Bug in interpolate.dmrg_cross

[MazenAli]

Describe the bug
DMRG cross fails due to shape mismatch.

To Reproduce
Too long to include everything here.

This is the call

cross_func = lambda idx: torch.tensor(self._cross_func(idx, eval_point)[i, j])
tensor_shape = [2]*num_sites
tt = tn.interpolate.dmrg_cross(cross_func, tensor_shape, verbose=True)

This is the printed output and traceback:

Sweep 1: 
	LR supercore 1,2
		number evaluations 8
		rank updated 2 -> 2, local error 1.566035e+02
	LR supercore 2,3
		number evaluations 16
		rank updated 2 -> 4, local error 8.495978e-01
	LR supercore 3,4
		number evaluations 32
		rank updated 2 -> 5, local error 8.368856e-01
	LR supercore 4,5
		number evaluations 40
		rank updated 2 -> 5, local error 1.144697e+00
	LR supercore 5,6
		number evaluations 20
		rank updated 2 -> 4, local error 4.816946e-16
	RL supercore 5,6
		number evaluations 20
		rank updated 4 -> 2, local error 3.952443e-16
	RL supercore 4,5
		number evaluations 40

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
Cell In[15], line 1
----> 1 J = get_tensor_networks()
      2 print(J)

File /app/tn.py:318, in TN.get_tensor_networks(self)
    316     cross_func = lambda idx: torch.tensor(self._cross_func(idx, eval_point)[i, j])
    317     tensor_shape = [2]*num_sites
--> 318     tt = tn.interpolate.dmrg_cross(cross_func, tensor_shape, verbose=True)
    320     col.append(tt)
    322 row.append(col)

File /usr/local/lib/python3.12/site-packages/torchtt/interpolate.py:693, in dmrg_cross(function, N, eps, nswp, x_start, kick, dtype, device, eval_vect, rmax, verbose)
    690 if radd > 0:
    691     U = tn.cat(
    692         (U, tn.zeros((U.shape[0], radd), dtype=dtype, device=device)), 1)
--> 693     U = U @ Rtemp.T
    694     V = V.t()
    696 # print('kkt new',tn.linalg.norm(supercore-U@V))
    697 # compute err (dx)

RuntimeError: mat1 and mat2 shapes cannot be multiplied (10x4 and 5x4)

[ion-g-ion]

Hi,
Thanks for reporting it. Since I could not reproduce it (it seems that it appears only for some functions), I did some blind changes in order to avoid the multiplication error. It is pushed in the latest git version. I hope this solves it...

BR,
Ion