In this report, we consider the numerical solution of two challenging
Covid-19 ordinary differential equation (ODE) models that have 
discontinuities. The discontinuities are associated with modelling
the introduction of measures to slow the spread of the virus.
One of the models has a time-dependent discontinuity; this means 
that at a given point in time, a discontinuity is introduced
into the model. The other has a state-dependent discontinuity;
in this case, the time at which the discontinuity arises depends on 
the value of one of the solution components, and thus it is not known 
a priori. These discontinuities make the models quite challenging 
for standard ODE solvers to solve. As well, the presence of 
exponentially growing solution components adds to the difficulties 
faced by standard ODE solvers. We also consider variations on 
these problems where the change in the model is not discontinuous 
but happens quickly over a short period of time in order to better 
model how the public reacts to the introduction of public health 
measures.

In this report, we present an investigation of performance of a 
collection of ODE solvers (we consider 21 solvers) available in 
four popular software environments: R, Python, Scilab, and Matlab, 
when applied to solve these Covid-19 models.

We first focus on straightforward implementations of the models 
where the user employs the solver to attempt to solve the problems 
using default settings, e.g., default tolerances, and simple 
implementations for the discontinuities, i.e., the introduction 
of `if' statements into the functions that define the right-hand 
sides of the ODE systems. Such implementations of the models
and usage of the solvers are typical of what Covid-19 researchers 
might employ in attempting to solve their models. We then follow 
with an investigation of approaches for solving the models that 
make better use of the capabilities of the solvers.

We also highlight a number of issues with the way that some of the 
solvers are implemented in some of the software environments. For 
example, the treatment of output points, i.e., the points in the 
domain where solution values are required, is an issue for some 
of the solvers in some of the software environments. 

We show that the standard use of ODE solvers available within  
widely used software environments, applied to simple 
implementations of these Covid-19 models, will frequently deliver 
numerical solutions that have no significant digits of accuracy.
Furthermore, the solvers give no indication that the returned 
solutions are inaccurate. We also show that these straightforward 
treatments of the models are frequently also inefficient. We show 
that the more advanced treatments of the models can result in 
more efficient computations while at the same time providing
more accurate approximate solutions.