class brainpy.integrators.fde.GLShortMemory(f, alpha, inits, num_memory, dt=None, name=None, state_delays=None)[source]#

Efficient Computation of the Short-Memory Principle in Grünwald-Letnikov Method [1].

According to the explicit numerical approximation of Grünwald-Letnikov, the fractional-order derivative \(q\) for a discrete function \(f(t_K)\) can be described as follows:

\[{{}_{k-\frac{L_{m}}{h}}D_{t_{k}}^{q}}f(t_{k})\approx h^{-q} \sum\limits_{j=0}^{k}C_{j}^{q}f(t_{k-j})\]

where \(L_{m}\) is the memory lenght, \(h\) is the integration step size, and \(C_{j}^{q}\) are the binomial coefficients which are calculated recursively with

\[C_{0}^{q}=1,\ C_{j}^{q}=\left(1- \frac{1+q}{j}\right)C_{j-1}^{q},\ j=1,2, \ldots k.\]

Then, the numerical solution for a fractional-order differential equation (FODE) expressed in the form


can be obtained by

\[x(t_{k})=f(x(t_{k-1}))h^{q}- \sum\limits_{j=1}^{k}C_{j}^{q}x(t_{k-j}).\]

for \(0 < q < 1\). The above expression requires infinity memory length for numerical solution since the summation term depends on the discritized time \(t_k\). This implies relatively high simulation times.

To reduce the computational time, the upper bound of summation needs to be modified by \(k=v\), where

\[\begin{split}v=\begin{cases} k, & k\leq M,\\ L_{m}, & k > M. \end{cases}\end{split}\]

This is known as the short-memory principle, where \(M\) is the memory window with a width defined by \(M=\frac{L_{m}}{h}\). As was reported in [2], the accuracy increases by increaing the width of memory window.


>>> import brainpy as bp
>>> a, b, c = 10, 28, 8 / 3
>>> def lorenz(x, y, z, t):
>>>   dx = a * (y - x)
>>>   dy = x * (b - z) - y
>>>   dz = x * y - c * z
>>>   return dx, dy, dz
>>> integral = bp.fde.GLShortMemory(lorenz,
>>>                                 alpha=0.96,
>>>                                 num_step=500,
>>>                                 inits=[1., 0., 1.])
>>> runner = bp.integrators.IntegratorRunner(integral,
>>>                                          monitors=list('xyz'),
>>>                                          inits=[1., 0., 1.],
>>>                                          dt=0.005)
>>> import matplotlib.pyplot as plt
>>> plt.plot(runner.mon.x.flatten(), runner.mon.z.flatten())
  • f (callable) – The derivative function.

  • alpha (int, float, jnp.ndarray, bm.ndarray, sequence) – The fractional-order of the derivative function. Should be in the range of (0., 1.).

  • num_memory (int) –

    The length of the short memory.

    Changed in version 2.1.11.

  • inits (sequence) – A sequence of the initial values for variables.

  • dt (float, int) – The numerical precision.

  • name (str) – The integrator name.


__init__(f, alpha, inits, num_memory, dt=None, name=None, state_delays=None)[source]#


__init__(f, alpha, inits, num_memory[, dt, ...])


Move all variable into the CPU device.


Move all variables into the GPU device.

load_state_dict(state_dict[, warn, compatible])

Copy parameters and buffers from state_dict into this module and its descendants.

load_states(filename[, verbose])

Load the model states.

nodes([method, level, include_self])

Collect all children nodes.

register_implicit_nodes(*nodes[, node_cls])

register_implicit_vars(*variables[, var_cls])


Reset function of the delay variables.

save_states(filename[, variables])

Save the model states.


Set the integral function.


Returns a dictionary containing a whole state of the module.


Moves all variables into the given device.


Move all variables into the TPU device.

train_vars([method, level, include_self])

The shortcut for retrieving all trainable variables.


Flattens the object as a PyTree.

tree_unflatten(aux, dynamic_values)

Unflatten the data to construct an object of this class.

unique_name([name, type_])

Get the unique name for this object.

vars([method, level, include_self, ...])

Collect all variables in this node and the children nodes.



All arguments when calling the numer integrator of the differential equation.



The numerical integration precision.


The integral function.


Name of the model.


The parameters defined in the differential equation.


State delays.


The variables defined in the differential equation.