I've got a version of the HH model (via a NESTML file) which seems to be requiring an exponential Euler solver (ie the one commonly used for e^x dependent biologic systems) to most efficiently solve the equations*. From my reading of the NESTML code generator tools source, only rk45 and forward-Euler are supported. The difference between Euler and Exponential Euler doesn't seem that significant, so I wanted to see if I could get the exponential version implemented....however, what's become unclear to me is whether or not this is a limitation of NEST, NESTML or the underlying solvers and how I'd go about getting a different solver implemented. Are there any guides around on how one would implement a new solver and/or information which would help me figure this out.
thx.
jrd
*=forward Euler does work for some (might work with even smaller timestep) cases, but it requires a time step of 0.01ms. I'm comparing with a Brian2 version of the model running with a 0.1ms time step with it's natively supported exponential Euler solver which is what is making me look into other solvers for my NESTML-based code.
Hi,
Thanks for writing in. Just to be clear about terminology: for numerically stiff systems of ODEs, there exists a class of (numerical) solvers called "exponential Euler", such as Exponential Rosenbrock-Euler.
There are also linear sets of ODEs, the solution of which can be calculated algebraically (with pen and paper) and will typically end up containing terms like exp(-dt/tau).
Which of these two did you wish to use? In case of the latter, the good news is that that's already implemented and should be working out-of-the-box. For the former, I would be happy to add support for an ERE (or a similar) solver if it makes sense for your model, or you need it to reproduce results. Please let me know what you would deem a suitable ERE solver, preferably one with a C++ interface and an open-source license.
Cheers! Charl
On Mon, Feb 24, 2025, at 15:45, justin.r.davis@essie.ufl.edu wrote:
I've got a version of the HH model (via a NESTML file) which seems to be requiring an exponential Euler solver (ie the one commonly used for e^x dependent biologic systems) to most efficiently solve the equations*. From my reading of the NESTML code generator tools source, only rk45 and forward-Euler are supported. The difference between Euler and Exponential Euler doesn't seem that significant, so I wanted to see if I could get the exponential version implemented....however, what's become unclear to me is whether or not this is a limitation of NEST, NESTML or the underlying solvers and how I'd go about getting a different solver implemented. Are there any guides around on how one would implement a new solver and/or information which would help me figure this out.
thx.
jrd
*=forward Euler does work for some (might work with even smaller timestep) cases, but it requires a time step of 0.01ms. I'm comparing with a Brian2 version of the model running with a 0.1ms time step with it's natively supported exponential Euler solver which is what is making me look into other solvers for my NESTML-based code. _______________________________________________ NEST Users mailing list -- users@nest-simulator.org To unsubscribe send an email to users-leave@nest-simulator.org
Yes, it's the former I'm looking to implement. Specifically, I'm trying to replicate some Brian2 results which implement the exponential integrator via: https://brian2.readthedocs.io/en/stable/_modules/brian2/stateupdaters/expone...
Which I believe is just a 1st order integrator (as opposed to the higher order ERE methods) such as shown in Eqn 8 (I assume the existing forward euler looks something like Eqn 4): https://en.wikipedia.org/wiki/Numerical_methods_for_ordinary_differential_eq...
Assuming the processor is already splitting the equations into linear/non-linear parts (ie eqn 7), I don't think a separate solver library would be needed for the 1st order integrator.
jrd
-----Original Message----- From: Charl Linssen nest-users@turingbirds.com Sent: Monday, February 24, 2025 9:59 AM To: users@nest-simulator.org Subject: [NEST Users] Re: Exponential Euler Support in NESTML
[External Email]
Hi,
Thanks for writing in. Just to be clear about terminology: for numerically stiff systems of ODEs, there exists a class of (numerical) solvers called "exponential Euler", such as Exponential Rosenbrock-Euler.
There are also linear sets of ODEs, the solution of which can be calculated algebraically (with pen and paper) and will typically end up containing terms like exp(-dt/tau).
Which of these two did you wish to use? In case of the latter, the good news is that that's already implemented and should be working out-of-the-box. For the former, I would be happy to add support for an ERE (or a similar) solver if it makes sense for your model, or you need it to reproduce results. Please let me know what you would deem a suitable ERE solver, preferably one with a C++ interface and an open-source license.
Cheers! Charl
On Mon, Feb 24, 2025, at 15:45, justin.r.davis@essie.ufl.edu wrote:
I've got a version of the HH model (via a NESTML file) which seems to be requiring an exponential Euler solver (ie the one commonly used for e^x dependent biologic systems) to most efficiently solve the equations*. From my reading of the NESTML code generator tools source, only rk45 and forward-Euler are supported. The difference between Euler and Exponential Euler doesn't seem that significant, so I wanted to see if I could get the exponential version implemented....however, what's become unclear to me is whether or not this is a limitation of NEST, NESTML or the underlying solvers and how I'd go about getting a different solver implemented. Are there any guides around on how one would implement a new solver and/or information which would help me figure this out.
thx.
jrd
*=forward Euler does work for some (might work with even smaller timestep) cases, but it requires a time step of 0.01ms. I'm comparing with a Brian2 version of the model running with a 0.1ms time step with it's natively supported exponential Euler solver which is what is making me look into other solvers for my NESTML-based code. _______________________________________________ NEST Users mailing list -- users@nest-simulator.org To unsubscribe send an email to users-leave@nest-simulator.org
_______________________________________________ NEST Users mailing list -- users@nest-simulator.org To unsubscribe send an email to users-leave@nest-simulator.org
Hi Justin,
The integration method you describe is implemented as part of NESTML’s new compartmental feature, which can be used to construct multi-compartment models in NEST (if you only want to simulate single-compartment models, you can do so by only adding one compartment when initialising the model)
Note that we are currently still in the process of creating tutorials / examples for this feature, so the user documentation is a bit sparse. However, you may have a look at the tests, where e.g. compiling the file https://github.com/nest/nestml/blob/master/tests/nest_compartmental_tests/re..., using the function `generate_nest_compartmental_target` will result in forward integration code for the ion channels as you describe (note that this uses the forward exponential method for e.g. ion channel state-variables, but implicit Euler for the voltage. You can check the python code to compile and build models here https://github.com/nest/nestml/blob/master/tests/nest_compartmental_tests/re....
For examples on how to create and run compartmental models once the nestml code is compiled, please see https://nest-simulator.readthedocs.io/en/stable/auto_examples/compartmental_... and https://nest-simulator.readthedocs.io/en/stable/auto_examples/compartmental_....
Fyi, currently, the compartmental model does not support plasticity and integrate-and-fire mechanisms as standard nestml models. Support for this will be added in the next version.
Best regards, Willem
---------------------- Dr. Willem Wybo, group leader Peter Grünberg Institute 15 - Neuromorphic Software Ecosystems Forschungszentrum Jülich www.fz-juelich.de/pgi/PGI-15 ----------------------
On 24. Feb 2025, at 17:08, Justin R. Davis justin.r.davis@essie.ufl.edu wrote:
Yes, it's the former I'm looking to implement. Specifically, I'm trying to replicate some Brian2 results which implement the exponential integrator via: https://brian2.readthedocs.io/en/stable/_modules/brian2/stateupdaters/expone...
Which I believe is just a 1st order integrator (as opposed to the higher order ERE methods) such as shown in Eqn 8 (I assume the existing forward euler looks something like Eqn 4): https://en.wikipedia.org/wiki/Numerical_methods_for_ordinary_differential_eq...
Assuming the processor is already splitting the equations into linear/non-linear parts (ie eqn 7), I don't think a separate solver library would be needed for the 1st order integrator.
jrd
-----Original Message----- From: Charl Linssen nest-users@turingbirds.com Sent: Monday, February 24, 2025 9:59 AM To: users@nest-simulator.org Subject: [NEST Users] Re: Exponential Euler Support in NESTML
[External Email]
Hi,
Thanks for writing in. Just to be clear about terminology: for numerically stiff systems of ODEs, there exists a class of (numerical) solvers called "exponential Euler", such as Exponential Rosenbrock-Euler.
There are also linear sets of ODEs, the solution of which can be calculated algebraically (with pen and paper) and will typically end up containing terms like exp(-dt/tau).
Which of these two did you wish to use? In case of the latter, the good news is that that's already implemented and should be working out-of-the-box. For the former, I would be happy to add support for an ERE (or a similar) solver if it makes sense for your model, or you need it to reproduce results. Please let me know what you would deem a suitable ERE solver, preferably one with a C++ interface and an open-source license.
Cheers! Charl
On Mon, Feb 24, 2025, at 15:45, justin.r.davis@essie.ufl.edu wrote: I've got a version of the HH model (via a NESTML file) which seems to be requiring an exponential Euler solver (ie the one commonly used for e^x dependent biologic systems) to most efficiently solve the equations*. From my reading of the NESTML code generator tools source, only rk45 and forward-Euler are supported. The difference between Euler and Exponential Euler doesn't seem that significant, so I wanted to see if I could get the exponential version implemented....however, what's become unclear to me is whether or not this is a limitation of NEST, NESTML or the underlying solvers and how I'd go about getting a different solver implemented. Are there any guides around on how one would implement a new solver and/or information which would help me figure this out.
thx.
jrd
*=forward Euler does work for some (might work with even smaller timestep) cases, but it requires a time step of 0.01ms. I'm comparing with a Brian2 version of the model running with a 0.1ms time step with it's natively supported exponential Euler solver which is what is making me look into other solvers for my NESTML-based code. _______________________________________________ NEST Users mailing list -- users@nest-simulator.org To unsubscribe send an email to users-leave@nest-simulator.org _______________________________________________ NEST Users mailing list -- users@nest-simulator.org To unsubscribe send an email to users-leave@nest-simulator.org _______________________________________________ NEST Users mailing list -- users@nest-simulator.org To unsubscribe send an email to users-leave@nest-simulator.org
------------------------------------------------------------------------------------------------ ------------------------------------------------------------------------------------------------ Forschungszentrum Jülich GmbH 52425 Jülich Sitz der Gesellschaft: Jülich Eingetragen im Handelsregister des Amtsgerichts Düren Nr. HR B 3498 Vorsitzender des Aufsichtsrats: MinDir Stefan Müller Geschäftsführung: Prof. Dr. Astrid Lambrecht (Vorsitzende), Dr. Stephanie Bauer (stellv. Vorsitzende), Prof. Dr. Ir. Pieter Jansens ------------------------------------------------------------------------------------------------ ------------------------------------------------------------------------------------------------
Hi Justin,
and as an addendum, for some literature comparing the exponential method to forward Euler, the exact method (for linear systems), and other historical solvers in neuroscience you can see the old Rotter & Diesmann 1999 [1] and the recent Plesser 2024 [2], and a trail of publications in between.
Regards, Markus
[1] https://link.springer.com/article/10.1007/s004220050570 [2] https://www.frontiersin.org/journals/neuroinformatics/articles/10.3389/fninf...
Regards, Markus
On 2/24/25 17:39, Willem Wybo wrote:
Hi Justin,
The integration method you describe is implemented as part of NESTML’s new compartmental feature, which can be used to construct multi-compartment models in NEST (if you only want to simulate single-compartment models, you can do so by only adding one compartment when initialising the model)
Note that we are currently still in the process of creating tutorials / examples for this feature, so the user documentation is a bit sparse. However, you may have a look at the tests, where e.g. compiling the file https://github.com/nest/nestml/blob/master/tests/nest_compartmental_tests/re... https://github.com/nest/nestml/blob/master/tests/nest_compartmental_tests/resources/cm_default.nestml, using the function `generate_nest_compartmental_target` will result in forward integration code for the ion channels as you describe (note that this uses the forward exponential method for e.g. ion channel state-variables, but implicit Euler for the voltage. You can check the python code to compile and build models here https://github.com/nest/nestml/blob/master/tests/nest_compartmental_tests/re... https://github.com/nest/nestml/blob/master/tests/nest_compartmental_tests/resources/cm_default.nestml.
For examples on how to create and run compartmental models once the nestml code is compiled, please see https://nest-simulator.readthedocs.io/en/stable/auto_examples/compartmental_... https://nest-simulator.readthedocs.io/en/stable/auto_examples/compartmental_model/receptors_and_current.html and https://nest-simulator.readthedocs.io/en/stable/auto_examples/compartmental_... https://nest-simulator.readthedocs.io/en/stable/auto_examples/compartmental_model/two_comps.html.
Fyi, currently, the compartmental model does not support plasticity and integrate-and-fire mechanisms as standard nestml models. Support for this will be added in the next version.
Best regards, Willem
Dr. Willem Wybo, group leader Peter Grünberg Institute 15 - Neuromorphic Software Ecosystems Forschungszentrum Jülich www.fz-juelich.de/pgi/PGI-15
On 24. Feb 2025, at 17:08, Justin R. Davis justin.r.davis@essie.ufl.edu wrote:
Yes, it's the former I'm looking to implement. Specifically, I'm trying to replicate some Brian2 results which implement the exponential integrator via: https://brian2.readthedocs.io/en/stable/_modules/brian2/stateupdaters/expone...
Which I believe is just a 1st order integrator (as opposed to the higher order ERE methods) such as shown in Eqn 8 (I assume the existing forward euler looks something like Eqn 4): https://en.wikipedia.org/wiki/Numerical_methods_for_ordinary_differential_eq...
Assuming the processor is already splitting the equations into linear/non-linear parts (ie eqn 7), I don't think a separate solver library would be needed for the 1st order integrator.
jrd
-----Original Message----- From: Charl Linssen nest-users@turingbirds.com Sent: Monday, February 24, 2025 9:59 AM To: users@nest-simulator.org Subject: [NEST Users] Re: Exponential Euler Support in NESTML
[External Email]
Hi,
Thanks for writing in. Just to be clear about terminology: for numerically stiff systems of ODEs, there exists a class of (numerical) solvers called "exponential Euler", such as Exponential Rosenbrock-Euler.
There are also linear sets of ODEs, the solution of which can be calculated algebraically (with pen and paper) and will typically end up containing terms like exp(-dt/tau).
Which of these two did you wish to use? In case of the latter, the good news is that that's already implemented and should be working out-of-the-box. For the former, I would be happy to add support for an ERE (or a similar) solver if it makes sense for your model, or you need it to reproduce results. Please let me know what you would deem a suitable ERE solver, preferably one with a C++ interface and an open-source license.
Cheers! Charl
On Mon, Feb 24, 2025, at 15:45, justin.r.davis@essie.ufl.edu wrote:
I've got a version of the HH model (via a NESTML file) which seems to be requiring an exponential Euler solver (ie the one commonly used for e^x dependent biologic systems) to most efficiently solve the equations*. From my reading of the NESTML code generator tools source, only rk45 and forward-Euler are supported. The difference between Euler and Exponential Euler doesn't seem that significant, so I wanted to see if I could get the exponential version implemented....however, what's become unclear to me is whether or not this is a limitation of NEST, NESTML or the underlying solvers and how I'd go about getting a different solver implemented. Are there any guides around on how one would implement a new solver and/or information which would help me figure this out.
thx.
jrd
*=forward Euler does work for some (might work with even smaller timestep) cases, but it requires a time step of 0.01ms. I'm comparing with a Brian2 version of the model running with a 0.1ms time step with it's natively supported exponential Euler solver which is what is making me look into other solvers for my NESTML-based code. _______________________________________________ NEST Users mailing list -- users@nest-simulator.org To unsubscribe send an email to users-leave@nest-simulator.org
NEST Users mailing list -- users@nest-simulator.org To unsubscribe send an email to users-leave@nest-simulator.org _______________________________________________ NEST Users mailing list -- users@nest-simulator.org To unsubscribe send an email to users-leave@nest-simulator.org
Forschungszentrum Jülich GmbH 52425 Jülich Sitz der Gesellschaft: Jülich Eingetragen im Handelsregister des Amtsgerichts Düren Nr. HR B 3498 Vorsitzender des Aufsichtsrats: MinDir Stefan Müller Geschäftsführung: Prof. Dr. Astrid Lambrecht (Vorsitzende), Dr. Stephanie Bauer (stellv. Vorsitzende), Prof. Dr. Ir. Pieter Jansens
NEST Users mailing list -- users@nest-simulator.org To unsubscribe send an email to users-leave@nest-simulator.org
Dear Justin,
This would certainly be cool to have as a feature. It would be ideal if it could be implemented as part of ODE-toolbox (https://ode-toolbox.readthedocs.org/), which is a Python package that we use to generate solver code for neuron and synapse models that are written in the NESTML domain-specific language (https://nestml.readthedocs.org/).
Right now ODE-toolbox supports "exact" integration (as Markus described in his response), as well as forward Euler solvers. If the exponential Euler feature is implemented in ODE-toolbox, it would be immediately available for all neuron and synapse models written in NESTML. I would be happy to look at this together. Please drop me a line if you'd like to schedule a meeting!
With kind regards, Charl
On Mon, Feb 24, 2025, at 17:08, Justin R. Davis wrote:
Yes, it's the former I'm looking to implement. Specifically, I'm trying to replicate some Brian2 results which implement the exponential integrator via: https://brian2.readthedocs.io/en/stable/_modules/brian2/stateupdaters/expone...
Which I believe is just a 1st order integrator (as opposed to the higher order ERE methods) such as shown in Eqn 8 (I assume the existing forward euler looks something like Eqn 4): https://en.wikipedia.org/wiki/Numerical_methods_for_ordinary_differential_eq...
Assuming the processor is already splitting the equations into linear/non-linear parts (ie eqn 7), I don't think a separate solver library would be needed for the 1st order integrator.
jrd
-----Original Message----- From: Charl Linssen nest-users@turingbirds.com Sent: Monday, February 24, 2025 9:59 AM To: users@nest-simulator.org Subject: [NEST Users] Re: Exponential Euler Support in NESTML
[External Email]
Hi,
Thanks for writing in. Just to be clear about terminology: for numerically stiff systems of ODEs, there exists a class of (numerical) solvers called "exponential Euler", such as Exponential Rosenbrock-Euler.
There are also linear sets of ODEs, the solution of which can be calculated algebraically (with pen and paper) and will typically end up containing terms like exp(-dt/tau).
Which of these two did you wish to use? In case of the latter, the good news is that that's already implemented and should be working out-of-the-box. For the former, I would be happy to add support for an ERE (or a similar) solver if it makes sense for your model, or you need it to reproduce results. Please let me know what you would deem a suitable ERE solver, preferably one with a C++ interface and an open-source license.
Cheers! Charl
On Mon, Feb 24, 2025, at 15:45, justin.r.davis@essie.ufl.edu wrote:
I've got a version of the HH model (via a NESTML file) which seems to be requiring an exponential Euler solver (ie the one commonly used for e^x dependent biologic systems) to most efficiently solve the equations*. From my reading of the NESTML code generator tools source, only rk45 and forward-Euler are supported. The difference between Euler and Exponential Euler doesn't seem that significant, so I wanted to see if I could get the exponential version implemented....however, what's become unclear to me is whether or not this is a limitation of NEST, NESTML or the underlying solvers and how I'd go about getting a different solver implemented. Are there any guides around on how one would implement a new solver and/or information which would help me figure this out.
thx.
jrd
*=forward Euler does work for some (might work with even smaller timestep) cases, but it requires a time step of 0.01ms. I'm comparing with a Brian2 version of the model running with a 0.1ms time step with it's natively supported exponential Euler solver which is what is making me look into other solvers for my NESTML-based code. _______________________________________________ NEST Users mailing list -- users@nest-simulator.org To unsubscribe send an email to users-leave@nest-simulator.org
NEST Users mailing list -- users@nest-simulator.org To unsubscribe send an email to users-leave@nest-simulator.org _______________________________________________ NEST Users mailing list -- users@nest-simulator.org To unsubscribe send an email to users-leave@nest-simulator.org
-
Charl Linssen -
Justin R. Davis -
justin.r.davis＠essie.ufl.edu
-
Markus Diesmann -
Willem Wybo