For a more compact instructions, try setting http://vergil.chemistry.gatech.edu/nu-psicode/install-v1.2rc2.html to linux/source/3.6/nightly . The fourth line (print it instead of executing with backticks) is loading various values into your cmake cache. You can always override or extend that command (e.g., {backtick}psi4-path-advisor --gcc{backtick}-objdir-postfix -DMAX_AM_ERI=8 -DENABLE_CheMPS2=OFF)
At the moment, you may need to install psi4-dev package and then force install libint am8 b/c I haven’t rebuilt the latter with the new compilers. But considering Libint is a quite-static C lib, I don’t anticipate interop problems.
Or if you post your cmake error, I can probably help through that route.
There’s a libint AM=8 conda package, but you do have to build Psi4 against it, not just use a psi4 conda package and switch out the libint package. conda install libint am8 -c psi4/label/dev should do the trick. Then pass -DLibint_DIR=/conda_install/envs/conda_env/share/cmake/Libint to your cmake command.
Dear Lori,
I have now run:
conda install libint am8 -c psi4/label/dev
in my home directory.
When I type “ls” though, I don’t see any evidence of a psi4.
I don’t know where else the program might be.
The process was also suspiciously quick, so I worry that it didn’t get installed yet.
Do you have any advice for what I should do?
With best wishes,
Nike
There won’t be a psi4 package (you have to compile psi4 using the am8 libint package) but your libint should have a [am8] somewhere when you’ve activated the environment and run conda list. Post the result of that command, and I can offer a diagnosis.
I guess I was worried about not knowing where to find /psi4, but it turns out I was ignorant off the “conda list” command (I didn’t use conda when I installed the other packages such as CFOUR, MRCC, DIRAC, NECI, etc.).
The installation did seem rather short (a handful of minutes) so I’d be surprised if my Psi4 is working based on just those few minutes of installation. But perhaps the compilation was done on your computer and those few minutes on my computer were just for downloading the pre-compiled binaries.
To clarify, conda packages are all precompiled binaries. So the installation step is performing a “relocatable” installation by substituting the installation prefix in all text and binary files and rewriting the binary files’ rpath property.
Your conda list looks pretty good. You have all psi4’s dependencies and libint is am8. You also have the psi4 package itself. It might be a good idea to uninstall that (conda uninstall psi4 --force) just so that (since it’s in your path) you don’t accidentally run it instead of your compiled psi4.
You can now build your psi4 source code, being careful to include cmake ... -DLibint_DIR=${CONDA_PREFIX}/share/cmake/Libint -DMAX_AM_ERI=8 in your build setup. It should say “Libint 8” in color in the cmake output.
Alternatively, you can install a helper package conda install psi4-dev -c psi4/label/dev and follow these lines (select Linux-source-nightlybuild from http://vergil.chemistry.gatech.edu/nu-psicode/install-v1.2rc2.html) to guide your build, tacking on the -DMAX_AM_ERI=8 at the end of the psi4-path-advisor line outside the backticks.
Dear Lori,
Thank you very much for your tremendous help.
I do not need to modify the code or anything, so I wonder if I can just use the Conda pre-compiled binaries that you say I have correct.
What I mean is, if I can avoid doing the “conda uninstall psi4 --force” and building the psi4 code, and still have AM=8 working, I would be quite delighted to just start doing some 7-zeta calculations.
Regrettably, until this is in place, psi4 needs to get the MAX_AM_ERI from the Libint headers, so you’ll have to build psi4 from source, pointing the build (with -DLibint_DIR) to the am8 libint conda pkg.
Then followed Linux -> Source -> Nightlybuild from the webpage you gave, and did:
git clone https://github.com/psi4/psi4.git && cd psi4
conda create -n p4dev psi4-dev python=3.6 -c psi4/label/dev
conda activate p4dev
`psi4-path-advisor --gcc` -DMAX_AM_ERI=8
cd objdir && make -j`getconf _NPROCESSORS_ONLN`
I did not see Libint 8 in color anywhere in the cmake output.
I also got this error:
CMake Error at simint/CMakeLists.txt:7 (message):
Maximum angular momentum set is greater than I have available (7)
Furthermore I ran out of memory when trying this compilation on the cluster’s head node.
The “interactive” nodes have a 3 hour time limit, and the compilation took around that amount of time (I was waiting for well over an hour when I started getting those memory errors).
Can you give an estimate of roughly how long it will take to compile?
Is the CMake Error I mentioned above, a concern at all if I want to use AM=8 ?
Do you not have pre-compiled binaries with AM=8, that I can just do some tests on right away? I understand that the conda packages are pre-compiled binaries, but you also say that I have to compile directly from source by myself as well. So it seems I’m using conda to get the pre-compiled libint with AM=8, then I have to delete the installed Psi4 and compile it from scratch using the libint=8 package that was just downloaded.
-- LibintConfig missing component: requested AM 8 > 6 detected in /home/nike/.local/easybuild/software/2017/Core/miniconda3/4.3.27/envs/p4dev/lib/libint.so /home/nike/.local/easybuild/software/2017/Core/miniconda3/4.3.27/envs/p4dev/lib/libderiv.so
-- Suitable Libint could not be located, Building Libint instead.
When I run “conda list”, I also don’t see “am8 0.1 0 psi4/label/dev” anymore, so this might be part of the problem.
Maybe it’s because I followed your instruction to run “conda uninstall psi4 --force” before doing:
Ah yes, after the psi4-dev installation, need to replace the am6 libint that that brings with an am8 libint through conda install libint am8 -c psi4/label/dev. Check that [am8] now appears in conda list. Then run the psi4-path-advisor line (removing the incomplete objdir/ from the previous build beforehand). Then build. Without building libint internally (which indeed can take several hours), psi4 builds in a few minutes and w/i 30 min on travis-ci, so shouldn’t take too long.
@jturney took another look, and drop in replacement is tricky b/c array dims are expected to be available at compile-time. If the above doesn’t work for you, I’ll build a psi4+am8 conda package you can use. I won’t build it regularly, but hopefully you and others aren’t at the intersection of high AM and bleeding-edge features.
Now you are telling me “after the psi4-dev installation, need to replace the am6 libint that that brings with an am8 libint through conda install libint am8 -c psi4/label/dev” but that command seems to be what I was already doing in Step #2.
Yes it would be helpful if there can be a psi4+am8 conda package which can be installed simpler, I do NOT need it to be updated regularly because I’m only using it to generate integrals (and hopefully FCIDUMP files and some post-HF using the MRCC interface).
I do wish to keep trying the installation though, but I think something in the above steps #1-9 is not right. At present I’ve never used Psi4 before, so I am just experimenting to see if it can do what I need.
I have now tried exactly that and at step 8 there was “psi4-path-advisor: error: unrecognized arguments: -DMAX_AM_ERI=8”
One interesting thing in the output below is that I was asked if I really wanted libint to be “downgraded”, but I clicked “yes” because the thing it was being downgraded to said am8 in it.
this forum doesn’t do backticks well. so -DMAX_AM_ERI=8 is an argument to cmake (which is the result of the psi4-path-advisor command), not to psi4-path-advisor itself. So (8) will be
Ok, it took me a long time to figure out how to re-start this procedure after already having done the first few steps and then having logged out part-way through and come back on a different day.
There was a command I had (which was a very lucky guess!) to add in order to continue where I left off.
To organize better for myself, the “current” state-of-the-art instructions are:
cd ~/.local/easybuild/software/2017/Core/miniconda3/4.3.27/bin/
psi4 --test
At step 8, I get this:
-- Detecting CXX compile features - done
CMake Error at CMakeLists.txt:13 (include):
include could not find load file:
psi4OptionsTools
CMake Error at CMakeLists.txt:14 (include):
include could not find load file:
TestRestrict
CMake Error at CMakeLists.txt:17 (include):
include could not find load file:
custom_color_messages
CMake Error at CMakeLists.txt:18 (test_restrict):
Unknown CMake command "test_restrict".
-- Configuring incomplete, errors occurred!
So I am still struggling.
At this point I wonder if you could indeed make that psi4 conda package with am8, which I can install much easier, because this is taking me longer than I thought (I started this thread on 14 June, but actually I asked a question about the same thing back on 17 July 2017 Libint max angular momentum error , so it’s actually been 2 days short of 1 full year since I first attempted installing Psi4. It does NOT need to be updated nightly, I just need it once, I will not be using any features that are presently in flux, I just need to calculate the integrals and dump them to an FCIDUMP!
I guess I would still learn need to learn how to compile it myself in order to work my way up to AM11, since it appears there’s reluctance to make these bigg(er) conda packages. So I would still be interested in how to get past the above error even if one day an easy conda package were made available to do am8 with psi4 using just one “conda install” command. Would the rest of the procedure still be the same even with am9 ?
Dear Lori,
Regarding the error I reported in my most recent reply, is it reproducible, or do you think it’s something only I am getting?
With best wishes!
Nike
Dear Lori,
I have completed the calculations up to 6-zeta in CFOUR and wish to continue to 7-zeta and 8-zeta, for which I would like to compile Psi4 with am=8. I am getting an error at Step 8 of the procedure in my 15 June reply, is it something you could help me with?
With best wishes!
Nike