Plug in turbulence - TMX

pitch-370004 · in warm_bubble

Status: Thinking — at the foot of the hill Shaping — halfway up the hill Ready — at the top of the hill In progress — halfway down the hill Done — at the bottom, over the hill Shelved — off the hill Somebody is building this in the cycle it was bet into — an overrun is re-bet, not extended.
Priority: Medium
Appetite: Bet 12 over 3 people, which buys 4.0 weeks · 4.0 in tasks A budget in person-weeks, not an estimate: what this is worth, not how long it takes.
Tags: icon4py, tmx, turbulence, physics-driver, port
Owner: yiluchen1066
Assignees: yiluchen1066, jcanton, havogt
Reviewers: OngChia, jcanton, yiluchen1066, msimberg
Review waived: —
Parent: warm_bubble
Blocked by: —
PRs: C2SM/icon4py#1301, C2SM/icon4py#1359, C2SM/icon4py#1360, C2SM/icon4py#1457, C2SM/icon4py#1458, C2SM/icon4py#1475
Cycle: 37 A bet is for one cycle. Work that overruns is re-bet, not extended by default.
Start date: 14.07.2026
End date: —
Scheduled: 2026-07-13 → 2026-08-07
Blocks: Scientific validation, Distributed driver run
Progress: 0/7 wk

No No-gos section. The pitch template asks for one — it is what keeps the appetite honest.

Progress 0/7 wk

Review tmx port In progress3 wk · jcanton
Integrate tmx into phys driver In progress4 wk · yiluchen1066

Problem

The original approach question: interface with Kokkos/TMX vs port TMX to GT4py - is resolved. @jcanton has ported TMX to GT4py in PR#1359, but it ships as a standalone granule with its own state objects and savepoint tests, it is not wired into anything that runs. Our PhysicsDriver already runs muphys as a Component; Tmx needs the same treatment so we can aim for running warm bubble experiment in icon4py as one of our milestones.

Appetite

Solution

The Physics driver PR#1301 needs to be merged (working the 3rd round of review).
The Tmx port PR#1359 needs to be verified and reviewed.
Wrap the TMX granule as a second Component behind the existing driver protocol, using the same adaper shape muphys uses. New files only inside the tmx package, so re-merges of the still moving PR#1359 stay clean, plus a small opt-in change to the driver
Two static states (metrics + interpolation) for TMX are assembled by pure fetch from the field factories, exactly like DiffusionMetricState/Diffusion. The interpolation coefficients already exist; the seven metrics fields which are not used in other component should be added to the common MetricsFactory as proviers (formulas from mo_vertical_grid.f90), so they are veirified in the common metrics datatest and reusable by future physics.
The data flow will mirror Muphys:
- a TmxState.gather gathers T/p/u/v/air_mass/cv_air from the prognostic.
- a TmxComponent.__call__ wraps the the tmx.run(), which returns tendencies
- a TmxState.scatter updates the tedencies to prognostics states
Testing:
- The Tmx granule itself is already verified independently inside the PR #1359. We will run a full APE_aes with muphys + TMX completes, verify with exclaim_ape_aesPhys_v06

Rabbit holes

TMX interfaces with jsbach for exchanges with land. We are not touching that this cycle. We are implementing a simplified land instead.

No-gos

Progress

Reported at the review of 11.08.2026, in the deck's own words.

Physics driver with Muphys and TMX:

https://github.com/C2SM/icon4py/pull/1301 Merged - verified with v07 data
https://github.com/C2SM/icon4py/pull/1360 To be finished - verifying with v08

tmx port (claude):

~15k lines in 48 hours
tests passing with good data (v08)

The bet was placed at the cycle-37 table and was not re-bet at cycle 38, but the work ran on into it and is still open. @havogt joined during cycle 38 and began carving #1359 into pieces that can be reviewed on their own; @OngChia is reviewing them as they arrive:

https://github.com/C2SM/icon4py/pull/1457 Merged 14.09.2026 - the package, its configuration and the static states
https://github.com/C2SM/icon4py/pull/1458 Open - the prescribed surface fluxes
https://github.com/C2SM/icon4py/pull/1475 Open - the diagnostics component

#1359 itself stays open as the trunk the pieces are cut from, and #1360 waits on it.

## Problem

The original approach question: interface with Kokkos/TMX vs port TMX to GT4py - is resolved. @jcanton has ported TMX to GT4py in [PR#1359](https://github.com/C2SM/icon4py/pull/1359), but it ships as a standalone granule with its own state objects and savepoint tests, it is not wired into anything that runs. Our ```PhysicsDriver``` already runs muphys as a ```Component```; Tmx needs the same treatment so we can aim for running warm bubble experiment in icon4py as one of our milestones.

## Appetite

## Solution

- The Physics driver [PR#1301](https://github.com/C2SM/icon4py/pull/1301) needs to be merged (working the 3rd round of review).

- The Tmx port [PR#1359](https://github.com/C2SM/icon4py/pull/1359) needs to be verified and reviewed.

- Wrap the TMX granule as a second ```Component``` behind the existing driver protocol, using the same adaper shape muphys uses. New files only inside the tmx package, so re-merges of the still moving [PR#1359](https://github.com/C2SM/icon4py/pull/1359) stay clean, plus a small opt-in change to the driver

- Two static states (metrics + interpolation) for TMX are assembled by pure fetch from the field factories, exactly like ```DiffusionMetricState/Diffusion```. The interpolation coefficients already exist; the seven metrics fields which are not used in other component should be added to the common MetricsFactory as proviers (formulas from ```mo_vertical_grid.f90```), so they are veirified in the common metrics datatest and reusable by future physics.

- The data flow will mirror ```Muphys```:
    - a ```TmxState.gather``` gathers T/p/u/v/air_mass/cv_air from the prognostic.
    - a ```TmxComponent.__call__``` wraps the the tmx.run(), which returns tendencies
    - a ```TmxState.scatter``` updates the tedencies to prognostics states

- Testing:
    - The Tmx granule itself is already verified independently inside the PR #1359. We will run a full APE_aes with ```muphys``` + ```TMX``` completes, verify with ```exclaim_ape_aesPhys_v06```

## Rabbit holes

- TMX interfaces with jsbach for exchanges with land. We are not touching that this cycle. We are implementing a [simplified land](https://hackmd.io/i8bToj2ZRg6GFwDdU40UqA) instead.

## No-gos

## Progress

Reported at the review of 11.08.2026, in the deck's own words.

Physics driver with Muphys and TMX:

- https://github.com/C2SM/icon4py/pull/1301 Merged - verified with v07 data
- https://github.com/C2SM/icon4py/pull/1360 To be finished - verifying with v08

tmx port (claude):

- ~15k lines in 48 hours
- tests passing with good data (v08)

The bet was placed at the cycle-37 table and was not re-bet at cycle 38, but the
work ran on into it and is still open. @havogt joined during cycle 38 and began
carving #1359 into pieces that can be reviewed on their own; @OngChia is
reviewing them as they arrive:

- https://github.com/C2SM/icon4py/pull/1457 Merged 14.09.2026 - the package,
  its configuration and the static states
- https://github.com/C2SM/icon4py/pull/1458 Open - the prescribed surface fluxes
- https://github.com/C2SM/icon4py/pull/1475 Open - the diagnostics component

#1359 itself stays open as the trunk the pieces are cut from, and #1360 waits on
it.