On numerical integration of the Hodgkin and Huxley equations for a membrane action potential

https://doi.org/10.1016/0022-5193(74)90054-XGet rights and content

Abstract

Several methods of numerical integration of the Hodgkin and Huxley equations (6·3°C) for a membrane action potential have been compared for speed and accuracy of computations. Conversational languages make it possible to write compact programs for these equations which will run on minicomputers with only 4000 words of core memory.

The shape of the computed action potential was found to be rather method-invariant but the latency (for a given stimulus) was found to depend on the method of integration and the size of the time increment. For stimuli near threshold, the computed response was found to be especially sensitive to these two variables. A new threshold value independent of integration method, has been determined to be − 6·50790 mV by linear extrapolations of threshold-time step relations to a zero step size.

The simple Euler method was found to be fast but gave large latency errors compared to the true solution. The Runge-Kutta (RK) method (in the sequential mode, defined in the section on numerical integration methods) was slower but gave smaller latency errors. The Adams corrector-predictor method was still slower, gave less stable solutions, and had large latency errors. A hybrid program which combined a RK numerical integration for the membrane potential with analytic expressions for the variables m, n, and h gave errors indistinguishable from the RK method, but ran much faster. Modified Euler methods run in the parallel mode gave very accurate solutions with moderate computation times. Solutions run with the RK method in the parallel mode took about twice as long but gave no appreciable improvement in accuracy.

We were able to develop a simple method for correcting the Euler integration, resulting in quite accurate solutions which were essentially independent of step size and stimuli. Because this method is not only accurate but also runs about as fast as the simple Euler, it appears to be the method of choice for most purposes.

References (15)

  • R. Fitzhugh et al.

    Biophys. J.

    (1964)
  • R.H. Adrian et al.

    J. Physiol., Lond.

    (1970)
  • K.S. Cole et al.

    J. Soc. indust. appl. Math.

    (1955)
  • F.A. Dodge

    A study of ionic permeability changes underlying excitation in myelinated nerve fibers of the frog

  • R. Fitzhugh

    Bull. math. Biophys.

    (1955)
  • R. Fitzhugh

    J. gen. Physiol.

    (1966)
  • R. Fitzhugh et al.

    J. Soc. indust. appl. Math.

    (1959)
There are more references available in the full text version of this article.

Cited by (80)

  • A comparison of six numerical methods for integrating a compartmental Hodgkin-Huxley type model

    2021, Applied Numerical Mathematics
    Citation Excerpt :

    Integration methods listed in Table 3 are intended for different contexts. At the same time there exist myriad examples of these methods being used to integrate generic HH models [5,6,8,10,23,28]. To be clear, we integrated ion channel states represented by equation (3) with ODE methods forward Euler, backward Euler and trapezoidal rule in place of PDE methods forward-time central-space (FTCS), backward-time central-space (BTCS) and HCN, respectively.

  • On the accuracy and computational cost of spiking neuron implementation

    2020, Neural Networks
    Citation Excerpt :

    Moreover, this kind of algorithms has not been incorporated into the SN simulation software, but only adaptive time step algorithms. Although several works, since Moore and Ramon (1974), have faced such a problem, this has not been treated as a multiobjective optimization one, and a general and unique solution has been intended to be found with no convincing results. Finding the optimal simulation parameters for SNs is still an open issue because we have an undetermined number of specific problems, a considerable number of NMs, firing frequencies, simulation windows, SN models, stimulation current types and intensities, and an infinite number of time steps.

  • Efficient partitioned numerical integrators for myocardial cell models

    2020, Applied Mathematics and Computation
    Citation Excerpt :

    The methods with the shortest and second-shortest execution times are given in Table 3 along with the ratio of the times. A direct comparison of RL (MRC1) and MRH (proposed by Moore and Ramon in [9]) is given in Table 4, in which the winning MME time is highlighted and the winning ratio of the slower time to the faster time is given. From these tables, we are able to make following observations.

  • Action potential duration dispersion and alternans in simulated heterogeneous cardiac tissue with a structural barrier

    2007, Biophysical Journal
    Citation Excerpt :

    No-flux boundary conditions were used. The activation and inactivation variables were computed from their analytic formula (40). To increase computation speed, lookup tables were used for evaluation of the voltage-dependent rate constants or steady-state values and time constants.

View all citing articles on Scopus
View full text