Parallel algorithm for mixed HPC systems to determine the solutions in the bimatrix informational extended games
Închide
Articolul precedent
Articolul urmator
239 3
Ultima descărcare din IBN:
2024-03-15 02:05
SM ISO690:2012
HANCU, Boris, GLADEI, Anatol. Parallel algorithm for mixed HPC systems to determine the solutions in the bimatrix informational extended games. In: Conference on Applied and Industrial Mathematics: CAIM 2022, Ed. 29, 25-27 august 2022, Chişinău. Chișinău, Republica Moldova: Casa Editorial-Poligrafică „Bons Offices”, 2022, Ediţia a 29, pp. 109-111. ISBN 978-9975-81-074-6.
EXPORT metadate:
Google Scholar
Crossref
CERIF

DataCite
Dublin Core
Conference on Applied and Industrial Mathematics
Ediţia a 29, 2022
Conferința "Conference on Applied and Industrial Mathematics"
29, Chişinău, Moldova, 25-27 august 2022

Parallel algorithm for mixed HPC systems to determine the solutions in the bimatrix informational extended games


Pag. 109-111

Hancu Boris, Gladei Anatol
 
Moldova State University
 
 
Disponibil în IBN: 21 decembrie 2022


Rezumat

According to [1] we can construct and implement on HPC systems the following parallel algorithm to find the all informational extended equilibrium profiles in bimatrix games. Algorithm 1. Using the MPI programming model we generate the MPI Communicator with linear topology and dimension κ1 · κ2. Root process, using the MPI Bcast function, broadcasts to all MPI process the initial matrices A = ||aij ||j∈J i∈I , and B = ||bij ||j∈J i∈I of the bimatrix game Γ = ⟨A,B⟩ . 2. Using the MPI programming model and open source library ScaLAPACK-BLACS, the processes grid {(α, β)}β=1,κ2 α=1, κ1 is initialized, and in parallelall fixed (α, β)−processes, using combinatorial algorithm, construct the informational extended strategies iα = iα1 iα2 ...iαj ...iα m and jβ = jβ 1 jβ 2 ...jβ i ...jβ n. 3. In parallel, all fixed MPI (α, β)-processes, using the OpenMP directives, construct utility matrices A(α, β) = ||aiαj jβ i ||j∈J i∈I and B(α, β) = ||biαj jβ i ||j∈J i∈I , generated by the informational extended strategies iα and jβ. 4. In parallel, the α-rank MPI process, for all α = 1, κ1, generates the ”beliver-probabilities” p(β/α) for all fixed β = 1, κ2 of the α-type player 1 and also, β-rank MPI process, for all β = 1, κ2, generates the ”beliver-probabilities” q(α/β) for all fixed α = 1, κ1 of the β-type player 2. 5. Using MPI and OpenMP programming models, for all α = 1, κ1, in parallel, the α-rank MPI process generates the sets L(α) of the lα = lα 1 lα 2 ...lα β ...lα κ2 strategies and constructs the payoff matrix A(α) of the α-type player 1 and the β-rank MPI process, for all β = 1, κ2, generates the sets C(β) of the cβ = cβ 1 cβ 2 ...cβ α...cβ κ1 strategies and constructs the payoff matrix B(β) of the β-type player 2. So all MPI (α, β)-processes have a pair of matrices (A(α),B(β)) . 7. In parallel, all MPI (α, β)-processes, using the OpenMP functions, ScaLAPACK routines and existing the sequential algorithm, determine all Nash equilibrium profiles in the bimatrix game with matrices (A(α),B(β)) . 8. Using ScaLAPACK-BLACS routines, the root MPI process gather from processes grid {(α, β)}β=1,κ2 α=1, κ1 the sets of Nash equilibrium profiles in the bimatrix game (A(α),B(β)) . In general case, to determine all sets of Bayes-Nash equilibrium profiles in bimatrix informational extended games a very large number (equal to nm×mn) of the bimatrix subgames in the non extended strategies are to be solved.