Cannot run UDM lclr with 256 or more components in iqtree

[pnatsi]

Hi,

I am trying to use the UDM lclr models with iqtree on a toy alignment of 100 species and 500 sites.

Our computer has 1.5TB RAM and 40 cores.

However, iqtree runs successfully only with up to 128 components, and not with 256 or more.
The command I use is:

iqtree -s aln1_100species_500sites.aln -mdef ../tools/EDCluster/Distributions/hogenom/udm_hogenom_0256_lclr_iqtree.nex -m Poisson+UDM0256LCLR -nt AUTO

and the error I get is:

ERROR: STACK TRACE FOR DEBUGGING:
ERROR: Not enough memory, allocation of -15032385536 bytes failed (bad_alloc)

Is this a known issue?
Is it possible to use the PMSF approximation in iqtree in combination with the UDM to overcome this?

Paschalis

[dschrempf]

Hi, thanks for you interest in EDM models and thanks for your report.

This was (is?) a long standing bug (integer overflow) in IQ-TREE.

Which version of IQ-TREE are you using? Can you try using the latest release of IQ-TREE2?

If the issue persists, please let me know. I do have a personal IQ-TREE version with this bug fixed, and I can file an upstream bug report.

[pnatsi]

Thanks for your answer. I was indeed using an older version of IQ-TREE (2.0.3) However, even with the latest release (2.1.2) I get the same problem. Can I find the version that you also have somewhere? Or maybe you can file a bug report?

…

________________________________ From: Dominik Schrempf ***@***.***> Sent: Wednesday, April 14, 2021 9:17 AM To: dschrempf/EDCluster ***@***.***> Cc: Paschalis Natsidis ***@***.***>; Author ***@***.***> Subject: Re: [dschrempf/EDCluster] Cannot run UDM lclr with 256 or more components in iqtree (#1) Hi, thanks for you interest in EDM models and thanks for your report. This was (is?) a long standing bug (integer overflow) in IQ-TREE. Which version of IQ-TREE are you using? Can you try using the latest release of IQ-TREE2<https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fiqtree%2Fiqtree2%2Freleases%2Ftag%2Fv2.1.2&data=04%7C01%7C%7C56426abd9b7f4ec15b6f08d8ff0d0223%7C84df9e7fe9f640afb435aaaaaaaaaaaa%7C1%7C0%7C637539778616982908%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=InMWMNe%2BU24Vvrwyhb2E1lc9KKpgzhXS5MfCaydrRZ4%3D&reserved=0>? If the issue persists, please let me know. I do have a personal IQ-TREE version with this bug fixed, and I can file an upstream bug report. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub<https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fdschrempf%2FEDCluster%2Fissues%2F1%23issuecomment-819259177&data=04%7C01%7C%7C56426abd9b7f4ec15b6f08d8ff0d0223%7C84df9e7fe9f640afb435aaaaaaaaaaaa%7C1%7C0%7C637539778616992862%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=LtOnumFziXjyDiKxVdFpwvNW1Ar7qfg63DRqULin7HM%3D&reserved=0>, or unsubscribe<https://emea01.safelinks.protection.outlook.com/?url=https%3A%2F%2Fgithub.com%2Fnotifications%2Funsubscribe-auth%2FAEV4V75ZZRHK3FNBJADVSOLTIUXQHANCNFSM42WYKA3A&data=04%7C01%7C%7C56426abd9b7f4ec15b6f08d8ff0d0223%7C84df9e7fe9f640afb435aaaaaaaaaaaa%7C1%7C0%7C637539778616992862%7CUnknown%7CTWFpbGZsb3d8eyJWIjoiMC4wLjAwMDAiLCJQIjoiV2luMzIiLCJBTiI6Ik1haWwiLCJXVCI6Mn0%3D%7C1000&sdata=0%2BRa3HCc1fxZ12bUA8sML3YaSMirlP6CUdiwS4RAwEc%3D&reserved=0>.

[dschrempf]

I have submitted a bug report, see iqtree/iqtree2#24, and also Cibiv/IQ-TREE#171. An (older, v2.0.4) but working version can be downloaded now like so:

git clone https://github.com/Cibiv/IQ-TREE.git
cd IQ-TREE
git checkout linked_models

Compilation is explained here: http://www.iqtree.org/doc/Compilation-Guide

I hope that the issue gets fixed soon, and that you can work with the workaround for now!