Abstract
The present work concerns the COVID-19’s spread over The United States, Brazil and Colombia. Although countries show differences in economic development, but similarities such as continental dimension or social interaction, the spread of COVID-19 in them has some similarities. At the moment, the countries are living the disease with temporal delay. Thus, we used a database on WHO Coronavirus, Mathematical Modeling and Numerical Simulations to describe the most recent COVID-19 development patterns in these countries, which we saw.
1. Introduction
The SARS-CoV-2 coronavirus pandemic causes the COVID-19 disease, for which we have no immune response or vaccine. The origin of COVID-19 is believed to have occurred in Wuhan, China, in late 2019. From China, the disease was rapidly transmitted globally by individuals who travelled to Europe and The United States.
On the America continent, the first cases of COVID-19 appeared in The United States. On January 21, 2020, the American Center for Disease Control and Prevention (CDC) confirmed the first case of COVID-19 in a 35-year-old man from Snohomish County, Washington, who returned from a trip to the region around Wuhan [1], In February 2020, cases of COVID-19 were already emerging in several countries on the American continent. The coronavirus landed in Latin America on February 26, when Brazil confirmed a case in São Paulo, a 61-year-old man who returned from a trip to Italy [2]. The Colombia’s first case of COVID-19 took place on March 6, 2020, a 19-year-old female who traveled to Milan, Italy [3]. Currently, COVID-19 has reached almost all countries on the American continent.
According to the World Health Organization (WHO), on July 31, 2020, The United States, Brazil and Colombia were the countries of the American continent with the highest numbers of daily infected cases [4]. By this date, 9.152.173 cases and 351.121 deaths by COVID-19 had already occurred in the American continent, comprising over 53.5% of the total number of reported cases in the world. The United States, Brazil and Colombia present 4.388.566, 2.552.265 and 276.055 cases with 150.054, 90.134 and 9.454 deaths, respectively, comprising over 48%, 28% and 3% of the total number of reported cases in the American continent [4].
The United States, Brazil and Colombia are in different moments in the epidemiological process, Colombia is in the exponential growth phase, Brazil is probably in the peak of the epidemic, while The United States is already experiencing a second wave of SARS-CoV-2 coronavirus infection. It is noteworthy that these three countries have very different Human Development Indexes (HDI) and territorial dimensions and, even so, they have in common high infection rates by the SARS-CoV-2 coronavirus.
The objective of this article is to carry out a mathematical study of the possible trends of the epidemic by COVID-19 in these three countries. This article was based on the data provided by the WHO [4] and the Susceptibles-Infectious-Recovered-Dead model, the SIRD model [5][6].
The SIRD model is a classic compartmental model of the Kermack - McK-endrick type [7] [8]. Compartmental models divide the population into several different compartments, for example, Susceptible population, Infectious population, Recovered population, Dead population, among others, and specify how individuals move through the compartments over time. Despite being a simple mathematical model, the SIRD is one of the most applied mathematical models to understand the current health crisis. Obviously, more realistic and complex models would better describe the dynamics of this epidemic, but data and information about COVID-19 are lacking to implement them. Reviews of epidemiological models can be found in [5][6]
Regarding the adjustment of the parameters of the SIRD model for The United States, Brazil and Colombia, it appears that these parameters change frequently, depending on local political factors (closure of non-essential establishments, quarantine, movement restrictions…) [9], socio-economic factors (social and hygienic behaviors, lower per-capita income…) [10][11], climatic factors (temperature, humidity, average wind speed, UV index…) [12], among others. In this context, it was decided to adjust the parameters of the SIRD model over time, based on data made available by WHO, using non-linear least squares methods [13][14].
Regarding the numerical procedures to solve the SIRD model, first, the discretization of the ordinary differential equation system was performed using the Finite Difference Method. The resulting linear system of non-linear equations was solved iteratively by the Gauss-Seidel method until the convergence criterion was overcome. It was also found that all the matrices’s coefficients of the iterative processes satisfy the Sassenfeld convergence criterion [15][16] [17].
2. Materials and Methods
2.1. Data Base
There are several data base source about COVID-19, e. g., Brazil’s Coronavirus Panel [18], the National Institute of Health in Colombia [19], the Johns Hopkins Coronavirus Resource Center [20] and among other sources that exhibit the numbers of the infected, dead and more information as well.
But, to this work we used the WHO Coronavirus Disease (COVID-19) Dashboard [4] to keep integrity and uniformity of our numerical model.
From source WHO Coronavirus Dashboard we did not have the recovering data over time, so we used the assumption that says - the time for a person to be moved from the infected to the recovered compartmental can be about 14 days [21] [22]. Thus, we reconstructed the number of people in the compartments: Susceptible, Infected, Recovered and Dead.
In fact the coronavirus disease is not completely understanding at the moment, certainly there are additional parameters unknown that describing better the pattern it. So, we chose to take several data sets consolidated of the 14 days to do our simulations in order to get more realistic information. In this way, we got a simulations’ clustering which permitted to understanding the fundamentals of the behavior coronavirus disease over the countries.
How already mentioned in the introduction section, the disease begins at The United States, Brazil and Colombia countries in a different times. The government actions (good or bad), civil liberties, and mass testing come interfering with the situation of COVID-19 in countries. All of these conducts imply at the number variations of the infected and dead people, what is transcribed at the WHO Coronavirus Disease (COVID-19) data base.
From data base, it can be seen that The United States is living a second wave contamination. Brazil, maybe, arrived at the infecting peak on the August month. Colombia, clearly, is going up at the infected number yet. So, we have a time delay of the COVID-19’s spread in the countries. With these, we want to explain how disease spreading happens in the countries through numerical simulations. But with a cluster of simulations, not just a simulation alone.
2.2. Governing Equations
There are several mathematical modelling of the COVID-19 at the moment. But, to this work we considering the SIRD’s model given by equations:
where:
t, S, I, R, D are time, susceptible, infected, recovered and death variables, respectively;
N is an average population value;
β is an infection rate, and γR and γD are recover and death rates, these parameters are calculated per day.
At the initial time we have
and for end time, denoted by tF, we admitted

The condition (2.3) was considered because at this moment the disease finished and the equations’ steady state stays established.
2.3. SIRD’s Numerical Modelling Proposal
The model (2.1) has 4 ordinary differential equations (ODE). Specifically, any ODE can be written as
so that
therefore, all deductive numeric procedure at this work will be over equation (2.4).
The temporal discret grid considered is showed at the Figure (1). We denoted the time lapse like Δt, and the end time tF = (s − 1) Δt. The value k is a time counter, thereby k = s means temporal nodes’ total value.
Temporal discret grid.
From condition (2.2) we can write like that
and the Neumann condition (2.3) we have too

With respect to temporal derivative of the equations (2.4), they are approximated by forward second order finite difference in the node k
similarly, the equations (2.6) are approximated by backward second order finite difference at the last node s

These approximations were used to obtain improved numerical results.
So, writing equation (2.4) to node k we can see
and inserting formula (2.7) at the last equation (2.9), what it leads to
or finally

The equality (2.10) is an our temporal difference equation, that describe the COVID-19 behaviour. We want to explain that the equation (2.10) is the temporal evolution for Φl. Numerically, from Figure (2), setting s we have to vary k = 1,…, s − 2 to obtain the differences’ equations at the points k = 2,…, s − 1.
Computational temporal grid until s.
The first possible value is s = 3, so just a simple linear equation is found, and the is known at the time tF = Δt.
Nevertheless, with help the Figure (2) again, if s > 3 we will have to solve l systems of the (s − 2) difference equations to compute (labeled to star symbol) to tF = (s − 2) Δt. It remembering which at the square symbol the initial condition (2.5) is assumed and the triangle symbol the Neumann condition (2.6) is used as well. Particularly, the system mentioned is the kind of

Note that the system (2.11) is not a linear system. The terms
are known if the
values were explicit. But they are unknown! Thus, we use a iterative strategy to solve the system (2.11) how presented in [23].
Without loss of generality, if s = 4 the system (2.11) becomes
whose the iterative process associated can be like
where IT = 0, 1,…, ITMAX, and ITMAX means iteration maximum number. Particularly, we have a linear system to each specific IT. Furthermore, we choose the Gauss-Seidel method to solve (2.12) because the convergence is guaranteed, it is easy to see that the Sassenfeld criterium is verified. The stopped criteria value of the Gauss-Seidel method used in this work was 10-7.
When IT = 0 the system (2.12) becomes
so that:
is known (initial condition);
is known, of course;
is admitted how our prediction hypothesis.
Thus, we have the linear system’s solution by means of Gauss-Seidel method. Now, if IT = 1 we find
so that
is our prediction hypothesis again. It solving (2.14), we get
.
After this, if the 2Φ2 is less than one (which means an end to the spread of covid-19) we accept the 2Φl = Φl as numerical solution of the SIRD’s (2.4) at the time t = 2Δt. Otherwise, we set s = 5 and do the all process again to a new time t = 3Δt, and so on.
The numerical process finishes if we get the mΦ2 < 1 to some 0 < m < ITMAX. Finally, we have some gains with strategy provided here:
the governing equations system (2.4) is solved by a convergent methodology to any time;
our implicit methodology does not need a severe restriction on the Δt value.
The implicit method together with Gauss-Seidel and Lax’s law guarantee it. Besides, the tF time is discovered in the computational run’s time. To performance all simulation we developed a fortran 90 code that solve the SIRD’s governing equations from auxiliary conditions.
2.4. Parameters’ Optimization
The parameters’ optimization in the SIRD (2.1) is done as in [24, 25], that is, solving a nonlinear least square problem. We define the vector function u(t) = (S(t), I(t), R(t), D(t)), the vectors of parameters q = (β,γR, γD) and the known data y at times t1,...tn. Given a function F(u, q), we estimate the parameters q by solving the following nonlinear least square problem

The trust-region based method adapted in the Isqnonlin Matlab function is used for the constrained optimization problem (2.15).
3. Discussion and concluding remarks
In this section we introduce our analyse of the COVID-19 and simulations as well. The reader will see here that the country’s richness, its continental dimension or social interaction are not avoid the Sars-CoV-2 infection.
By other hand, it is good to explain we used normalized measures, what allow us to compare the disease spread at the countries. We setting N population values like 328.200.000, 211.000.000, 49.650.000 to The United States (USA), Brazil (BRA) and Colombia (COL), respectively. Thus the SIRD’s model variables kept at the [0,1] window. Also, the times counting begins in t = 0 which corresponds to 2020-01-20 (USA), 2020-02-26 (BRA) and 2020-03-06 (COL). We remember that our simulations were done with data taken every 14 days to each country. What it led to label the simulations like S1,…, S6.
The Figures (3)-(4) show the data for susceptible, infected, recovered and dead in the time. We preferred to display them this shape to see the COVID-19 rises at the countries. For latin countries, the infected number came going up, but it looks like the Brazil already reached the peak. The american population is living a second wave and walks to second peak, probably. Additionally, the recover and dead people increase. But with a recover number much higher than the death.
Susceptible (left), and infected (right) normalized data
Recovered (left), and dead (right) normalized data
The Figure (5 - left) shows the susceptible decrease dynamic of The United States, and our simulations. We started simulations at the date 2020-05-04. From Figure (5 - right), the zoom Figure (5 - left), it is possible to see a changing of the disease’s drop behavior near of t = 139 (date 2020-06-07) by our simulations. Here, in the neighborhood of the date, it seems that there is an inflection point. This is more evident, if we look at the Figure (6) as well.
USA-Susceptible reconstructed data from WHO Coronavirus Disease (COVID-19) Dashboard [4] and simulations S1,…,S6.
USA-Infected reconstructed data from WHO Coronavirus Disease (COVID-19) Dashboard [4] and simulations S1,…,S6.
We emphasize that the date 2020-06-07 corresponds to a 14 days after the George Floyd’s assassinate (2020-05-25) at the Minneapolis-US. Maybe, the inflection point could be correlated to not social distancing of the USA people manifestation about the assassinate. Which could to evidence the start of the second wave of Sars-CoV-2 infection.
The Figure (6 - right) shows the inverse tendency over infected people. The COVID-19 changes from deceleration to acceleration status. For we this is a characteristic of a new infection wave. The simulations predicts a new peak arising at the country, which can be seen at the Figure (6 - left). However, it is seen that the peak is being attenuating at the country.
Beyond of that, The United States have indicated a value climb at the recover and death’s compartments. But, we emphasized the recovered number’s magnitude is higher than the death number, see the Figure (7). Particularly, by our last simulations, it can be seen a damping of the accumulated numbers as well.
USA-Recovered(left)/Dead(right) reconstructed data from WHO Coronavirus Disease (COVID-19) Dashboard [4] and our simulations S1,…,S6.
For Brazil, the Figure (8) shows the susceptibles’ decelerating dynamic, and our simulations too. The our simulations started in the date 2020-05-12 because the COVID-19 cases became more manifest. Different of The United States, by means of our predictions, Brazil and Colombia do not live a second infection wave so far. Colombia, whose simulations started in the date 2020-05-14, presents the same patter of Brazil’s susceptible, see the Figure (9). However, the Colombia’s Covid-19 disease is more accelerated than Brazil, it’s further evident in the last days, see the Figure (8 - left) and Figure (9 - right).
Brazil-Susceptible reconstructed data from WHO Coronavirus Disease (COVID-19) Dashboard [4] and simulations S1,…,S6.
Colombia-Susceptible reconstructed data from WHO Coronavirus Disease (COVID-19) Dashboard [4] and simulations S1,…,S6.
Colombia do not rich the disease peak, yet. This can be seen through of the Figure (10). The COVID-19’s spread come increasing and its peak is still far away. Nevertheless, to Brazil it is possible to infer the disease peak and plateau happen between August and September in the present year, see the Figure (11). We assume it because at the months May and June the COVID-19 became decelerating.
Colombia-Infected reconstructed data from WHO Coronavirus Disease (COVID-19) Dashboard [4] and simulations S1,…,S6.
Brazil-Infected reconstructed data from WHO Coronavirus Disease (COVID-19) Dashboard [4] and simulations S1,…,S6.
Analogous to The United States, the recovered and death Brazilians numbers climbs at the values, Figure (12) but with different magnitudes between them as well. This patter happens to Colombia as well, look the Figure (13).
Brazil-Recovered(left)/Dead(right) reconstructed datafrom WHO Coronavirus Disease (COVID-19) Dashboard [4] and our simulations S1,…,S6.
Colombia-Recovered(left)/Dead(right) reconstructed data from WHO Coronavirus Disease (COVID-19) Dashboard [4] and our simulations S1,…,S6.
We calculated the basic reproduction number for the last simulations on July. The United States is fighting with disease, its R0 comes decreasing, see the Figure (14 - left). The Brazil’s R0 came dropping but it changed, Figure (14 - center). If it persists, it’s possible increasing infection again. The Colombia’s R0 is oscillating, Figure (14 - right), and all care are necessary at the currenty time.
Basic reproduction number R0 at the time.
Finally, we emphasize that the COVID-19’s dynamic of the Brazil and Colombia are the same, but as delay. Of course the next days are crucials, the countries cannot relax your fight actions against COVID-19. The our simulations predict that there are some control over COVID-19 disease in theses countries, yet. Clearly, we do not know the specific proceedings which provide it. But if they relax, the countries could have a second wave infection similar to the one the United States is currently experiencing.
Resumo. O presente trabalho trata do espalhamento da COVID-19 sobre os Estados Unidos, Brasil e Colômbia. Embora os países exibam diferençias no desenvolvimento econômico, mas similaridades tais como dimensão continental ou interação social, a propagação do COVID-19 neles tem algumas semelhanças. Agora, os países estão vivendo a doença com atraso temporal. Assim, usamos um banco de dados de Coronavirus da OMS, Modelagem Matemática e Simulações Numéricas para descrever o padrão mais recente de desenvolvimento do COVID-19 nesses países, que nós observamos.
Palavras-chave. Modelagem Matemática, Simulaçõs Numéricas, COVID-19.
Footnotes
1 This work was supported by the Universidade Estadual de Londrina-BR. The fourth and fifth author acknowledge the support of the Centro de Modelacion Matemática y Computacion Científica of the Universidad del Atlantico.
↵2 ercirilo{at}uel.br
↵3 plnatti{at}uel.br
↵4 nromeiro{at}uel.br
↵5 miguelcaro{at}mail.uniatlantico.edu.co
↵6 jeinny.peralta{at}unad.edu.co