PD DrSci. Boris Khoromskij (PhD),
FIELD OF RESEARCH
Tensor-Structured Numerical Methods in High-Dimensional Scientific Computing
-
Development of efficient multilinear algebra in
Quantics Tensor Train/Chain and
Tensor Networks formats.
-
Tensor-product approximation of
multi-dimensional nonlocal (integral)
operators , e-tA, A-1, ,
the convolution
product and Green's functions in Rd , d > 3.
- Numerical tensor-structured methods for
the Hartree-Fock and post Hartree-Fock equations, molecular dynamics, DMRG optimization,
multi-dimensional FEM/BEM and stochastic PDEs.
-
Fast and robust numerical multi-linear algebra
based on the low-rank Tucker, canonical and mixed Tucker-canonical tensor formats.
- Tensor iterative methods and low-rank spectrally close preconditioners for multidimensional elliptic problems with application to PDE-constrained optimal control problems and for the Poisson-Boltzmann equation in protein modeling.
-
Machine learning for tensor approximation in many dimensions.
Former research topics
- Data-sparse approximation to the integral and pseudodifferential operators
using the H -matrix and related techniques.
Boundary/edge concentrated FEM/BEM.
-
Efficient
O(N logq N)-complexity representation of dense
N x N matrices approximating boundary/volume potentials (BEM).
O(N logq N) -computation and storage of the FEM elliptic
inverse and Schur-complement matrices (data-sparse direct methods).
- Robust domain decomposition methods
for elliptic/parabolic
problems, multilevel interface preconditioners. Coupling of FEM and BEM. Application
to the skin modeling problem (medicine).
-
Functional analysis and data-sparse O(N logq N) -approximation
to the elliptic Poincare-Steklov
operators.
Low order h and
hp -FEM methods on the boundary concentrated meshes.
Application to the Laplace, biharmonic, Lame and Stokes equations
as well as to elliptic equations with piecewise analytic coefficients.
-
High order finite difference and finite element schemes by the
Richardson extrapolation
techniques on a sequence of grids.
-
Local-global convergence of Newton-type methods
for solving nonlinear operator equations. Applications to
the quasi-linear elliptic boundary
value problems, inverse scattering problems and to
eigenvalue problems for
integral-differential operators arising in magnetostatics and theoretical
physics.