; GC offered MPI-P)
A35 James Kermode (JK) Convene a small, representative independent advisory group on MLIP / GPU data standards (NVIDIA follow-up) and report at the next meeting
A36 ASE-TSC Make an ASE release at the end of the workshop week, with release notes covering the ASE 4 work and recommending continued use of the existing v3 I/O-format plugin API. Done: ASE v3.29 has been released, with many thanks and congratulations to Ask, Yuji and Adam. The release notes make clear that the _4 features should not yet be relied upon, but the intended recommendation that maintainers continue using the existing v3 I/O-format plugin API was not included in the announcement; AJJ to follow up to have this added to the online release notes and circulated to the mailing list.
A37 Alin-Marin Elena (AE) Draft an ASE 4 roadmap / timeline (ordered list of steps; dates optional)
A38 Alin-Marin Elena (AE) / Docs Group Redraft the contribution statement on AI-assisted contributions (cf. the Leiden declaration); update the developer guide, then add an agents.md file. Update: a draft AI-assisted coding statement is open as merge request !4167 for review.
A39 ASE-TSC Confirm AR’s (Andrew Rosen) maintainer / TSC status. Done: AR has confirmed that he is no longer a TSC member; he remains a Steering Committee member.

Carried forward from Meeting #2: A24 (Swiss foundation option — optional, not yet pursued).

Minutes prepared by James Kermode (Chair), June 2026. Status: Approved.

Updated: