Up: Home page for Qhull (local)
Up: Qhull manual: contents
To: ProgramsOptionsOutputFormatsGeomviewPrintQhullPrecisionTraceFunctions (local)
To: synopsis • input • outputs • controls • graphics • notes • conventions • options


[halfspace]qhalf -- halfspace intersection about a point

The intersection of a set of halfspaces is a polytope. The polytope may be unbounded. See Preparata & Shamos ['85] for a discussion. In low dimensions, halfspace intersection may be used for linear programming.

Example: rbox c | qconvex FV n | qhalf Fp

Print the intersection of the facets of a cube. rbox c generates the vertices of a cube. qconvex FV returns a feasible point for halfspace intersection. This feasible or interior point, qconvex FV, is the average of the cube's vertices (i.e., the origin). qconvex n returns the halfspaces that define the cube. qhalf Fp computes the intersection of the halfspaces about the feasible point. The intersection is the vertices of the original cube.

Example: rbox c | qconvex FQ n | qhalf H0.1,0.1,0.1

Compute the intersection of the facets of a cube and print a summary ('s'). Option 'FQ' prints the qconvex command as an input comment for the summary. 'qhalf Hn,n,n' specifies the feasible point as [0.1, 0.1, 0.1]. 'qhalf H0' would specify the feasible point as the origin.

Example: rbox c d G0.55 | qconvex FQ FV n | qhalf Fp

Print the intersection of the facets of a cube and a diamond. There are 24 facets and 14 intersection points. Four facets define each diamond vertex. Six facets define each cube vertex.

Example: rbox c d G0.55 | qconvex FQ FV n | qhalf Fp Qt

Same as above except triangulate before computing the intersection points. Three facets define each intersection point. There are two duplicates of the diamond and four duplicates of the cube.

Example: rbox 10 s t10 | qconvex FQ FV n | qhalf Fp Fn

Print the intersection of the facets of the convex hull of 10 cospherical points. Include the intersection points and the neighboring intersections. As in the previous examples, the intersection points are nearly the same as the original input points.

In Qhull, a halfspace is defined by the points on or below a hyperplane. The distance of each point to the hyperplane is less than or equal to zero.

Qhull computes a halfspace intersection by the geometric duality between points and halfspaces. See halfspace examples, qhalf notes, and option 'p' of qhalf outputs.

Qhalf's outputs are the intersection points (Fp) and the neighboring intersection points (Fn). For random inputs, halfspace intersections are usually defined by more than d halfspaces. See the sphere example.

The identity pipeline for Qhull starts with points, produces the halfspaces for their convex hull, and intersects these halfspaces, returning the original points. For example, 'rbox c' is the unit cube.

	rbox c | qconvex FV n | qhalf Fp
	3
	8
	  -0.5    0.5    0.5
	   0.5    0.5    0.5
	  -0.5    0.5   -0.5
	   0.5    0.5   -0.5
	   0.5   -0.5    0.5
	  -0.5   -0.5    0.5
	   0.5   -0.5   -0.5
	  -0.5   -0.5   -0.5

You can try triangulated output ('Qt') and joggled input ('QJ'). It demonstrates that triangulated output is more accurate than joggled input.

If you use 'Qt' (triangulated output), all halfspace intersections are simplicial (e.g., three halfspaces per intersection in 3-d). In 3-d, if more than three halfspaces intersect at the same point, triangulated output will produce duplicate intersections, one for each additional halfspace. See the third example, or add 'Qt' to the sphere example.

If you use 'QJ' (joggled input), all halfspace intersections are simplicial. This may lead to nearly identical intersections. For example, either replace 'Qt' with 'QJ' above, or add 'QJ' to the sphere example. See Merged facets or joggled input.

The 'qhalf' program is equivalent to 'qhull H'. It disables the following Qhull options: d n v Qbb QbB Qf Qg Qm Qr Qv Qx Qz TR E V Fa FA FC FD FS Ft FV Gt Q0,etc.

Copyright © 1995-2020 C.B. Barber


»qhalf synopsis

qhalf -- halfspace intersection about a point.
    input (stdin): [dimension, 1, interior point]
                       dimension+1, number of halfspaces, coefficients+offset
    comments start with a non-numeric character

options:
    Hn,n - specify coordinates of interior point
    Qt   - triangulated output
    QJ   - joggled input instead of merged facets
    Tv   - verify result: structure, convexity, and redundancy
    .    - concise list of all options
    -    - one-line description of each option
    -?   - this message
    -V   - version

output options (subset):
    s    - summary of results (default)
    Fp   - intersection coordinates
    Fv   - non-redundant halfspaces incident to each intersection
    Fx   - non-redundant halfspaces
    G    - Geomview output (dual convex hull)
    m    - Mathematica output (dual convex hull)
    o    - OFF file format (dual convex hull)
    QVn  - print intersections for halfspace n, -n if not
    TI file - input file, may be enclosed in single quotes
    TO file - output file, may be enclosed in single quotes

examples:
    rbox d | qconvex FQ n | qhalf s H0,0,0 Fp
    rbox c | qconvex FQ FV n | qhalf s i
    rbox c | qconvex FQ FV n | qhalf s o

»qhalf input

The input data on stdin consists of:

Use I/O redirection (e.g., qhalf < data.txt), a pipe (e.g., rbox c | qconvex FV n | qhalf), or the 'TI' option (e.g., qhalf TI data.txt).

Qhull needs a feasible point to compute the halfspace intersection. A feasible point is clearly inside all of the halfspaces. A point is inside a halfspace if its distance to the corresponding hyperplane is negative.

The feasible point may be listed at the beginning of the input (as shown above). If not, option 'Hn,n' defines the feasible point as [n,n,0,...] where 0 is the default coordinate (e.g., 'H0' is the origin). Use linear programming if you do not know the feasible point (see halfspace notes),

The input to qhalf is a set of halfspaces that are defined by their hyperplanes. Each halfspace is defined by d coefficients followed by a signed offset. This defines a linear inequality. The coefficients define a vector that is normal to the halfspace. The vector may have any length. If it has length one, the offset is the distance from the origin to the halfspace's boundary. Points in the halfspace have a negative distance to the hyperplane. The distance from the feasible point to each halfspace is likewise negative.

The halfspace format is the same as Qhull's output options 'n', 'Fo', and 'Fi'. Use option 'Fd' to use cdd format for the halfspaces.

For example, here is the input for computing the intersection of halfplanes that form a cube.

rbox c | qconvex FQ FV n TO data

RBOX c | QCONVEX FQ FV n
3 1
     0      0      0
4
6
     0      0     -1   -0.5
     0     -1      0   -0.5
     1      0      0   -0.5
    -1      0      0   -0.5
     0      1      0   -0.5
     0      0      1   -0.5

qhalf s Fp < data


Halfspace intersection by the convex hull of 6 points in 3-d:

  Number of halfspaces: 6
  Number of non-redundant halfspaces: 6
  Number of intersection points: 8

Statistics for: RBOX c | QCONVEX FQ FV n | QHALF s Fp

  Number of points processed: 6
  Number of hyperplanes created: 11
  Number of distance tests for qhull: 11
  Number of merged facets: 1
  Number of distance tests for merging: 45
  CPU seconds to compute hull (after input):  0

3
3
8
  -0.5    0.5    0.5
   0.5    0.5    0.5
  -0.5    0.5   -0.5
   0.5    0.5   -0.5
   0.5   -0.5    0.5
  -0.5   -0.5    0.5
  -0.5   -0.5   -0.5
   0.5   -0.5   -0.5

»qhalf outputs

The following options control the output for halfspace intersection.

 
Intersections
FN
list intersection points for each halfspace. The first line is the number of halfspaces. Each remaining line starts with the number of intersection points. Redundant halfspaces have 0 intersection points. For the cube example, each halfspace has four intersection points.
Fn
list neighboring intersections for each intersection point. The first line is the number of intersection points. Each following line starts with the number of neighboring intersections. For the cube example, each intersection point has three neighboring intersections.

In 3-d, a non-simplicial intersection has more than three neighboring intersections. For random data (e.g., the sphere example), non-simplicial intersections are the norm. Option 'Qt' produces three neighboring intersections per intersection by duplicating the intersection points. Option QJ' produces three neighboring intersections per intersection by joggling the hyperplanes and hence their intersections.

Fp
print intersection coordinates. The first line is the dimension and the second line is the number of intersection points. The following lines are the coordinates of each intersection point. The "infinity" point, [-10.101,-10.101,...] (qh_INFINITE), indicates an unbounded intersection.
FI
list intersection IDs. The first line is the number of intersections. The IDs follow, one per line.
 
 
Halfspaces
Fx
list non-redundant halfspaces. The first line is the number of non-redundant halfspaces. The other lines list one halfspace per line. A halfspace is non-redundant if it defines a facet of the intersection. Redundant halfspaces are ignored. For the cube example, none of the halfspaces are redundant.
Fv
list non-redundant halfspaces incident to each intersection point. The first line is the number of non-redundant halfspaces. Each remaining line starts with the number of non-redundant halfspaces incident to that point. For the cube example, each intersection point is incident to three halfspaces.
i
list non-redundant halfspaces incident to each intersection point. The first line is the number of intersection points. Each remaining line lists the incident, non-redundant halfspaces for that intersection point. For the cube example, each intersection point is incident to three halfspaces.
Fc
list coplanar halfspaces for each intersection point. The first line is the number of intersection points. Each remaining line starts with the number of coplanar halfspaces. A coplanar halfspace is listed for one intersection point even though it is coplanar to multiple intersection points.
Qi Fc
list redundant halfspaces for each intersection point. The first line is the number of intersection points. Each remaining line starts with the number of redundant halfspaces. Use options 'Qc Qi Fc' to list coplanar and redundant halfspaces.
 
 
General
s
print summary for the halfspace intersection. Use 'Fs' if you need numeric data.
o
print vertices and facets of the dual convex hull. The first line is the dimension. The second line is the number of vertices, facets, and ridges. The vertex coordinates are next, followed by the facets, one per line.
p
print vertex coordinates of the dual convex hull. Each vertex corresponds to a non-redundant halfspace. Its coordinates are the negative of the hyperplane's coefficients divided by the offset plus the inner product of the coefficients and the feasible point (-c/(b+a.p). Options 'p Qc' includes coplanar halfspaces. Options 'p Qi' includes redundant halfspaces.
m
Mathematica output for the dual convex hull in 2-d or 3-d.
FM
Maple output for the dual convex hull in 2-d or 3-d.
G
Geomview output for the dual convex hull in 2-d, 3-d, or 4-d.

»qhalf controls

These options provide additional control:

Qt
triangulated output. If a 3-d intersection is defined by more than three hyperplanes, Qhull produces duplicate intersections -- one for each extra hyperplane.
QRn
randomly rotate the input with a random seed of n. If n=0, the seed is the time. If n=-1, use time for the random seed, but do not rotate the input.
QJ
joggle the input instead of merging facets. In 3-d, this guarantees that each intersection is defined by three hyperplanes.
f
facet dump. Print the data structure for each intersection (i.e., facet)
TFn
report summary after constructing n intersections
QVn
select intersection points for halfspace n (marked 'good')
QGn
select intersection points that are visible to halfspace n (marked 'good'). Use -n for the remainder.
Qbk:0Bk:0
remove the k-th coordinate from the input. This computes the halfspace intersection in one lower dimension.
Tv
verify result
TI file
input data from file. The filename may not use spaces or quotes.
TO file
output results to file. Use single quotes if the filename contains spaces (e.g., TO 'file with spaces.txt'
Qs
search all points for the initial simplex. If Qhull can not construct an initial simplex, it reports a descriptive message. Usually, the point set is degenerate and one or more dimensions should be removed ('Qbk:0Bk:0'). If not, use option 'Qs'. It performs an exhaustive search for the best initial simplex. This is expensive is high dimensions.

»qhalf graphics

To view the results with Geomview, compute the convex hull of the intersection points ('qhull FQ H0 Fp | qhull G'). See Halfspace examples.

»qhalf notes

See halfspace intersection for precision issues related to qhalf.

If you do not know a feasible point for the halfspaces, use linear programming to find one. Assume, n halfspaces defined by: aj*x1+bj*x2+cj*x3+dj<=0, j=1..n. Perform the following linear program:

max(x5) aj*x1+bj*x2+cj*x3+dj*x4+x5<=0, j=1..n

Then, if [x1,x2,x3,x4,x5] is an optimal solution with x4>0 and x5>0 we get:

aj*(x1/x4)+bj*(x2/x4)+cj*(x3/x4)+dj<=(-x5/x4) j=1..n and (-x5/x4)<0,

and conclude that the point [x1/x4,x2/x4,x3/x4] is inside all the halfspaces. Since x5 is optimal, this feasible point is "clearly inside" the halfspaces (good for precision errors).

After finding a feasible point, the rest of the intersection algorithm is from Preparata & Shamos ['85, p. 316, "A simple case ..."]. Translate the halfspaces so that the feasible point is the origin. Calculate the dual polytope. The dual polytope is the convex hull of the vertices dual to the original faces in regard to the unit sphere (i.e., halfspaces at distance d from the origin are dual to vertices at distance 1/d). Then calculate the resulting polytope, which is the dual of the dual polytope, and translate the origin back to the feasible point [S. Spitz, S. Teller, D. Strawn].

»qhalf conventions

The following terminology is used for halfspace intersection in Qhull. This is the hardest structure to understand. The underlying structure is a convex hull with one vertex per non-redundant halfspace. See convex hull conventions and Qhull's data structures.

»qhalf options

qhalf -- compute the intersection of halfspaces about a point
    http://www.qhull.org

input (stdin):
    optional interior point: dimension, 1, coordinates
    first lines: dimension+1 and number of halfspaces
    other lines: halfspace coefficients followed by offset
    comments:    start with a non-numeric character

options:
    Hn,n - specify coordinates of interior point
    Qc   - keep coplanar halfspaces
    Qi   - keep other redundant halfspaces
    QJ   - joggled input instead of merged facets
    Qt   - triangulated output

Qhull control options:
    Qa   - allow input with fewer or more points than coordinates
    Qbk:0Bk:0 - remove k-th coordinate from input
    QJn  - randomly joggle input in range [-n,n]
    QRn  - random rotation (n=seed, n=0 time, n=-1 time/no rotate)
    Qs   - search all halfspaces for the initial simplex

Qhull extra options:
    QGn  - print intersection if visible to halfspace n, -n for not
    QVn  - print intersections for halfspace n, -n if not
    Qw   - allow option warnings
    Q12  - allow wide facets and wide dupridge
    Q14  - merge pinched vertices that create a dupridge

T options:
    TFn  - report summary when n or more facets created
    TI file - input file, may be enclosed in single quotes
    TO file - output file, may be enclosed in single quotes
    Ts   - statistics
    Tv   - verify result: structure, convexity, and in-circle test
    Tz   - send all output to stdout

Trace options:
    T4   - trace at level n, 4=all, 5=mem/gauss, -1= events
    Ta   - annotate output with message codes
    TAn  - stop qhull after adding n vertices
     TCn - stop qhull after building cone for point n
     TVn - stop qhull after adding point n, -n for before
    Tc   - check frequently during execution
    Tf   - flush each qh_fprintf for debugging segfaults
    TPn - turn on tracing when point n added to hull
     TMn  - turn on tracing at merge n
     TWn - trace merge facets when width > n

Precision options:
    Cn   - radius of centrum (roundoff added).  Merge facets if non-convex
     An  - cosine of maximum angle.  Merge facets if cosine > n or non-convex
           C-0 roundoff, A-0.99/C-0.01 pre-merge, A0.99/C0.01 post-merge
    Rn   - randomly perturb computations by a factor of [1-n,1+n]
    Un   - max distance below plane for a new, coplanar halfspace
    Wn   - min facet width for outside halfspace (before roundoff)

Output formats (may be combined; if none, produces a summary to stdout):
    f    - facet dump
    G    - Geomview output (dual convex hull)
    i    - non-redundant halfspaces incident to each intersection
    m    - Mathematica output (dual convex hull)
    o    - OFF format (dual convex hull: dimension, points, and facets)
    p    - vertex coordinates of dual convex hull (coplanars if 'Qc' or 'Qi')
    s    - summary (stderr)

More formats:
    Fc   - count plus redundant halfspaces for each intersection
         -   Qc (default) for coplanar and Qi for other redundant
    Fd   - use cdd format for input (homogeneous with offset first)
    FF   - facet dump without ridges
    FI   - ID of each intersection
    Fm   - merge count for each intersection (511 max)
    FM   - Maple output (dual 2-d or 3-d convex hull)
    Fn   - count plus neighboring intersections for each intersection
    FN   - count plus intersections for each halfspace
    FO   - options and precision constants
    Fp   - dim, count, and intersection coordinates
    FP   - nearest halfspace and distance for each redundant halfspace
    FQ   - command used for qhalf
    Fs   - summary: #int (8), dim, #halfspaces, #non-redundant, #intersections
                      output: #non-redundant, #intersections, #coplanar
                                  halfspaces, #non-simplicial intersections
                    #real (2), max outer plane, min vertex
    Fv   - count plus non-redundant halfspaces for each intersection
    Fx   - non-redundant halfspaces

Geomview output (2-d, 3-d and 4-d; dual convex hull)
    Ga   - all points (i.e., transformed halfspaces) as dots
     Gp  -  coplanar points and vertices as radii
     Gv  -  vertices (i.e., non-redundant halfspaces) as spheres
    Gc   - centrums
    GDn  - drop dimension n in 3-d and 4-d output
    Gh   - hyperplane intersections
    Gi   - inner planes (i.e., halfspace intersections) only
     Gn  -  no planes
     Go  -  outer planes only
    Gr   - ridges

Print options:
    PAn  - keep n largest facets (i.e., intersections) by area
    Pdk:n- drop facet if normal[k] <= n (default 0.0)
    PDk:n- drop facet if normal[k] >= n
    PFn  - keep facets whose area is at least n
    Pg   - print good facets (needs 'QGn' or 'QVn')
    PG   - print neighbors of good facets
    PMn  - keep n facets with most merges
    Po   - force output.  If error, output neighborhood of facet
    Pp   - do not report precision problems

    .    - list of all options
    -    - one line descriptions of all options
    -?   - help with examples
    -V   - version

Up: Home page for Qhull (local)
Up: Qhull manual: contents
To: ProgramsOptionsOutputFormatsGeomviewPrintQhullPrecisionTraceFunctions (local)
To: synopsis • input • outputs • controls • graphics • notes • conventions • options


The Geometry Center Home Page

Comments to: qhull@qhull.org
Created: Sept. 25, 1995 --- Last modified: see top