Minimal working example required

[LucidaLu]

Hello, I am wondering if you can provide a minimal working example of fixed-angle QAOA?

I tried to read this project and the qtensor library, but failed to obtain the expected result.

Here is the code I am trying:

from qtensor import QAOA_energy
import networkx as nx
import json

with open("angles.json", 'r') as fangles:
    ANGLES = json.load(fangles)

G = nx.from_edgelist([[0, 1], [0, 2], [0, 3], [1, 2], [1, 3], [2, 3]])
opt = max(sum(0 if s >> x & 1 == s >> y & 1 else 1 for x, y in G.edges) for s in range(1 << G.number_of_nodes()))

p = 5

gamma, beta = ANGLES['3'][str(p)]['gamma'], ANGLES['3'][str(p)]['beta']

E = QAOA_energy(G, gamma, beta)

print(gamma, '\n', beta, '\n', E, E/opt, opt)

# outputs 
# [0.3595894069747186, 0.7065726026028729, 0.8225565476596165, 1.0047548948734175, 1.154274581864511] 
# [0.6318520293950315, 0.5226524412470716, 0.3897420609614237, 0.2754121638618102, 0.1491700495831238] 
# [2.15570478] [0.5389262] 4

# {"GW computation time (sec)": 1.5017328262329102, "edges": [[0, 1], [0, 2], [0, 3], [1, 2], [1, 3], [2, 3]], "maxcut": 4.0, "GWave": 3.69, "Nvertices": 4, "QAOA_1": 3.4880307849489687, "AR_1": 0.8720076962372422, "WS computation time (sec)": 0.031711578369140625, "QAOA_2": 3.443514444229522, "AR_2": 0.8608786110573805, "QAOA_3": 3.7257461621272516, "AR_3": 0.9314365405318129, "QAOA_4": 3.7994349269828813, "AR_4": 0.9498587317457203, "QAOA_5": 3.9023289956861937, "AR_5": 0.9755822489215484, "QAOA_6": 3.821206750231719, "AR_6": 0.9553016875579298, "QAOA_7": 3.8566061089160772, "AR_7": 0.9641515272290193, "QAOA_8": 3.902939231923547, "AR_8": 0.9757348079808867, "QAOA_9": 3.8698211110331737, "AR_9": 0.9674552777582934, "QAOA_10": 3.909461386351469, "AR_10": 0.9773653465878672, "QAOA_11": 3.8962754265951487, "AR_11": 0.9740688566487872}
# the first item in ALL_3r_graphs.json, saying that QAOA_5=3.9023289956861937

I must have made some stupid mistake, but I just can't see it. I'd really appreciate your help! Thanks!

[danlkv]

Hi! There's a bit of rescaling required. Just divide the angles by np.pi and it should work.