Thermal
Thermal FEA predicts temperature, heat flow, and stored thermal energy under prescribed temperature, heat flux, convection, and volumetric heat generation. RunMat separates thermal material ownership from boundary conditions: conductivity, density, and specific heat belong to solid/cell material regions; temperatures and fluxes target compatible regions on the same admitted geometry.
The typed .fea contracts and tetrahedral thermal finite-element assembly are implemented.
Compiled-study execution supports a thermal prefix feeding a same-mesh thermomechanical
preload and supported buckling, prestressed modal, direct harmonic or modal-reduced harmonic
consumer. This is not blanket admission of standalone thermal or arbitrary coupled graphs.
Authored models now preserve explicit thermal reference temperatures, and
thermomechanical preload binds a named retained thermal state to that compiled path.
The authored thermal-to-modal fixture passes actual-CAD execution after pre-execution driver/worker
replacement and cold import from declared objects. Other outer-root consumer/order combinations
and mid-phase restart remain unqualified; the equations below describe implemented assembly,
not a validation claim. See the
thermal-prestress example and
state-transfer contract.
Retained thermal field histories
The thermal pipeline exposes temperature (K), temperature gradient (K/m), heat flux
(W/m²), and thermal_energy_storage_rate_density (W/m³) to the shared history evaluator.
The last quantity is a scalar element field: the volume average of
rho * c_p * dT/dt, stored in thermal.energy_storage_rate snapshots. It is distinct from
heat_rate, whose dimension is watts. A volume integral of storage-rate density has units of
power; an average retains W/m³. Exact requested history times are inserted into the solve schedule.
The insulated uniform-heating reference uses Q = rho * c_p, giving a 1 K/s temperature rise
and storage-rate density equal to Q for accepted time steps, independently of element volume.
Both Tet4 and Tet10 history outputs are checked against that analytic value. This does not qualify
the unfinished public thermal executor.
Native coordination boundary
The native library has a single-step thermal coordinator for steady or backward-Euler Tet4/Tet10 studies. It schedules the solve, shared observation partitions and reductions, retained-history checkpoints/finalization, accuracy evaluation, every declared temperature state, and final publication. Local and remote task-pool entrypoints use the same coordinator. This is task placement, not a distributed thermal linear solve.
Observations consume the published thermal fields; they do not rerun the solve. A final-time field retains its physical time coordinate. A complete history and a final field are separate publication obligations: publishing one frame cannot satisfy a history declaration. Accuracy validation binds the observation catalog to the original thermal workload and mesh.
Typed thermal-prestress completion retains the thermal prefix in local runtime publication and remote CLI conversion. Portable thermal roots authenticate their complete declared object closure; tasks transfer their declared input inventories rather than all historical catalog objects. The authored modal fixture covers actual-CAD cold replay after pre-execution replacement. The full outer-root consumer/order matrix, genuine durable whole-chain qualification, bounded driver imports, checkpoint-prefix lineage for evicted samples and restart remain open gates. Resource reservations are conservative admission bounds, not measured memory-use evidence. Convergence, energy balance, thermoelastic transfer and formulation-specific release evidence remain open.
Authored observations and selective reads
Thermal-prestress runtime publication projects retained non-history observations and finalized
series reductions into the ordinary run. Histories retain their publication catalog and typed
object roots instead of being flattened into run.observation_histories. Discover them with
storage::load_observation_publication(run_id); select a retained coordinate with
analysis_observation_history_sample_by_run_id_op(run_id, observation_id, sample, context).
These are Rust runtime APIs, not new .fea syntax or plotting operations.
ManifestAndChunks reads the history metadata and selected sample partition; it does not read
unrelated history samples or canonical thermal fields. WholeArtifact necessarily reads the
whole history object. Both return the same typed sample, including physical coordinate, units,
source identity and reduction attribution. Missing, altered or unretained samples fail rather
than being interpolated or recomputed. Cold runtime reads use persisted objects, not the worker
session. Initial source authentication and persistence still inspect the complete declared closure;
selective query behavior does not establish bounded admission memory.
What heat equation does RunMat assemble?
For temperature , density , specific heat , conductivity , and volumetric source ,
RunMat supports prescribed temperature on , applied inward heat flux on , and convection on :
Here is the outward unit normal. Positive prescribed flux adds heat to the body; the recovered heat-flux vector is , whose outward normal component is therefore on . Convection uses the outward heat-loss convention.
Tetrahedral shape functions produce conductivity and consistent-capacity operators
Steady analysis solves . The implemented transient assembly uses backward Euler for one step,
Prescribed values are eliminated consistently. A steady connected component needs a temperature anchor or positive Robin/convection weight; otherwise RunMat reports an underconstrained thermal nullspace instead of inventing a reference temperature.
Which thermal procedures are implemented?
| Procedure | Spatial/temporal realization | Required material/state |
|---|---|---|
thermal steady | Tet4/Tet10 Galerkin conduction plus a sparse linear solve | Conductivity and a well-posed reference/Robin boundary |
thermal transient | The same spatial operator with consistent capacity and backward Euler | Conductivity, density, specific heat, time samples, and initial temperature |
Recovery computes solver-mesh temperature gradients and heat-flow/energy-balance evidence; canonical volume and boundary projection are required by the registered mesh-backed branches.
How do I run a steady-state heat-transfer study?
model heated_part {
assign material.aluminum to geometry.part.region.body
constraint sink_temperature {
thermal.temperature {
target = geometry.part.region.heat_sink
value = 293.15 K
}
}
load applied_flux {
thermal.heat_flux {
target = geometry.part.region.hot_face
value = 25000 W/m^2
}
}
}
study steady_temperature {
geometry = geometry.part
mesh = mesh.part
model = model.heated_part
analysis temperature { thermal steady {} }
output temperature = analysis.temperature.field.node.temperature
}The solid material must provide thermal conductivity. Density and specific heat are also required
when the capacity operator participates. thermal.temperature is an essential constraint;
thermal.heat_flux and thermal.convection are natural boundaries.
How do I model convection to ambient air?
load exterior_convection {
thermal.convection {
target = geometry.part.region.exterior_faces
ambient_temperature = 298.15 K
coefficient = 35 W/m^2/K
}
}The coefficient contributes both a positive boundary matrix and the source. Check the actual exposed area and whether one coefficient is defensible across the entire selected region. Natural convection, forced convection, and contact conductance are different models; do not reuse a film coefficient as a contact resistance.
How do I apply internal heat generation?
load die_loss {
heat_generation on geometry.module.region.die = 8000000 W/m^3
}Heat generation targets a volume and carries . Heat flux targets a surface and carries . RunMat projects each source with an integration rule and records scalar conservation evidence; an unmapped source or projection outside tolerance fails.
How do I run transient thermal analysis with a power cycle?
signal power_cycle : heat_flux over time {
samples = [
{ at = 0 s, value = 0 W/m^2 },
{ at = 2 s, value = 25000 W/m^2 },
{ at = 12 s, value = 25000 W/m^2 },
{ at = 14 s, value = 0 W/m^2 },
]
interpolation = linear
extrapolation = reject
}
excitation cycling_flux {
heat_flux {
target = geometry.part.region.hot_face
signal = signal.power_cycle
}
}
analysis warmup {
thermal transient {
start_time = 0 s
end_time = 20 s
time_step = 0.1 s
}
}The declared time samples are part of the study. Choose small enough to resolve load breakpoints and the shortest relevant diffusion time. Supply an initial temperature when zero kelvin is not the intended state; initial state is never inferred from an ambient-temperature load.
Why is my thermal model singular or physically wrong?
Common causes are:
- no prescribed temperature or positive convection term on a connected steady component;
- missing conductivity, density, or specific heat on a participating material region;
- a face selected where a solid heat source is required, or the reverse;
- overlapping temperature fields that prescribe conflicting nodal values;
- omitted contact/interface resistance between separate bodies;
- confusing positive inward prescribed flux with the outward component of recovered heat flux;
- time steps or mesh spacing too coarse for the thermal gradient.
RunMat separately reports missing material coverage, conflicting temperatures, unmapped sources, complex ordinates unsupported by the thermal path, and failed projection conservation.
What thermal outputs and validation checks should I use?
Request temperature fields and derived heat-flow quantities at locations supported by the procedure. Check total applied power against boundary outflow and stored-energy change:
Also perform mesh convergence at interfaces and hot spots, time-step convergence for transient work, material-property sensitivity over the actual temperature range, and boundary-condition sensitivity. Radiation is not among the currently documented thermal boundary kinds; if radiation matters, its absence must be treated as a model limitation rather than hidden in an effective convection value.
See thermal constraints, thermal loads, thermomechanical analysis, electrothermal analysis, outputs, and current status.