Divergences for computing exp()

I am working on a multivariate hierarchical model where I need to compute exp of sampled values (the sampled parameters are on the log scale).
However, computing the exp() gives many divergences. The issue is not with sampling the parameters, but computing the exponential of theta in the transformed parameters block.

How do I overcome the divergences? Or is there something incorrect with the way the exponential is computed?

 data {
      int<lower=1> J;    
      int<lower =1> K;   
      vector[K] mu_a;           
  matrix[J,K] a;
   vector<lower = 0>[K] tau;       // sd of b
  cholesky_factor_corr[K] Omega;
  matrix[K,J] z;                  //standard normal variates
 transformed parameters {
         matrix[J,K] b;
         matrix[J,K] theta_raw;
        b = (diag_pre_multiply(tau, Omega)*z)';     //covariance matrix
            theta_raw = exp(a + b);              
model {
         tau ~ cauchy(0,1);  
         Omega ~ lkj_corr_cholesky(2); 
         to_vector(z) ~ normal(0,1);        
       for(j in 1:J){
          a[j,] ~ normal(mu_a, 1); 

The R code used is

stan_data <- list(J = J, 
                    K = 8, 
                  mu_a = c(log(1.4e6), log(6e-11), log(6),log(5), log(1e4), log(3),log(4e4),log(1e-4)),                  
fit <- stan("model_name.stan",          
                  data = stan_data,
                  chains = 1, iter = 1000
                  control = list(adapt_delta = 0.8, max_treedepth = 10),