Romanization spreading on historical interregional networks in Northern Tunisia

PID (if applicable): doi:10.1007/s41109-022-00492-w

Problem Statement

Understand the Romanization Process of Northern Africa

Object of Research and Objective

The research topic of this project is at the interface between mathematics and archaeology and is based on an interdisciplinary collaboration between scientists from Zuse Institute Berlin, Mathematics Institute at Free University Berlin and German Archaeological Institute (Deutsches Archäologisches Institut). Our aim is to develop a novel network-based approach for inferring a well-defined range of socio-physical/historical phenomena from spatio-temporal data prevalent in Archaeology. The resulting framework is tailored to model processes such as innovation spreading, migrations and settlement development based on real-world 4D archaeological data.

In particular, within this project we focus on the archaeological evidence for the Romanization of Northern Africa. The starting point for this process is in 146 BC, when the region known nowadays as Tunisia, was annexed by Rome in the aftermath of the Third Punic War. The African province expanded further in the following centuries, with its greatest extent around 117 AD. Apart from an expected gradually developed adaptation due to the cultural exchange, some changes were actively introduced and enforced by the Roman conquerors, such as administrative structures, infrastructure and architecture.

For the romanization process only fragmented information is available. Based on this data we infer the interriogonal network and a temporal spreading rate. To do so, we have first to divide the cities into subregions based on expert knowledge. To infer the interriogonal network we are using PMALA.

Procedure

0. Extract archaeological data into a matrix: each row corresponds to a city and each column to a time-frame, the entry tells us the status of a city for a given time-frame.
1. Divide cities based on archaeological knowledge into 4 subregions. Subregions are shaped by features of the landscape, climate, economy, culture and politics.
2. Compute the number of romanized cities of each region for each time-frame. This is the main data data, that is passed to the algorithm.
3. Fit a spreading curve of an SI spreading with temporal infection rate to the data with PMALA.
4. Extract the best solutions that PMALA found.
5. Compare the top 5%

Involved Disciplines

Archaeology (wikidata:Q23498)

Epidemiology (wikidata:Q133805)

Mathematics (wikidata:Q395)  

$\Rightarrow$ Numerical Analysis (wikidata:Q11216)     

$\Rightarrow$ Statistics (wikidata:Q12483)

Data Streams

Excel Files with archaeological data from Archaeology to Mathematics.

Model

Metapopulation SI model with temporal infection rate.

Equations:

$\frac{d{s}_m(t)}{dt}=-s_m(t)\alpha (t)\sum _{n=1,n\ne m}^{N_R}(\frac{W_{m,n}}{P_m}+\frac{W_{n,m}}{P_n})i_n(t)-s_m(t)\alpha (t)\frac{W_{m,m}}{P_m}{i}_m(t)$

$i_m(t)=P_m-s_m(t)$

Short Description of Equations:

$s_m(t)\alpha (t)\sum _{n=1,n\ne m}^{N_R}(\frac{W_{m,n}}{P_m}+\frac{W_{n,m}}{P_n})i_n(t)$ : change in number of romanized cities in $m$ caused by interactions with other regions

$s_m(t)\alpha (t)\frac{W_{m,m}}{P_m}{i}_m(t)$ : influence that cities within the same region have on each other

$i_m(t)=P_m-s_m(t)$ : conservation of population number in every region $m$

Constrained Minimisation Problem:

$\min {\mathrm{\backslash limits}}_{\sigma =(G,\alpha )}\sum _{i=1}^{N_T}\sum _{m=1}^{N_R}\frac{({\omega }_{m,t_i}-\varphi (t_i|\sigma ,{\omega }_{\bullet ,0}){)}^2}{C_{m,t_i}^2}$ $\mathrm{\forall }m\ne n,0\le G_{m,n}\le 2,0\le G_{m,m}\le 1,t\ge 0,0\le \alpha (t)\le \mathrm{\infty }$

Short Description of Constrained Minimisation Problem:

$\varphi (t_i|\sigma ,{\omega }_{\bullet ,0})$ : approximate solution of the model with ${\omega }_{m,t_i}$ the observed number of romanized cities in the $mth$ region at time point $t_i$ and paramater set $\sigma$

$C_{m,t_i}$ : maximum between the data and the standard deviation for region $m$ observed in time

Discretization

• Time: interval of length of 50 years
• Space: divide cities into subregions

Variables

Name	Unit	Symbol
Time	time interval	$t$
Suseptible cities (not-romanized cities)	-	$s$
Infected cities (romanized cities)	-	$i$
spreading rate for a fixed time interval	per time interval	$\alpha (t)$
Connectivity between regions $m$ and $n$	-	$W_{m,n}$
Number of cities in region $m$	-	$P_m$
Contact network $(W/P)$	-	$G$
Connectivity network $(W)$	-	$W$
Number of network nodes	-	$N_R$
Flux	-	$H$

Process Information

Process Steps

Name	Description	Input	Output	Method	Parameter	Environment	Mathematical Area
Prepare Data	Collect, discretise and reduce archaeological data	xlsx file	npy file	-	-	Python 2.7.16	Archaeology, interdisciplinary
Create Mathematical Model	Literature Research, try models on data	Ideas	$\frac{d{s}_m(t)}{dt}$	-	-	-	Epidemiology
Solve inverse problem	Find parameters whose corresponding spreading curve describes best the data	npy file	$G$ $\alpha (t)$	PMALA	n_PMALA	Python 2.7.16	Numerical Analysis
Analysis of simulations	Get best solutions found by algorithm and compare them	$G$ $\alpha (t)$	$G_{opt}$ ${\alpha }_{opt}(t)$	t-SNE, manual inspection	perplexity	Python 2.7.16	Statistics

Applied Methods

ID	Name	Process Step	Parameter	realised/implemented by
doi:10.1002/mats.202100017	prescaled Metroplois-adjusted Langevin algorithm (PMALA)	Solve the inverse Problem	n_PMALA : number of time steps	Python 2.7.16
wikidata:Q18387205	t-Distributed Stochastic Neighbor Embedding (t-SNE)	Analysis of Simulation	perplexity = 30	Python 2.7.16

Software used

ID	Name	Description	Version	Programming Language	Dependencies	versioned	published	documented
sw:14460	Python	Python Interpreter	2.7.16	C	gcc	Yes	Yes	Yes
sw:6294	NumPy	Array Computing	1.18.1	Python, C, Fortran	Python, distutils, zlib, gcc, gfortran	Yes	Yes	Yes

Hardware

ID	Name	Processor	Compiler	#Nodes	#Cores
-	ZIB z1 Cluster	Intel(R) Xeon(R), AMD EPYC	gnu	104 (CPU), 36 (GPU)	4348 (CPU), 2072 (GPU)

Input Data

ID	Name	Size	Data Structure	Format Representation	Format Exchange	binary/text	proprietary	to publish	to archive
-	Romanization Time	small	numpy array	matrix	npy	binary	no	2022	2030

Output Data

ID	Name	Size	Data Structure	Format Representation	Format Exchange	binary/text	proprietary	to publish	to archive
doi:10.34780/d469-84h2	Simulations	middle	Lists	-	-	text	no	--	2030
doi:s41109-022-00492-w	Results	small	list of matrices and vectors	-	-	text	2022	2030
GitHub follows	Code	small	python	-	-	-	-	-

Reproducibility

Mathematical Reproducibility

We are using a stochastic algorithm. Each simulations is different. However, we run many simulations to obtain a reproducible result.

Runtime Reproducibility

The simulation time is almost constant.

Reproducibility of Results

See mathematical reproducibility. The best solutions are almost the same.

Reproducibility on original Hardware

See above

Reproducibility on other Hardware

a) Serial Computation

b) Parallel Computation

See above

Transferability to

a) similar model parameters (other initial and boundary values)

b) other models

Legend

The following abbreviations are used in the document to indicate/resolve IDs:

doi: DOI / https://dx.doi.org/

sw: swMATH / https://swmath.org/software/

wikidata: https://www.wikidata.org/wiki/