---
license:
  - cc-by-4.0
  - cc-by-sa-4.0
  - odbl
license_details: >-
  Three shards under three incompatible licences; see the licence table below.
  Load one config at a time.
language:
  - en
tags:
  - materials-science
  - physics-simulation
  - finite-element
  - mpm
  - wood
  - mechanical-properties
  - 3d-assets
pretty_name: Macro Material Properties for 3D Physics Simulation
size_categories:
  - 10K<n<100K
configs:
  - config_name: permissive
    default: true
    data_files: partnext_material_rho_E_nu_CC-BY-4.0.csv
  - config_name: share_alike
    data_files: partnext_material_rho_E_nu_CC-BY-SA-4.0.csv
  - config_name: odbl
    data_files: partnext_material_rho_E_nu_ODbL.csv
---

# Macro Material Properties for 3D Physics Simulation
### Grade-level rho / E / nu priors with per-quantity provenance

**21,160 rows — 20,315 distinct material grades across 15 macro material classes** —
each carrying **density ρ, Young's modulus E and Poisson's ratio ν**, assembled from
56 public sources for annotating 3D assets and driving rigid-body and deformable
physics simulation (Genesis, Isaac Gym, FEM, MPM).

The distinguishing feature of this table is not its size. It is that **every single
value states what it actually is** — measured, derived, or filled — and where it is not
a direct measurement, **the error has been quantified against data inside the table
itself**, not asserted.

All fields, including the per-row `citation_gap_note`, are in English.

Read `LIMITATIONS.md` before using this for deformable simulation.

---

## 1 · Veracity — what is actually measured

| | ρ | E | ν |
|---|---|---|---|
| **reported by the cited source, a handbook, or a grade standard** | **20,329 (96.1 %)** | 1,912 (9.0 %) | 732 (3.5 %) |
| derived from this row's density | – | 19,033 (89.9 %) | – |
| handbook softwood/hardwood group median | – | – | 19,138 (90.4 %) |
| alloy-family / class prior / solver clamp / computed | 831 | 215 | 1,290 |

Only **514 rows (2.4 %)** have all three quantities stated directly by a source.

**ρ is the only quantity measured at scale.** E and ν are estimates on ~90 % of rows.
This is a property of the upstream sources — the Global Wood Density Database, which
supplies 85 % of the rows, publishes density only. What this table adds is that each
estimate is labelled for what it is and its error is measured.

The `value_provenance` column states the origin of each quantity separately, e.g.
`rho=source;E=derived_from_rho;nu=handbook_group`. The `citation_gap_note` column then
says, for each non-source value, exactly where it came from and what its measured error
is. **636 rows carry a value that the cited paper does not report at all** — an
alloy-family standard substituted upstream after the original was judged corrupt. Those
are labelled `family_standard`, never `source`, and say so in the note.

## 2 · Plausibility — how it was checked

Four independent checks, none of which relies on assertion:

**Internal consistency of wood (19,294 rows).** Longitudinal sound speed √(E/ρ) has
median **4,296 m s⁻¹**, P1 4,056, P99 4,727. This is the known species-invariant value
for wood, so E and ρ are mutually consistent row by row — a check that needs no external
reference at all.

**Isotropic elasticity closure (287 rows with a source-reported shear modulus).**
G = E/(2(1+ν)) holds to a **median 0.24 %** on the 273 rows in isotropic classes. The same
identity over-predicts by 3.3–8.9× on wood and fibre laminates — which is how the anisotropy
caveat in `LIMITATIONS.md` is quantified rather than guessed. (The few outliers inside the
isotropic set are themselves anisotropic materials, e.g. foliated gneiss.)

**Prior-knowledge spot check.** 101 checks over 68 well-characterised timber species
spanning temperate/tropical, softwood/hardwood and the full density range:
**ρ 96/101, E 95/101** inside published ranges.

**Numerical conditions, every row.** 0 rows with ν ∉ [0, 0.5), non-positive bulk or
shear modulus, or out-of-range wave speed.

Two calibration steps sit behind the wood values, both with factors derived from data
inside the table and both validated: the density regression is calibrated against USDA
measured MOE (leave-one-out median bias +0.0 %), and all wood is expressed on a single
12 % moisture basis. `LIMITATIONS.md` §2–3 gives the factors and their derivation.

## 3 · Traceability — every value has a recorded origin

- **96.2 % of source-reported cells were round-tripped**: recomputed from the original
  upstream data file and matched. GWDD 18,046/18,046 recomputed from 109,626 raw
  measurement records under the documented aggregation rule; CIRAD 1,010/1,010;
  SciGlass 890/890; USDA 444/444; Materials Project 342/342; StressEng 677/677.
- **Every filled cell traces to an upstream `fill_log` entry** — 40,936 cells, 0 unmatched.
  Where a value differs from the raw log entry it is because one of the two calibration
  steps in `LIMITATIONS.md` §2–3 applies, and the row says so.
  The log records the route and the reference for each fill (e.g. `C_SG_REGRESSION` →
  USDA Wood Handbook FPL-GTR-190 Ch. 5).
- **Grade names**: 84.0 % verified against a formal designation system — 17,597 botanical
  binomials resolved to `accepted` in World Flora Online, plus CHEMICAL_FORMULA, AA, AISI/SAE
  and steel grade codes. The remaining 16 % are trade or descriptive names with no standard
  designation. Only 2 non-grade-level rows survive in the published shards (both named
  `[proxy]`); the other 23 class-level envelopes were withheld.
- The remaining 848 source cells (3.8 %) come from ~50 small sources, mostly single
  open-access papers contributing 2–96 cells each, cited but not independently re-derived.

## 4 · Physics simulation — what you can run

**Rigid body (Genesis, Isaac Gym): ρ and E are ready, friction is not in this table.**
ρ is measured on 96 % of rows and is what sets mass and inertia; E only enters contact
stiffness, where order-of-magnitude accuracy suffices. A rigid-body material in Genesis or
Isaac also needs static and dynamic friction and a restitution coefficient, and **this
release carries none of them** — supply them from your own contact model. Restitution in
particular is a contact-pair property, not a material constant, and no source here reports it.

**Deformable (FEM / MPM): no row diverges, and the caveats are per-row.**

| tier | rows | share | meaning |
|---|---|---|---|
| **F1** direct | 653 | 3.1 % | no caveat |
| **F2a** clamp ν | 38 | 0.2 % | ν > 0.45 would lock linear tetrahedra — use `poissons_ratio_fem_safe`. 48 rows carry a clamped value in total; the other 10 are wood and appear under F2b |
| **F2b** anisotropy collapsed | 18,627 | 88.0 % | magnitude usable, **deformation shape wrong**: transverse stiffness too high by 3–9× |
| **F2c** ν is a proxy | 1,168 | 5.5 % | numerically fine, directionality lost |
| **F3** E error known large | 672 | 3.2 % | low-density wood, E over-predicted 30–100 % |
| **F4** blocked | **2** | 0.01 % | an 8 kPa hydrogel and a 2.5 kPa knit sag > 20 % under self-weight at 0.2 m — real behaviour, not a data error |

By class: metals, ceramics, glass, stone, plastics, foam, paper and engineered wood are
F1/F2c throughout. Wood is F2b because of the isotropic collapse. Elastomers need the
clamp column.

`poissons_ratio_fem_safe` = min(ν, 0.45) is a **solver parameter, not a material property**.
The cost of clamping is real — K/G falls from 49.7 at ν = 0.49 to 9.7 at ν = 0.45 — so with
mixed / F-bar elements or a near-incompressible constitutive model, use `poissons_ratio`.

**Linear-elastic validity windows** (beyond these, a linear model is the wrong physics
regardless of parameter accuracy): wood ≈ 0.3–0.5 % strain, laminates ≈ 0.5–1 %, foams
≈ 2 %, fabrics 1–5 % (the window the moduli were fitted in), elastomers require a
hyperelastic model above ~10–20 %.

---

## Three license shards

CC BY-SA and ODbL are both share-alike and **cannot be merged into one derived dataset**,
so the table ships as three. The permissive shard alone covers all 15 classes.

| config | rows | licence | contents |
|---|---|---|---|
| `permissive` (default) | 19,715 | CC BY 4.0 | CC0 / CC BY / MIT / US-Gov public domain. All 15 classes. |
| `share_alike` | 1,021 | CC BY-SA 4.0 | CIRAD tropical timber + Wikidata cross-checks |
| `odbl` | 424 | ODC-ODbL | SciGlass optical and technical glasses |

33 further rows are withheld — 12 Ashby teaching-licence class envelopes, 10 internal
consensus rows, 10 rows whose recorded licence string ("open research dataset") is not a
licence identifier, and one manufacturer datasheet. No real material is lost, but note the side effect:
those rows were the class-level fallbacks, so a part identified only as "wood" has no
single row to fall back on and must be matched against 19,294 species. `class_defaults.csv`
is provided for that case: a per-class median computed from the permissive shard itself.

## Columns

| column | meaning |
|---|---|
| `material_class` | one of 15 macro classes |
| `material_grade` | grade / species / designation — unique together with `source_id` |
| `density_kg_m3` | ρ, kg·m⁻³ |
| `youngs_modulus_GPa` | E, GPa |
| `poissons_ratio` | ν as reported or filled |
| `poissons_ratio_fem_safe` | min(ν, 0.45) — solver parameter, not a material property |
| `value_provenance` | `rho=…;E=…;nu=…`, per-quantity origin |
| `citation_gap_note` | for every non-source value: real origin and measured error |
| `source_id`, `source_citation`, `source_doi_or_url`, `source_license`, `redistribution_clearance` | provenance and licensing |

`class_defaults.csv` is a separate 15-row helper for retrieval: per macro class, the median
ρ / E / ν over the permissive shard with a p10–p90 band, a row count, and a
`fallback_reliability` flag. Use it only when a part's class is known but no grade can be
matched — the band shows what falling back costs. Three classes are flagged: `glass`
(only 3 rows are permissive, 424 are in the `odbl` shard), `textile_woven` (mixes
fabric-scale and fibre-direction rows five orders of magnitude apart, so its median is
meaningless) and `wood_solid` (the isotropic-collapse caveat applies to every row).

`value_provenance` vocabulary: `source` · `handbook` / `handbook_group` ·
`named_standard` · `family_standard` (**cited paper does not report this quantity**) ·
`derived_from_rho` · `class_prior` / `class_prior_self` · `sim_clamp` · `computed`.

## Citation

Cite the sources listed in each row's `source_citation`. The largest contributors are:

- **Fischer, F.J., Chave, J., Zanne, A.E. et al. (2026).** Global Wood Density Database v.2
  (GWDD v.2), v2.2, Zenodo, [10.5281/zenodo.20815517](https://doi.org/10.5281/zenodo.20815517)
  — CC BY 4.0. Accompanying article: Fischer et al. (2026), *Beyond species means — the
  intraspecific contribution to global wood density variation*, New Phytologist,
  [10.1111/nph.70860](https://doi.org/10.1111/nph.70860). Supplies 18,046 rows.
  The source file was identified by checksum: the copy used here
  (`gwdd_v2.2.csv`, 73,386,310 bytes, md5 `646454bd0ebcecb201fe081d0efde8fe`) is
  byte-identical to the file of that name in Zenodo record 20815517.
- **CIRAD wood density database** (Vieilledent et al.), CC BY-SA 4.0 — 1,010 rows.
- **USDA FPL Wood Handbook FPL-GTR-190**, US Government work — 262 rows.
- **StressEng** (Kumar, Kabra & Cole 2024, *Scientific Data* 11:1273), CC BY 4.0 — 749 rows.
- **SciGlass** (EPAM Systems), ODC-ODbL — 424 rows.
- **Materials Project** (Jain et al. 2013), CC BY 4.0 — 114 rows.
- **MIL-HDBK-5J** (2003) and **Smith (1976)** *J. Res. NBS* 80A(1), 45–49, US Government works.