diff --git a/src/opt/mpi/include/rarray.cc.h b/src/opt/mpi/include/rarray.cc.h
index 3e3002a..a952704 100644
--- a/src/opt/mpi/include/rarray.cc.h
+++ b/src/opt/mpi/include/rarray.cc.h
@@ -212,6 +212,9 @@ namespace CNORXZ
 	template <class Index1, class Index2>
 	void RCArray<T>::load(const Sptr<Index1>& i1, const Sptr<Index2>& i2) const
 	{
+	    VCHECK(i1->lex());
+	    VCHECK(i2->lex());
+	    /*
 	    const SizeT rsize = getRankedSize(mGeom);
 	    if(mMap.size() != rsize){
 		mMap.resize(rsize);
@@ -228,6 +231,7 @@ namespace CNORXZ
 		if(recvr == getRankNumber()) {  }
 	    }, pos(i1), pos(i2) ) );
 	    // MPI_Sendrecv()!!!
+	    */
 	}
 	
     } // namespace mpi
diff --git a/src/opt/mpi/tests/CMakeLists.txt b/src/opt/mpi/tests/CMakeLists.txt
index 4de0e9b..ec733b6 100644
--- a/src/opt/mpi/tests/CMakeLists.txt
+++ b/src/opt/mpi/tests/CMakeLists.txt
@@ -13,3 +13,7 @@ add_dependencies(mpirautest cnorxz cnorxzmpi test_lib)
 target_link_libraries(mpirautest ${GTEST_BOTH_LIBRARIES} ${CMAKE_THREAD_LIBS_INIT} ${MPI_LIBS} cnorxz cnorxzmpi test_lib)
 set(MPI_TEST_COMMAND mpirun -n 4 mpirautest)
 add_test(NAME mpirautest COMMAND ${MPI_TEST_COMMAND})
+
+add_executable(riexp rindex_exp.cc)
+add_dependencies(riexp cnorxz cnorxzmpi test_lib)
+target_link_libraries(riexp ${GTEST_BOTH_LIBRARIES} ${CMAKE_THREAD_LIBS_INIT} ${MPI_LIBS} cnorxz cnorxzmpi test_lib)
diff --git a/src/opt/mpi/tests/rindex_exp.cc b/src/opt/mpi/tests/rindex_exp.cc
new file mode 100644
index 0000000..736baaf
--- /dev/null
+++ b/src/opt/mpi/tests/rindex_exp.cc
@@ -0,0 +1,189 @@
+// -*- C++ -*-
+
+// experiment...
+
+#include <cstdlib>
+#include <iostream>
+
+#include "cnorxz.h"
+#include "cnorxz_mpi.h"
+#include "test_numbers.h"
+#include "rrange.cc.h"
+
+namespace
+{
+    using namespace CNORXZ;
+    using Test::Numbers;
+    using namespace CNORXZ::mpi;
+
+    class Env
+    {
+    public:
+	
+	Env()
+	{
+	    CXZ_ASSERT(getNumRanks() == 4, "exptected 4 ranks");
+	    Vector<Int> xs(12);
+	    Vector<Int> ts(16);
+	    for(SizeT i = 0; i != xs.size(); ++i){
+		const Int x = static_cast<Int>(i) - static_cast<Int>(xs.size()/2);
+		xs[i] = x;
+	    }
+	    for(SizeT i = 0; i != ts.size(); ++i){
+		const Int t = static_cast<Int>(i) - static_cast<Int>(ts.size()/2);
+		ts[i] = t;
+	    }
+	    mSRange = CRangeFactory(4).create();
+	    mXRange = URangeFactory<Int>(xs).create();
+	    mTRange = URangeFactory<Int>(ts).create();
+	    Vector<RangePtr> rs { mTRange, mXRange, mXRange, mXRange };
+	    mGRange = YRangeFactory(rs).create();
+	    RangePtr g1 = CRangeFactory(1).create();
+	    RangePtr g2 = CRangeFactory(2).create();
+	    Vector<RangePtr> gs { g2, g1, g1, g2 };
+	    mGeom = YRangeFactory(gs).create();
+	    mRRange = rrange(mGRange, mGeom);
+	}
+
+	RangePtr mSRange;
+	RangePtr mXRange;
+	RangePtr mTRange;
+	RangePtr mGRange;
+	RangePtr mGeom;
+	RangePtr mRRange;
+    };
+}
+
+
+int main(int argc, char** argv)
+{
+    MPI_Init(&argc, &argv);
+
+    Env env;
+    const SizeT myrank = getRankNumber();
+    const SizeT Nranks = getNumRanks();
+    
+    typedef UIndex<Int> UI;
+    typedef MIndex<UI,UI,UI,UI> LocI;
+    typedef MIndex<CIndex,CIndex,CIndex,CIndex> RankI;
+    RIndex<LocI,RankI> rgi(env.mRRange);
+    RIndex<LocI,RankI> rgj(env.mRRange);
+    LocI gi(env.mGRange);
+    LocI gj(env.mGRange);
+    RankI ri(env.mGeom);
+    constexpr auto C0 = CSizeT<0> {};
+    constexpr auto C2 = CSizeT<2> {};
+    constexpr auto C3 = CSizeT<3> {};
+
+    const SizeT LSize = env.mRRange->sub(1)->size();
+    
+    const SizeT blocks = env.mSRange->size();
+    Vector<Double> data(LSize*blocks);
+    Vector<Double> buf;
+    Vector<Double*> map(env.mRRange->size(),nullptr);
+    for(SizeT i = 0; i != data.size(); ++i){
+	data[i] = static_cast<Double>(LSize*myrank*blocks+i);
+    }
+    Vector<Vector<SizeT>> cnt(Nranks);
+    for(auto& c: cnt){
+	c.resize(Nranks);
+    }
+    Vector<Vector<Double>> sendbuf(Nranks);
+    for(auto& sb: sendbuf){
+	sb.reserve(data.size());
+    }
+    // First loop: setup send buffer
+    for(rgi = 0, gi = 0; rgi.lex() != rgi.lmax().val(); ++rgi, ++gi){
+	gj = gi.lex();
+	*gj.pack()[C0] = (gj.pack()[C0]->lex() + 1) % gj.pack()[C0]->lmax().val();
+	*gj.pack()[C2] = (gj.pack()[C2]->lex() + 1) % gj.pack()[C2]->lmax().val();
+	*gj.pack()[C3] = (gj.pack()[C3]->lex() + 1) % gj.pack()[C3]->lmax().val();
+	gj();
+	rgj = gj.lex();
+	if(rgi.rank() != rgj.rank()){
+	    if(rgj.rank() == myrank){
+		// j = sender
+		const Double* d = data.data()+rgj.local()->pos()*blocks;
+		assert(static_cast<Int>(*d) % 4 == 0);
+		sendbuf[rgi.rank()].insert(sendbuf[rgi.rank()].end(), d, d+blocks);
+	    }
+	    ++cnt[rgi.rank()][rgj.rank()];
+	}
+    }
+    
+    // Initialize target buffer:
+    SizeT bufsize = 0;
+    for(const auto& c: cnt[myrank]){
+	bufsize += c*blocks;
+    }
+    buf.resize(bufsize);
+    
+    // Transfer data:
+    for(SizeT o = 0; o != Nranks; ++o){
+	const SizeT dstr = (myrank + o) % Nranks;
+	const SizeT srcr = (myrank - o + Nranks) % Nranks;
+	SizeT off = 0;
+	for(SizeT p = 0; p != srcr; ++p){
+	    off += cnt[myrank][p];
+	}
+	MPI_Status stat;
+	MPI_Sendrecv(sendbuf[dstr].data(), cnt[dstr][myrank]*blocks, MPI_DOUBLE, dstr, 0,
+		     buf.data()+off*blocks, cnt[myrank][srcr]*blocks, MPI_DOUBLE, srcr, 0,
+		     MPI_COMM_WORLD, &stat);
+    }
+    Vector<Vector<SizeT>> ext = cnt;
+
+    // Second loop: Assign map to target buffer positions:
+    for(auto& c: cnt){
+	c = Vector<SizeT>(Nranks,0);
+    }
+    for(rgi = 0, gi = 0; rgi.lex() != rgi.lmax().val(); ++rgi, ++gi){
+	gj = gi.lex();
+	*gj.pack()[C0] = (gj.pack()[C0]->lex() + 1) % gj.pack()[C0]->lmax().val();
+	*gj.pack()[C2] = (gj.pack()[C2]->lex() + 1) % gj.pack()[C2]->lmax().val();
+	*gj.pack()[C3] = (gj.pack()[C3]->lex() + 1) % gj.pack()[C3]->lmax().val();
+	gj();
+	rgj = gj.lex();
+
+	if(rgi.rank() == myrank){
+	    if(rgj.rank() != myrank){
+		SizeT off = 0;
+		for(SizeT p = 0; p != rgj.rank(); ++p){
+		    off += ext[myrank][p];
+		}
+		map[rgj.pos()] = buf.data()+off*blocks + cnt[rgi.rank()][rgj.rank()]*blocks;
+		++cnt[rgi.rank()][rgj.rank()];
+	    }
+	    map[rgi.pos()] = data.data() + rgi.local()->pos()*blocks;
+	}
+    }
+
+    // Third loop: Check:
+    for(rgi = 0, gi = 0; rgi.lex() != rgi.lmax().val(); ++rgi, ++gi){
+	gj = gi.lex();
+	*gj.pack()[C0] = (gj.pack()[C0]->lex() + 1) % gj.pack()[C0]->lmax().val();
+	*gj.pack()[C2] = (gj.pack()[C2]->lex() + 1) % gj.pack()[C2]->lmax().val();
+	*gj.pack()[C3] = (gj.pack()[C3]->lex() + 1) % gj.pack()[C3]->lmax().val();
+	gj();
+	rgj = gj.lex();
+
+	if(rgi.rank() == myrank){
+	    assert(map.data()[rgj.pos()] != nullptr);
+
+	    const Double vn = *map[rgj.pos()]/blocks;
+	    const SizeT xp = static_cast<SizeT>(vn);
+	    const SizeT orank = xp / env.mRRange->sub(1)->size();
+	    assert(env.mRRange->sub(1)->size() == 16*12*12*12/4);
+	    if(myrank == 0){
+		std::cout << " pos = " << rgj.pos() << " , val = " << *map[rgj.pos()]
+			  << " , val_norm = " << vn << " , origin rank =  "
+			  << orank << std::endl;
+	    }
+	    assert(orank == rgj.rank());
+	    assert(vn == rgj.pos());
+	}
+    }
+    
+    MPI_Finalize();
+    return 0;
+}