LS-DYNA keyword input (.k, .key, .dyn)

The LS-DYNA keyword input format: a keyword-driven ASCII deck of *NODE, *ELEMENT_*, *PART and *SET_* cards, split over as many *INCLUDEd files as the model needs. OpenRadioss reads it natively too, so it is the input format of two explicit solvers.

Format name	lsdyna
Extensions	.k, .key, .dyn
Read / Write	✓ / ✓
Extra dependencies	—

meshio++ reads the mesh (nodes, elements, parts and sets) and nothing else. Materials, sections, contacts, loads and every other keyword are skipped without a message: they are not geometry. The results of a run, in the binary d3plot family, are read by lsdyna_d3plot.

Reading & writing

import meshioplusplus

mesh = meshioplusplus.read("crash.k")           # follows *INCLUDE
meshioplusplus.write("out.k", mesh)             # nodes, elements, *PART, *SET_*

for r in mesh.regions:
    print(r.kind, r.name, r.tag, r.dim, len(r.entries))

Both engines (the C++ core and the Python reference) read the whole format and write byte-identical files. Buffers are read by the Python reader.

Card formats

A card is one line whose fields sit in fixed columns. Four layouts exist, they can be mixed in one file, and the reader tells them apart per card:

Layout	How it is selected	Fields
Standard	the default	the widths each card declares: *NODE is I8, 3E16, 2F8, an element card is ten I8, a set or *PART card is I10
Long	*KEYWORD LONG=Y (or S, K) for the file, or a + suffix on one keyword (*NODE+)	every field, integer and real, is 20 columns
I10	*KEYWORD I10=Y for the file, or a % suffix on one keyword	only the 8-column integer fields widen to 10; a different mechanism from long
Free	a comma anywhere on the line	comma-separated, any width

A - suffix (*NODE-) puts one keyword back to standard inside a long file. A *KEYWORD card sets the default for the file it appears in and for the files that file includes. Fortran real spellings are accepted (1.5D+01, 2.5-1 for 2.5e-1).

Elements

Card	meshio++ cell	Notes
*ELEMENT_SOLID	hexahedron, wedge, pyramid, tetra, tetra10	see the collapse table; one card (eid pid n1..n8) or two lines (eid pid, then up to ten nodes)
*ELEMENT_SHELL	triangle, quad	n4 == n3 or blank is a triangle; n5..n8 (mid-side nodes) are ignored with a warning
*ELEMENT_TSHELL	as *ELEMENT_SOLID	thick shells are hexahedra
*ELEMENT_BEAM	line	the orientation node n3 is dropped; an optional second card is skipped
*ELEMENT_DISCRETE	line
*ELEMENT_MASS	vertex	the mass is dropped

*ELEMENT_SHELL accepts the _THICKNESS, _BETA, _MCID and _OFFSET variants (one extra card per option, skipped); *ELEMENT_SOLID_ORTHO skips its two extra cards. Any other variant, and the *ELEMENT_SOLID_TET4TOTET10 / _H8TOH20 conversion directives, is skipped with a warning.

Element ids are unique per family (solids, shells, beams and so on may reuse a number, as LS-DYNA allows) and a duplicate within a family is an error. A card that references an undefined node is an error.

Degenerate elements

LS-DYNA has no tetra, pyramid or wedge card: they are hexahedra with repeated nodes, collapsed on read and expanded on write.

Element	LS-DYNA nodes	Read	Written
tetra	n1 n2 n3 n4 n4 n4 n4 n4	tetra (n1 n2 n3 n4)	this form
tetra	n1 n2 n3 n3 n5 n5 n5 n5	tetra (n1 n2 n3 n5)
pyramid	n1 n2 n3 n4 n5 n5 n5 n5	pyramid	this form
wedge	n1 n2 n3 n4 n5 n5 n6 n6	wedge (n1 n5 n2 n4 n6 n3)	this form
wedge	n1 n2 n3 n3 n5 n6 n7 n7	wedge (n1 n2 n3 n5 n6 n7)
wedge	any side edge collapsed in both faces, e.g. n1 n2 n3 n1 n5 n6 n7 n5	wedge (n1 n2 n3 n5 n6 n7), the faces' cyclic order kept
shell triangle	n1 n2 n3 n3	triangle	this form

The wedge spellings are read because they occur in real decks (the second is what most preprocessors write; the edge-collapsed form, added in v16.7.0 and shared with the Radioss reader, is what Radioss decks use). The node order is mapped so that the reading has a positive volume in the meshio++ convention; the tests check this numerically. A degenerate pattern not in the table stays a hexahedron.

Parts and sets

A *PART becomes a cell region: the title is the name (Part <pid> when the title is blank), the pid is the tag, and the dim is the highest topological dimension of its elements. An empty part is kept. The *PART_CONTACT, _COMPOSITE and _INERTIA variants read the same first two cards.

Keyword	Region
*SET_NODE	point
*SET_SOLID, _SHELL, _TSHELL, _BEAM, _DISCRETE	cell
*SET_PART	cell: the union of the listed parts' elements
*SET_SEGMENT	side: each segment's corner nodes are matched to a (cell, facet) pair

A set region has dim = -1 and the set id as tag. Its name is the title of a _TITLE set, otherwise <FAMILY> set <sid> (NODE set 20). The _LIST, _GENERATE, _LIST_GENERATE and _TITLE suffixes are all read. Two sets that would get the same name, tag and dimension (a titled solid set and shell set with the same title and id) are told apart by a [<family>] suffix.

A segment with n4 == n3 (or a blank n4) is a triangle. A segment shared by two cells goes to the lower-numbered one, a shell element's own face is facet 0, and a segment that matches no face is dropped with one warning. Entries of a set that name an undefined node or element are dropped with one warning as well.

Blocks

A cell block is one cell type of one element keyword, so blocks follow the sections of the deck: a *ELEMENT_SOLID that mixes hexahedra and wedges yields two blocks, and the next keyword extends the previous block when its first type matches. Cells are numbered block by block, which is the order a Cell region's indices use.

*INCLUDE

*INCLUDE, *INCLUDE_NO_TRANSFORM and *INCLUDE_TRANSFORM name one file per line (a line ending in + continues on the next). A name is looked up relative to the including file, then in each *INCLUDE_PATH / *INCLUDE_PATH_RELATIVE directory, then in the working directory; a backslash is accepted as a separator. Everything is collected into one state before ids are resolved, so a set may refer to elements defined in a later file. Nesting is limited to eight levels.

• A file that cannot be found is a warning, not an error: decks routinely include material and contact files that are not shipped with the mesh. An element that then references a missing node is an error.
• *INCLUDE_TRANSFORM offsets are not applied; the file is included untransformed, with a warning.

Skipped content

Content	Behaviour
Any other keyword	skipped silently
$ comment lines	skipped
*END	stops reading the file it appears in
*PARAMETER substitution (&name in a mesh card)	the card is skipped; one warning reports how many
PGP-encrypted blocks	skipped with a warning
20-node hexahedra (a third element line)	not supported

Writing

The writer emits *KEYWORD, the provenance comment, *NODE, one *ELEMENT_* keyword per cell block, one *PART per part, one *SET_*_LIST_TITLE per remaining region, and *END. Coordinates are written in the 16-column real field with the shortest scientific spelling that round-trips, and as many digits as fit otherwise (11 significant digits, 10 for a negative number). A mesh whose ids do not fit an 8-column field is refused rather than written in a different format.

• Parts. A cell region with a dimension (dim >= 0) is a part when none of its cells is already in an earlier part; its tag is the pid when that is positive and unused, otherwise the next free number. Every other cell region is written as a set, so no region is lost. Cells that no part claims form one part per cell block, named after the cell type. Every part gets the placeholder section and material ids 1, 1.
• Sets. A cell region that spans several element families is written as one set per family, with a warning. Point regions become *SET_NODE_LIST_TITLE, side regions *SET_SEGMENT_TITLE (the facet's corner nodes; a triangle repeats its last node).
• Cell types. vertex, line, triangle, quad, tetra, pyramid, wedge, hexahedron and tetra10 are written; anything else is a WriteError that names the type.
• A region name that starts with * or $ is written with a leading space so it is not read back as a keyword or a comment.

Quirks & limitations

• Sections and materials are placeholders. The output is meant for a mesher to LS-DYNA hand-off: add real *SECTION_* and *MAT_* cards before running it. It has not been loaded in LS-PrePost.
• Part sets and parts are not distinguished on write. A *SET_PART is read as the elements of its parts and written back as an element set.
• Empty parts come back as empty sets. A part with no elements has dim = -1, which the writer treats as a set.
• Shell nodes past n4, beam orientation nodes, element masses, *NODE constraints (tc, rc) and all part properties are dropped.
• Duplicate nodes are an error: LS-DYNA itself refuses them.
• Nodes are always read as three coordinates; a 2-D mesh is written with z = 0.