CODEWITHSUNDEEP
pythonnumpycross-productvectormath
Skyler
Skyler

Reputation: 420

Rollaxis error from numpy cross product in Python

I have been trying to determine the source of my error for this simple script which takes a numpy.array as input and produces a new lattice from the dataset

def reciprocalLat(lattice):
    for i,a in enumerate(lattice):
    print a
        b[i]=numpy.cross(a[(i+1)%3],a[(i+2)%3],axis=0)
#/numpy.dot(a[i],numpy.cross(a[(i+1)%3],a[(i+1)%3]),0)

When I try vary my lattice or even use stripped down examples and multiple different ways

like

print numpy.cross(lat[(1)%3],lat[(2)%3],axis=0)

or

print numpy.cross(lat[(1)%3],lat[(2)%3])

I just get this error 

ValueError: rollaxis: axis (0) must be >=0 and < 0

What is rollaxis doing in this problem and, what is it doing in the capacity of this problem and what exactly am I setting when I assign a value. How do I fix this error (how can something be less than 1 AND great than or equal to 1?)

My test matrix has been:

[['4.7480001450' '-2.3740000725' '0.0000000000']
 ['0.0000000000' '4.1118887427' '0.0000000000']
 ['0.0000000000' '0.0000000000' '15.4790000916']]

Upvotes: 1

Views: 1222

Answers (1)

hpaulj
hpaulj

Reputation: 231345

With the newer version of np.cross (which uses rollaxis instead of swap), I can produce this error with:

In [663]: np_cross.cross(lat[0,0],lat[0,1],axis=0)
---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-663-ee756043fbb9> in <module>()
----> 1 np_cross.cross(lat[0,0],lat[0,1],axis=0)

/home/paul/mypy/np_cross.py in cross(a, b, axisa, axisb, axisc, axis)
     96     b = asarray(b)
     97     # Move working axis to the end of the shape
---> 98     a = rollaxis(a, axisa, a.ndim)
     99     b = rollaxis(b, axisb, b.ndim)
    100     msg = ("incompatible dimensions for cross product\n"

/usr/lib/python3/dist-packages/numpy/core/numeric.py in rollaxis(a, axis, start)
   1340     msg = 'rollaxis: %s (%d) must be >=0 and < %d'
   1341     if not (0 <= axis < n):
-> 1342         raise ValueError(msg % ('axis', axis, n))
   1343     if not (0 <= start < n+1):
   1344         raise ValueError(msg % ('start', start, n+1))

ValueError: rollaxis: axis (0) must be >=0 and < 0

That is, when passing a scalar (or 0d) array to cross you get this rollaxis error. So make sure that you are passing vectors to cross (i.e. at least 1d arrays).

For example if latice is 2d

for i,a in enumerate(lattice):
    print a
    b[i]=numpy.cross(a[(i+1)%3],a[(i+2)%3],axis=0)

then a is 1d, and a[1] is a scalar.

Is this what your goal?

In [675]: lat
Out[675]: 
array([[  4. ,  -2.3,   0. ],
       [  0. ,   4.1,   0. ],
       [  0. ,   0. ,  15. ]])

In [676]: np.vstack([np_cross.cross(lat[(i+1)%3],lat[(i+2)%3]) for i,a in enumerate(lat)])
Out[676]: 
array([[ 61.5,   0. ,   0. ],
       [ 34.5,  60. ,  -0. ],
       [ -0. ,   0. ,  16.4]])

I was puzzled about your use of %3, but from your comment you are trying to do b[0,:] = cross(lat[1,:], lat[2,:]) etc.

But cross itself is doing that kind of permuted pairing. From the older cross:

        x = a[1]*b[2] - a[2]*b[1]
        y = a[2]*b[0] - a[0]*b[2]
        z = a[0]*b[1] - a[1]*b[0]

Upvotes: 2

Related Questions