HOOMD Schema¶
HOOMD-blue supports a wide variety of per particle attributes and properties. Particles, bonds, and types can be
dynamically added and removed during simulation runs. The hoomd
schema can handle all of these situations
in a reasonably space efficient and high performance manner. It is also backwards compatible with previous versions
of itself, as we only add new additional data chunks in new versions and do not change the interpretation
of the existing data chunks. Any newer reader will initialize new data chunks with default values when they are
not present in an older version file.
- Schema name
hoomd
- Schema version
1.4
Use-cases¶
The GSD schema hoomd
provides:
Every frame of GSD output is viable for restart from
init.read_gsd
No need for a separate topology file - everything is in one
.gsd
file.Support varying numbers of particles, bonds, etc…
Support varying attributes (type, mass, etc…)
Support orientation, angular momentum, and other fields that DCD cannot.
Simple interface for dump - limited number of options that produce valid files
Binary format on disk
High performance file read and write
Support logging computed quantities
Data chunks¶
Each frame the hoomd
schema may contain one or more data chunks. The layout and names of the chunks
closely match that of the binary snapshot API in HOOMD-blue itself (at least at the time of inception).
Data chunks are organized in categories. These categories have no meaning in the hoomd
schema
specification, and are simply an organizational tool. Some file writers may implement options that act on
categories (i.e. write attributes out to every frame, or just frame 0).
Values are well defined for all fields at all frames. When a data chunk is present in frame i, it defines the values for the frame. When it is not present, the data chunk of the same name at frame 0 defines the values for frame i (when N is equal between the frames). If the data chunk is not present in frame 0, or N differs between frames, values are assumed default. Default values allow files sizes to remain small. For example, a simulation with point particles where orientation is always (1,0,0,0) would not write any orientation chunk to the file.
N may be zero. When N is zero, an index entry may be written for a data chunk with no actual data written to the file for that chunk.
Name |
Category |
Type |
Size |
Default |
Units |
---|---|---|---|---|---|
Configuration |
|||||
uint64 |
1x1 |
0 |
number |
||
uint8 |
1x1 |
3 |
number |
||
float |
6x1 |
varies |
|||
Particle data |
|||||
attribute |
uint32 |
1x1 |
0 |
number |
|
attribute |
int8 |
NTxM |
[‘A’] |
UTF-8 |
|
attribute |
uint32 |
Nx1 |
0 |
number |
|
attribute |
int8 |
NTxM |
UTF-8 |
||
attribute |
float |
Nx1 |
1.0 |
mass |
|
attribute |
float |
Nx1 |
0.0 |
charge |
|
attribute |
float |
Nx1 |
1.0 |
length |
|
attribute |
int32 |
Nx1 |
-1 |
number |
|
attribute |
float |
Nx3 |
0,0,0 |
mass * length^2 |
|
property |
float |
Nx3 |
0,0,0 |
length |
|
property |
float |
Nx4 |
1,0,0,0 |
unit quaternion |
|
momentum |
float |
Nx3 |
0,0,0 |
length/time |
|
momentum |
float |
Nx4 |
0,0,0,0 |
quaternion |
|
momentum |
int32 |
Nx3 |
0,0,0 |
number |
|
Bond data |
|||||
topology |
uint32 |
1x1 |
0 |
number |
|
topology |
int8 |
NTxM |
UTF-8 |
||
topology |
uint32 |
Nx1 |
0 |
number |
|
topology |
uint32 |
Nx2 |
0,0 |
number |
|
Angle data |
|||||
topology |
uint32 |
1x1 |
0 |
number |
|
topology |
int8 |
NTxM |
UTF-8 |
||
topology |
uint32 |
Nx1 |
0 |
number |
|
topology |
uint32 |
Nx3 |
0,0,0 |
number |
|
Dihedral data |
|||||
topology |
uint32 |
1x1 |
0 |
number |
|
topology |
int8 |
NTxM |
UTF-8 |
||
topology |
uint32 |
Nx1 |
0 |
number |
|
topology |
uint32 |
Nx4 |
0,0,0,0 |
number |
|
Improper data |
|||||
topology |
uint32 |
1x1 |
0 |
number |
|
topology |
int8 |
NTxM |
UTF-8 |
||
topology |
uint32 |
Nx1 |
0 |
number |
|
topology |
uint32 |
Nx4 |
0,0,0,0 |
number |
|
Constraint data |
|||||
topology |
uint32 |
1x1 |
0 |
number |
|
topology |
float |
Nx1 |
0 |
length |
|
topology |
uint32 |
Nx2 |
0,0 |
number |
|
Special pairs data |
|||||
topology |
uint32 |
1x1 |
0 |
number |
|
topology |
int8 |
NTxM |
utf-8 |
||
topology |
uint32 |
Nx1 |
0 |
number |
|
topology |
uint32 |
Nx2 |
0,0 |
number |
Configuration¶
-
configuration/step
¶ - Type
uint64
- Size
1x1
- Default
0
- Units
number
Simulation time step.
-
configuration/dimensions
¶ - Type
uint8
- Size
1x1
- Default
3
- Units
number
Number of dimensions in the simulation. Must be 2 or 3.
-
configuration/box
¶ - Type
float
- Size
6x1
- Default
[1,1,1,0,0,0]
- Units
varies
Simulation box. Each array element defines a different box property. See the hoomd documentation for a full description on how these box parameters map to a triclinic geometry.
box[0:3]
: \((l_x, l_y, l_z)\) the box length in each direction, in length unitsbox[3:]
: \((xy, xz, yz)\) the tilt factors, unitless values
Particle data¶
Within a single frame, the number of particles N and NT are fixed for all chunks. N and NT may vary from one frame to the next. All values are stored in hoomd native units.
Attributes¶
-
particles/N
¶ - Type
uint32
- Size
1x1
- Default
0
- Units
number
Define N, the number of particles, for all data chunks
particles/*
.
-
particles/types
¶ - Type
int8
- Size
NTxM
- Default
[‘A’]
- Units
UTF-8
Implicitly define NT, the number of particle types, for all data chunks
particles/*
. M must be large enough to accommodate each type name as a null terminated UTF-8 character string. Row i of the 2D matrix is the type name for particle type i.
-
particles/typeid
¶ - Type
uint32
- Size
Nx1
- Default
0
- Units
number
Store the type id of each particle. All id’s must be less than NT. A particle with type id has a type name matching the corresponding row in
particles/types
.
-
particles/type_shapes
¶ - Type
int8
- Size
NTxM
- Default
empty
- Units
UTF-8
Store a per-type shape definition for visualization. A dictionary is stored for each of the NT types, corresponding to a shape for visualization of that type. M must be large enough to accommodate the shape definition as a null-terminated UTF-8 JSON-encoded string. See: Shape Visualization for examples.
-
particles/mass
¶ - Type
float (32-bit)
- Size
Nx1
- Default
1.0
- Units
mass
Store the mass of each particle.
-
particles/charge
¶ - Type
float (32-bit)
- Size
Nx1
- Default
0.0
- Units
charge
Store the charge of each particle.
-
particles/diameter
¶ - Type
float (32-bit)
- Size
Nx1
- Default
1.0
- Units
length
Store the diameter of each particle.
-
particles/body
¶ - Type
int32
- Size
Nx1
- Default
-1
- Units
number
Store the composite body associated with each particle. The value -1 indicates no body. The body field may be left out of input files, as hoomd will create the needed constituent particles.
-
particles/moment_inertia
¶ - Type
float (32-bit)
- Size
Nx3
- Default
0,0,0
- Units
mass * length^2
Store the moment_inertia of each particle \((I_{xx}, I_{yy}, I_{zz})\). This inertia tensor is diagonal in the body frame of the particle. The default value is for point particles.
Properties¶
-
particles/position
¶ - Type
float (32-bit)
- Size
Nx3
- Default
0,0,0
- Units
length
Store the position of each particle (x, y, z).
All particles in the simulation are referenced by a tag. The position data chunk (and all other per particle data chunks) list particles in tag order. The first particle listed has tag 0, the second has tag 1, …, and the last has tag N-1 where N is the number of particles in the simulation.
All particles must be inside the box:
\(x > -l_x/2 + (xz-xy \cdot yz) \cdot z + xy \cdot y\) and \(x < l_x/2 + (xz-xy \cdot yz) \cdot z + xy \cdot y\)
\(y > -l_y/2 + yz \cdot z\) and \(y < l_y/2 + yz \cdot z\)
\(z > -l_z/2\) and \(z < l_z/2\)
-
particles/orientation
¶ - Type
float (32-bit)
- Size
Nx4
- Default
1,0,0,0
- Units
unit quaternion
Store the orientation of each particle. In scalar + vector notation, this is \((r, a_x, a_y, a_z)\), where the quaternion is \(q = r + a_xi + a_yj + a_zk\). A unit quaternion has the property: \(\sqrt{r^2 + a_x^2 + a_y^2 + a_z^2} = 1\).
Momenta¶
-
particles/velocity
¶ - Type
float (32-bit)
- Size
Nx3
- Default
0,0,0
- Units
length/time
Store the velocity of each particle \((v_x, v_y, v_z)\).
-
particles/angmom
¶ - Type
float (32-bit)
- Size
Nx4
- Default
0,0,0,0
- Units
quaternion
Store the angular momentum of each particle as a quaternion. See the HOOMD documentation for information on how to convert to a vector representation.
-
particles/image
¶ - Type
int32
- Size
Nx3
- Default
0,0,0
- Units
number
Store the number of times each particle has wrapped around the box \((i_x, i_y, i_z)\). In constant volume simulations, the unwrapped position in the particle’s full trajectory is
\(x_u = x + i_x \cdot l_x + xy \cdot i_y \cdot l_y + xz \cdot i_z \cdot l_z\)
\(y_u = y + i_y \cdot l_y + yz \cdot i_z * l_z\)
\(z_u = z + i_z \cdot l_z\)
Topology¶
-
bonds/N
¶ - Type
uint32
- Size
1x1
- Default
0
- Units
number
Define N, the number of bonds, for all data chunks
bonds/*
.
-
bonds/types
¶ - Type
int8
- Size
NTxM
- Default
empty
- Units
UTF-8
Implicitly define NT, the number of bond types, for all data chunks
bonds/*
. M must be large enough to accommodate each type name as a null terminated UTF-8 character string. Row i of the 2D matrix is the type name for bond type i. By default, there are 0 bond types.
-
bonds/typeid
¶ - Type
uint32
- Size
Nx1
- Default
0
- Units
number
Store the type id of each bond. All id’s must be less than NT. A bond with type id has a type name matching the corresponding row in
bonds/types
.
-
bonds/group
¶ - Type
uint32
- Size
Nx2
- Default
0,0
- Units
number
Store the particle tags in each bond.
-
angles/N
¶ - Type
uint32
- Size
1x1
- Default
0
- Units
number
Define N, the number of angles, for all data chunks
angles/*
.
-
angles/types
¶ - Type
int8
- Size
NTxM
- Default
empty
- Units
UTF-8
Implicitly define NT, the number of angle types, for all data chunks
angles/*
. M must be large enough to accommodate each type name as a null terminated UTF-8 character string. Row i of the 2D matrix is the type name for angle type i. By default, there are 0 angle types.
-
angles/typeid
¶ - Type
uint32
- Size
Nx1
- Default
0
- Units
number
Store the type id of each angle. All id’s must be less than NT. A angle with type id has a type name matching the corresponding row in
angles/types
.
-
angles/group
¶ - Type
uint32
- Size
Nx2
- Default
0,0
- Units
number
Store the particle tags in each angle.
-
dihedrals/N
¶ - Type
uint32
- Size
1x1
- Default
0
- Units
number
Define N, the number of dihedrals, for all data chunks
dihedrals/*
.
-
dihedrals/types
¶ - Type
int8
- Size
NTxM
- Default
empty
- Units
UTF-8
Implicitly define NT, the number of dihedral types, for all data chunks
dihedrals/*
. M must be large enough to accommodate each type name as a null terminated UTF-8 character string. Row i of the 2D matrix is the type name for dihedral type i. By default, there are 0 dihedral types.
-
dihedrals/typeid
¶ - Type
uint32
- Size
Nx1
- Default
0
- Units
number
Store the type id of each dihedral. All id’s must be less than NT. A dihedral with type id has a type name matching the corresponding row in
dihedrals/types
.
-
dihedrals/group
¶ - Type
uint32
- Size
Nx2
- Default
0,0
- Units
number
Store the particle tags in each dihedral.
-
impropers/N
¶ - Type
uint32
- Size
1x1
- Default
0
- Units
number
Define N, the number of impropers, for all data chunks
impropers/*
.
-
impropers/types
¶ - Type
int8
- Size
NTxM
- Default
empty
- Units
UTF-8
Implicitly define NT, the number of improper types, for all data chunks
impropers/*
. M must be large enough to accommodate each type name as a null terminated UTF-8 character string. Row i of the 2D matrix is the type name for improper type i. By default, there are 0 improper types.
-
impropers/typeid
¶ - Type
uint32
- Size
Nx1
- Default
0
- Units
number
Store the type id of each improper. All id’s must be less than NT. A improper with type id has a type name matching the corresponding row in
impropers/types
.
-
impropers/group
¶ - Type
uint32
- Size
Nx2
- Default
0,0
- Units
number
Store the particle tags in each improper.
-
constraints/N
¶ - Type
uint32
- Size
1x1
- Default
0
- Units
number
Define N, the number of constraints, for all data chunks
constraints/*
.
-
constraints/value
¶ - Type
float
- Size
Nx1
- Default
0
- Units
length
Store the distance of each constraint. Each constraint defines a fixed distance between two particles.
-
constraints/group
¶ - Type
uint32
- Size
Nx2
- Default
0,0
- Units
number
Store the particle tags in each constraint.
-
pairs/N
¶ - Type
uint32
- Size
1x1
- Default
0
- Units
number
Define N, the number of special pair interactions, for all data chunks
pairs/*
.New in version 1.1.
-
pairs/types
¶ - Type
int8
- Size
NTxM
- Default
empty
- Units
UTF-8
Implicitly define NT, the number of special pair types, for all data chunks
pairs/*
. M must be large enough to accommodate each type name as a null terminated UTF-8 character string. Row i of the 2D matrix is the type name for particle type i. By default, there are 0 special pair types.New in version 1.1.
-
pairs/typeid
¶ - Type
uint32
- Size
Nx1
- Default
0
- Units
number
Store the type id of each special pair interaction. All id’s must be less than NT. A pair with type id has a type name matching the corresponding row in
pairs/types
.New in version 1.1.
-
pairs/group
¶ - Type
uint32
- Size
Nx2
- Default
0,0
- Units
number
Store the particle tags in each special pair interaction.
New in version 1.1.
State data¶
HOOMD stores auxiliary state information in state/*
data chunks. Auxiliary state encompasses internal state
to any integrator, updater, or other class that is not part of the particle system state but is also not a fixed
parameter. For example, the internal degrees of freedom in integrator. Auxiliary state is useful when restarting
simulations.
HOOMD only stores state in GSD files when requested explicitly by the user. Only a few of the documented state data chunks will be present in any GSD file and not all state chunks are valid. Thus, state data chunks do not have default values. If a chunk is not present in the file, that state does not have a well-defined value.
Note
HOOMD-blue versions 3.0 and newer write state data in an application defined
format in log/*
, not in state/*
. See the HOOMD-blue
documentation for details on the data chunks it reads and writes.
Name |
Type |
Size |
Units |
---|---|---|---|
HPMC integrator state |
|||
double |
1x1 |
length |
|
double |
1x1 |
number |
|
float |
NTx1 |
length |
|
uint8 |
NTx1 |
boolean |
|
float |
NTx1 |
length |
|
float |
NTx1 |
length |
|
float |
NTx1 |
length |
|
uint32 |
NTx1 |
number |
|
float |
sum(N)x3 |
length |
|
uint32 |
NTx1 |
number |
|
float |
sum(N)x3 |
length |
|
float |
NTx1 |
length |
|
uint32 |
NTx1 |
number |
|
float |
sum(N)x2 |
length |
|
uint32 |
NTx1 |
number |
|
float |
sum(N)x2 |
length |
|
float |
NTx1 |
length |
|
uint32 |
NTx1 |
number |
|
float |
sum(N)x2 |
length |
HPMC integrator state¶
NT is the number of particle types.
-
state/hpmc/integrate/d
¶ - Type
double
- Size
1x1
- Units
length
d is the maximum trial move displacement.
New in version 1.2.
-
state/hpmc/integrate/a
¶ - Type
double
- Size
1x1
- Units
number
a is the size of the maximum rotation move.
New in version 1.2.
-
state/hpmc/sphere/radius
¶ - Type
float
- Size
NTx1
- Units
length
Sphere radius for each particle type.
New in version 1.2.
-
state/hpmc/sphere/orientable
¶ - Type
uint8
- Size
NTx1
- Units
boolean
Orientable flag for each particle type.
New in version 1.3.
-
state/hpmc/ellipsoid/a
¶ - Type
float
- Size
NTx1
- Units
length
Size of the first ellipsoid semi-axis for each particle type.
New in version 1.2.
-
state/hpmc/ellipsoid/b
¶ - Type
float
- Size
NTx1
- Units
length
Size of the second ellipsoid semi-axis for each particle type.
New in version 1.2.
-
state/hpmc/ellipsoid/c
¶ - Type
float
- Size
NTx1
- Units
length
Size of the third ellipsoid semi-axis for each particle type.
New in version 1.2.
-
state/hpmc/convex_polyhedron/N
¶ - Type
uint32
- Size
NTx1
- Units
number
Number of vertices defined for each type.
New in version 1.2.
-
state/hpmc/convex_polyhedron/vertices
¶ - Type
float
- Size
sum(N)x3
- Units
length
Position of the vertices in the shape for all types. The shape for type 0 is the first N[0] vertices, the shape for type 1 is the next N[1] vertices, and so on…
New in version 1.2.
-
state/hpmc/convex_spheropolyhedron/N
¶ - Type
uint32
- Size
NTx1
- Units
number
Number of vertices defined for each type.
New in version 1.2.
-
state/hpmc/convex_spheropolyhedron/vertices
¶ - Type
float
- Size
sum(N)x3
- Units
length
Position of the vertices in the shape for all types. The shape for type 0 is the first N[0] vertices, the shape for type 1 is the next N[1] vertices, and so on…
New in version 1.2.
-
state/hpmc/convex_spheropolyhedron/sweep_radius
¶ - Type
float
- Size
NTx1
- Units
length
Sweep radius for each type.
New in version 1.2.
-
state/hpmc/convex_polygon/N
¶ - Type
uint32
- Size
NTx1
- Units
number
Number of vertices defined for each type.
New in version 1.2.
-
state/hpmc/convex_polygon/vertices
¶ - Type
float
- Size
sum(N)x2
- Units
length
Position of the vertices in the shape for all types. The shape for type 0 is the first N[0] vertices, the shape for type 1 is the next N[1] vertices, and so on…
New in version 1.2.
-
state/hpmc/convex_spheropolygon/N
¶ - Type
uint32
- Size
NTx1
- Units
number
Number of vertices defined for each type.
New in version 1.2.
-
state/hpmc/convex_spheropolygon/vertices
¶ - Type
float
- Size
sum(N)x2
- Units
length
Position of the vertices in the shape for all types. The shape for type 0 is the first N[0] vertices, the shape for type 1 is the next N[1] vertices, and so on…
New in version 1.2.
-
state/hpmc/convex_spheropolygon/sweep_radius
¶ - Type
float
- Size
NTx1
- Units
length
Sweep radius for each type.
New in version 1.2.
-
state/hpmc/simple_polygon/N
¶ - Type
uint32
- Size
NTx1
- Units
number
Number of vertices defined for each type.
New in version 1.2.
-
state/hpmc/simple_polygon/vertices
¶ - Type
float
- Size
sum(N)x2
- Units
length
Position of the vertices in the shape for all types. The shape for type 0 is the first N[0] vertices, the shape for type 1 is the next N[1] vertices, and so on…
New in version 1.2.
Logged data¶
Users may store logged data in log/*
data chunks. Logged data encompasses values computed at simulation time that
are too expensive or cumbersome to re-compute in post processing. This specification does not define specific chunk
names or define logged data. Users may select any valid name for logged data chunks as appropriate for their workflow.
For any named logged data chunks present in any frame frame the file: If a chunk is not present in a given frame i != 0, the implementation should provide the quantity as read from frame 0 for that frame. GSD files that include a logged data chunk only in some frames i != 0 and not in frame 0 are invalid.
By convention, per-particle and per-bond logged data should have a chunk name starting with log/particles/
and
log/bonds
, respectively. Scalar, vector, and string values may be stored under a different prefix starting with
log/
. This specification may recognize additional conventions in later versions without invalidating existing files.
Name |
Type |
Size |
Units |
---|---|---|---|
n/a |
NxM |
user-defined |
|
n/a |
NxM |
user-defined |
|
n/a |
NxM |
user-defined |
-
log/particles/user_defined
¶ - Type
user-defined
- Size
NxM
- Units
user-defined
This chunk is a place holder for any number of user defined per-particle quantities. N is the number of particles in this frame. M, the data type, the units, and the chunk name (after the prefix
log/particles/
) are user-defined.New in version 1.4.
-
log/bonds/user_defined
¶ - Type
user-defined
- Size
NxM
- Units
user-defined
This chunk is a place holder for any number of user defined per-bond quantities. N is the number of bonds in this frame. M, the data type, the units, and the chunk name (after the prefix
log/bonds/
) are user-defined.New in version 1.4.
-
log/user_defined
¶ - Type
user-defined
- Size
NxM
- Units
user-defined
This chunk is a place holder for any number of user defined quantities. N, M, the data type, the units, and the chunk name (after the prefix
log/
) are user-defined.New in version 1.4.