diff --git a/lib/gpu/lal_amoeba.cu b/lib/gpu/lal_amoeba.cu
index adcff0e648..c4f146a7c9 100644
--- a/lib/gpu/lal_amoeba.cu
+++ b/lib/gpu/lal_amoeba.cu
@@ -116,9 +116,13 @@ _texture( q_tex,int2);
   }                                                                         \
   if (offset==0 && ii<inum) {                                               \
     numtyp4 f, fp;                                                          \
-    f.x  = _fieldp[0]; f.y  = _fieldp[0]; f.z  = _fieldp[2];                \
-    fp.x = _fieldp[3]; fp.y = _fieldp[4]; fp.z = _fieldp[5];                \
+    f.x = _fieldp[0];                                                       \
+    f.y = _fieldp[1];                                                       \
+    f.z = _fieldp[2];                                                       \
     fieldp[ii] = f;                                                         \
+    fp.x = _fieldp[3];                                                      \
+    fp.y = _fieldp[4];                                                      \
+    fp.z = _fieldp[5];                                                      \
     fieldp[ii+inum] = fp;                                                   \
   }
 
@@ -152,7 +156,7 @@ _texture( q_tex,int2);
     tep[i]=t;                                                               \
   }
 
-#define store_answers_fieldp(_fieldp, ii, inum,tid, t_per_atom, offset, i,  \
+#define store_answers_fieldp(_fieldp, ii, inum, tid, t_per_atom, offset, i, \
                              fieldp)                                        \
   if (t_per_atom>1) {                                                       \
     for (unsigned int s=t_per_atom/2; s>0; s>>=1) {                         \
@@ -166,9 +170,13 @@ _texture( q_tex,int2);
   }                                                                         \
   if (offset==0 && ii<inum) {                                               \
     numtyp4 f, fp;                                                          \
-    f.x  = _fieldp[0]; f.y  = _fieldp[0]; f.z  = _fieldp[2];                \
-    fp.x = _fieldp[3]; fp.y = _fieldp[4]; fp.z = _fieldp[5];                \
+    f.x = _fieldp[0];                                                       \
+    f.y = _fieldp[1];                                                       \
+    f.z = _fieldp[2];                                                       \
     fieldp[ii] = f;                                                         \
+    fp.x = _fieldp[3];                                                      \
+    fp.y = _fieldp[4];                                                      \
+    fp.z = _fieldp[5];                                                      \
     fieldp[ii+inum] = fp;                                                   \
   }
 
@@ -890,8 +898,7 @@ __kernel void k_amoeba_udirect2b(const __global numtyp4 *restrict x_,
       _fieldp[3] += fip[0];
       _fieldp[4] += fip[1];
       _fieldp[5] += fip[2];
-
-    } // nbor
+    }  // nbor
 
   } // ii<inum
 
@@ -948,4 +955,3 @@ __kernel void k_special15(__global int * dev_nbor,
 
   } // if ii
 }
-
diff --git a/lib/gpu/lal_base_amoeba.cpp b/lib/gpu/lal_base_amoeba.cpp
index 88caec3972..6800288093 100644
--- a/lib/gpu/lal_base_amoeba.cpp
+++ b/lib/gpu/lal_base_amoeba.cpp
@@ -528,13 +528,14 @@ int** BaseAmoebaT::compute_udirect2b(const int ago, const int inum_full, const i
   // copy field and fieldp from device to host (_fieldp store both arrays, one after another)
 
   _fieldp.update_host(_max_fieldp_size*8,false);
-/*  
-  printf("GPU lib: _fieldp size = %d: max fieldp size = %d\n", this->_field.cols(), _max_fieldp_size);
+
+  printf("GPU lib: _fieldp size = %d: max fieldp size = %d\n",
+    this->_fieldp.cols(), _max_fieldp_size);
   for (int i = 0; i < 10; i++) {
     numtyp4* p = (numtyp4*)(&this->_fieldp[4*i]);
     printf("i = %d; field = %f %f %f\n", i, p->x, p->y, p->z);
   }
-*/
+
   return firstneigh; //nbor->host_jlist.begin()-host_start;
 }
 
diff --git a/src/GPU/pair_amoeba_gpu.cpp b/src/GPU/pair_amoeba_gpu.cpp
index d87e35cdf8..6501376dfa 100644
--- a/src/GPU/pair_amoeba_gpu.cpp
+++ b/src/GPU/pair_amoeba_gpu.cpp
@@ -367,10 +367,10 @@ void PairAmoebaGPU::induce()
   dfield0c(field,fieldp);
 
   // reverse comm to sum field,fieldp from ghost atoms to owned atoms
-
+/*
   crstyle = FIELD;
   comm->reverse_comm_pair(this);
-
+*/
 
   // set induced dipoles to polarizability times direct field
 
@@ -778,7 +778,7 @@ void PairAmoebaGPU::dfield0c(double **field, double **fieldp)
 
 void PairAmoebaGPU::udirect2b(double **field, double **fieldp)
 {
-  bool gpu_udirect2b_ready = false;
+  bool gpu_udirect2b_ready = true;
   if (!gpu_udirect2b_ready) {
     PairAmoeba::udirect2b(field, fieldp);
     return;
@@ -815,31 +815,33 @@ void PairAmoebaGPU::udirect2b(double **field, double **fieldp)
   if (!success)
     error->one(FLERR,"Insufficient memory on accelerator");
 
-  // get field and fieldp values from the GPU lib
+  // rebuild dipole-dipole pair list and store pairwise dipole matrices
+  // done one atom at a time in real-space double loop over atoms & neighs
+
+  udirect2b_cpu();
+
+  // accumulate the field and fieldp values from the GPU lib
+  //   field and fieldp may already have some nonzero values from kspace (udirect1)
 
   int nlocal = atom->nlocal;
   double *field_ptr = (double *)fieldp_pinned;
 
   for (int i = 0; i < nlocal; i++) {
     int idx = 4*i;
-    field[i][0] = field_ptr[idx];
-    field[i][1] = field_ptr[idx+1];
-    field[i][2] = field_ptr[idx+2]; 
+    field[i][0] += field_ptr[idx];
+    field[i][1] += field_ptr[idx+1];
+    field[i][2] += field_ptr[idx+2]; 
   }
 
   double* fieldp_ptr = (double *)fieldp_pinned;
   fieldp_ptr += 4*inum;
   for (int i = 0; i < nlocal; i++) {
     int idx = 4*i;
-    fieldp[i][0] = fieldp_ptr[idx];
-    fieldp[i][1] = fieldp_ptr[idx+1];
-    fieldp[i][2] = fieldp_ptr[idx+2];
+    fieldp[i][0] += fieldp_ptr[idx];
+    fieldp[i][1] += fieldp_ptr[idx+1];
+    fieldp[i][2] += fieldp_ptr[idx+2];
   }
-
-  // rebuild dipole-dipole pair list and store pairwise dipole matrices
-  // done one atom at a time in real-space double loop over atoms & neighs
-
-  udirect2b_cpu();
+  
 }
 
 /* ----------------------------------------------------------------------