optimize_acq returning same 10 candidates for next iteration

[ankush-kumar-mishra]

I am using BoTorch, for a 9D problem with constraints, that first 2 variable must sum to 1.0 and the next 6 variables must sum to 1.0 ). I am using qUCB with q= 10.

My Code:

epsilon = 1e-4
inequality_constraints = [
    (torch.tensor([0,1]), torch.tensor([1.0,1.0]), 1.0 + epsilon),  
    (torch.tensor([0,1]), torch.tensor([-1.0,-1.0]), -(1.0 + epsilon)), 
    (torch.tensor([2,3,4,5,6,7]), torch.tensor([1.0,1.0,1.0,1.0,1.0,1.0]), 1.0 + epsilon), 
    (torch.tensor([2,3,4,5,6,7]), torch.tensor([-1.0,-1.0,-1.0,-1.0,-1.0,-1.0]), -(1.0 + epsilon)), 
]

NUM_RESTARTS = 20
RAW_SAMPLES = 4096
MC_SAMPLES = 128

qUCB = qUpperConfidenceBound(
            model=model,
            beta=100.0,
            sampler=sampler
        )

candidates, acq_value = optimize_acqf(
            acq_function=qUCB,
            bounds=standard_bounds,
            q=batch_size,
            num_restarts=NUM_RESTARTS,
            raw_samples=RAW_SAMPLES,
            inequality_constraints=inequality_constraints,
            
        )
sampler = SobolQMCNormalSampler(sample_shape=(torch.Size([MC_SAMPLES])))

matern_kernel = ScaleKernel(MaternKernel(
            nu = 2.5,
            ard_num_dims = len(lower_limit),
            lengthscale_constraint=Interval(0.01, 0.5)
            )   
            )

model = SingleTaskGP(train_x_norm, train_obj, covar_module=matern_kernel)
model = model.to(**tkwargs)
mll = ExactMarginalLogLikelihood(model.likelihood, model)
fit_gpytorch_mll(mll)

The candidates that are being returned after optimizing_acq are all the same candidates. I am not sure why this is happening. It will be great to get some insights on this.