RuntimeError: Engine::lmax_exceeded -- angular momentum limit exceeded

1

I encountered the title error with below code

        mol_input = str(charge) + " " + str(spinmulti)
        for atom in mol.GetAtoms():
            mol_input += "\n" + atom.GetSymbol() \
                + " " + str(conf.GetAtomPosition(atom.GetIdx()).x)\
                + " " + str(conf.GetAtomPosition(atom.GetIdx()).y)\
                + " " + str(conf.GetAtomPosition(atom.Get  Idx()).z)
        try:
            geo = psi4.geometry(mol_input)

            #psi4.optimize(level, molecule=geo)
            energy, wf_o = psi4.optimize(level, molecule=geo, return_wfn=True)
            # single poinr energy calc.
            energy, wf_e = psi4.energy(level, molecule=geo, return_wfn=True)

            print("\t>>>>> frequency calculation will start, here")
            energy, wf_f = psi4.frequency(level, molecule=geo, return_wfn=True)

the output is below

mol_input = 0 1
I 1.9256644496641202 -0.032917443081063716 0.006780758800296458
C -0.18614923907723008 0.003182045137621483 -0.0006554942795642219
H -0.5910009858970279 -0.4559850791342856 0.9063130861999501
H -0.5622325375580698 1.0294328546154472 -0.0526369195110689
H -0.586281687131795 -0.5437123775377177 -0.8598014312096133
	>>>>> frequency calculation will start, here
RuntimeError: Engine::lmax_exceeded -- angular momentum limit exceeded

a .sdf file I used is below

I actually found this answer

The above one is bit old answer, I expect other counter-measures might exist…

The only way is to recompile psi4?
I found the option in CMakeList

psi4/psi4-1.7/CMakeLists.txt.

option_with_default(MAX_AM_ERI "Maximum angular momentum for integrals. The definition changed with Libint2 but still single integer. Default fine for production." 5)

I actully tried to perform re-installation. But, error happens

$ cmake -S. -Bbuild.new -DMAX_AM_ERI=7 -DLibint2_DIR=/home/xxxx/xxxx/software/psi4/psi4-1.7/build.new/stage/lib/cmake/libint2
...
...
-- Could NOT find Libint2 (missing: Libint2_DIR)
-- Suitable Libint2 could not be located, Building Libint2 7-7-4-7-7-5 instead.
CMake Warning (dev) at /usr/share/cmake-3.24/Modules/ExternalProject.cmake:3071 (message):

$ sudo make -j6 install
[sudo] password for tomohisa:
[  2%] Installing Psi4-detected BLAS/LAPACK
-- Install configuration: "Release"
[ 60%] Built target libxc_external
[ 60%] Built target libint2_external
[ 60%] Built target pybind11_external
[ 60%] Built target lapack_external
[ 80%] Built target gau2grid_external
[ 82%] Performing configure step for 'psi4-core'
loading initial cache file /home/tomohisa/mywin/mywork/minfo/software/psi4/psi4-1.7/build.new/psi4-core-prefix/tmp/psi4-core-cache-Release.cmake
-- Detecting MathOpenMP -- ?OpenMP=ON, ?MKL=, LANG=CXX, C/CXX/Fortran=GNU/GNU/
-- Using LAPACK: /usr/lib/x86_64-linux-gnu/libopenblas.so;...
-- Disabled HDF5
-- Found pybind11: /home/tomohisa/mywin/mywork/minfo/software/psi4/psi4-1.7/build.new/stage/include (found version "2.10.1")
-- Using pybind11: /home/tomohisa/mywin/mywork/minfo/software/psi4/psi4-1.7/build.new/stage/include (version 2.10.1 for Py3.9.10 and 17)
-- Using Python 3.9: /home/tomohisa/mywin/mywork/minfo/burnish/venv_py39_mol-predict/bin/python3.9
CMake Warning (dev) at /usr/share/cmake-3.24/Modules/FindPackageHandleStandardArgs.cmake:438 (message):
  The package name passed to `find_package_handle_standard_args` (libdl) does
  not match the name of the calling package (DL).  This can lead to problems
  in calling code that expects `find_package` result variables (e.g.,
  `_FOUND`) to follow a certain pattern.
Call Stack (most recent call first):
  /home/tomohisa/mywin/mywork/minfo/software/psi4/psi4-1.7/cmake/FindLibX.cmake:46 (find_package_handle_standard_args)
  /home/tomohisa/mywin/mywork/minfo/software/psi4/psi4-1.7/cmake/FindDL.cmake:8 (find_lib_x)
  CMakeLists.txt:74 (find_package)
This warning is for project developers.  Use -Wno-dev to suppress it.

-- Disabled ambit
-- Disabled CheMPS2
-- Disabled dkh
-- Disabled ecpint
-- Disabled libefp
-- Disabled erd
-- Using gau2grid: /home/tomohisa/mywin/mywork/minfo/software/psi4/psi4-1.7/build.new/stage/lib/libgg.a (version 2.0.7)
-- Disabled gdma
-- Libint2Config components requested: CXX_ho;eri_c2_d0_l4;eri_c2_d1_l3;eri_c2_d2_l3;eri_c3_d0_l4;eri_c3_d1_l3;eri_c3_d2_l3;eri_c4_d0_l7;eri_c4_d1_l2;eri_c4_d2_l2;gss;impure_sh;onebody_d0_l4;onebody_d1_l3;onebody_d2_l3
-- Libint2Config components found: CXX_ho;C;eri_c2_d0_l2;eri_c2_d0_l3;eri_c2_d0_l4;eri_c2_d0_l5;eri_c2_d0_l6;eri_c2_d1_l2;eri_c2_d1_l3;eri_c2_d1_l4;eri_c2_d1_l5;eri_c2_d2_l2;eri_c2_d2_l3;eri_c2_d2_l4;eri_c3_d0_l2;eri_c3_d0_l3;eri_c3_d0_l4;eri_c3_d0_l5;eri_c3_d0_l6;eri_c3_d1_l2;eri_c3_d1_l3;eri_c3_d1_l4;eri_c3_d1_l5;eri_c3_d2_l2;eri_c3_d2_l3;eri_c3_d2_l4;eri_c4_d0_l2;eri_c4_d0_l3;eri_c4_d0_l4;eri_c4_d0_l5;eri_c4_d1_l2;eri_c4_d1_l3;eri_c4_d1_l4;eri_c4_d2_l2;eri_c4_d2_l3;g12_d0_l2;g12_d0_l3;g12_d0_l4;g12_d1_l2;g12_d1_l3;g12_d1_l4;gss;impure_sh;onebody_d0_l2;onebody_d0_l3;onebody_d0_l4;onebody_d0_l5;onebody_d0_l6;onebody_d1_l2;onebody_d1_l3;onebody_d1_l4;onebody_d1_l5;onebody_d2_l2;onebody_d2_l3;onebody_d2_l4;shared
CMake Error at CMakeLists.txt:167 (find_package):
  Found package configuration file:

    /home/tomohisa/mywin/mywork/minfo/software/psi4/psi4-1.7/build.new/stage/lib/cmake/libint2/libint2-config.cmake

  but it set Libint2_FOUND to FALSE so package "Libint2" is considered to be
  NOT FOUND.


-- Configuring incomplete, errors occurred!

I need help…

2

If you are on linux, there are higher AM packages on conda. You can install with:

conda install psi4/label/dev::psi4 -c psi4/label/testing -c conda-forge

Otherwise, since it seems your optimization is running fine, could try finite-difference hessians for the frequency,

psi4.frequency(..., dertype=1)
3

Thanks for the advice. I tested the second method, it works.

4

I did as you suggest and my installation was upgraded, but there was no increase in the max AM in the updated version. Did I somehow get the wrong update? Thanks in advance for any suggestions.

5

Was the correct version of libint2 installed? Check your conda list libint2, you should get:

# Name                    Version                   Build  Channel
libint2                   2.7.2                h2fe1556_1    psi4/label/testing

What method/basis are you trying to run? Energy/gradient/hessian? IIRC, the high AM libint2 package doesn’t increase the max AM for Hessian integrals by much, but psi4 only uses analytic Hessians for HF and LDA functionals, so that’s not usually a big deal. If you want high AM HF Hessians, I would recommend using finite-differences by using the dertype=1 keyword like above. (depending on basis set you might still need the higher AM libint2 package from psi4/label/testing)

6

No, I seem to have version 2.7.1. I am doing CCSD(T) single point energies with the 6ZaP and 7ZaP basis sets, which have max AM=6 and 7, respectively. Would the 2.7.2 version of libint2 help with this! Again, thanks for your help.

7

Yeah, that build goes up to AM=7 for energies, which should have you covered.
Try installing to a fresh conda environment and pinning libint2=2.7.2 in the conda install command. Also, make sure you have python 3.8 or 3.10 (not 3.9 or 3.11).

8

What worked for me was, as a precaution, to do a fresh reinstall of psi4 1.8 and
then do conda install psi4/label/dev::psi4 -c psi4/label/testing -c conda-forge. Doing
this updated libint2 to version 2.7.2 and a test case with AM=6 ran fine.

9

I am trying the solution mentioned by jamont to enter “conda install psi4/label/dev[…] conda-forge”. Conda finds that libint2 needs to be newly installed. But conda has failed at Download and Extract of that exact package four separate times now, staying at 0% even while other conda packages are downloading fine.

Now I am trying to change “optimize(‘b3lyp-d3’)” to “optimize(‘b3lyp-d3’,dertype=1)” but the calculation is still crashing with “lmax_exceeded” error after the initial energy calculation but before the gradient calculation.

psi4TestOutput.dat (50.1 KB)
psi4TestInput.dat (448 Bytes)
TACHCr_2xC4.xyz (2.7 KB)

10

Does conda give any errors at all? Is it trying to install psi4/label/testing/linux-64::libint2-2.7.2-h2fe1556_1?
Note that the high AM build of libint2 is almost 500 MB, it could possibly take a while to download depending on your connection.

As another option, it is possible to manually download and conda install the linux-64/libint2-2.7.2-h2fe1556_1.tar.bz2 file from Files :: Anaconda.org.

Now I am trying to change “optimize(‘b3lyp-d3’)” to “optimize(‘b3lyp-d3’,dertype=1)” but the calculation is still crashing with “lmax_exceeded” error after the initial energy calculation but before the gradient calculation.

This doesn’t work because optimize already uses analytic gradients, which is what dertype=1 specifies. It’s possible to use dertype=0 (gradients by finite-difference of energies), but I would really recommend trying to get the high AM libint2 package installed, as an optimization with dertype=0 will be very slow.

11

Thanks Philip! libint2 eventually installed when I tried a bit later. Some combination of it being a large package, or its server not communicating well, was causing hangups in the installation.

After installing the latest libint2, my calculation proceeded fine.