arXiv subjects
Publications and source records attributed to Dinesh Singh.
In this paper we study subspaces which are invariant under squares and cubes (separately as well as jointly) of unicellular backward weighted shift operators on a separable Hilbert space. The finite-dimensional subspaces are characterized for all weights and the infinite-dimensional subspaces are characterized for two classes of weights.
Second-order optimization methods are among the most widely used optimization approaches for convex optimization problems, and have recently been used to optimize non-convex optimization problems such as deep learning models. The widely used second-order optimization methods such as quasi-Newton methods generally provide curvature information by approximating the Hessian using the secant equation. However, the secant equation becomes insipid in approximating the Newton step owing to its use of the first-order derivatives. In this study, we propose an approximate Newton sketch-based stochastic optimization algorithm for large-scale empirical risk minimization. Specifically, we compute a partial column Hessian of size ($d\times m$) with $m\ll d$ randomly selected variables, then use the \emph{Nyström method} to better approximate the full Hessian matrix. To further reduce the computational complexity per iteration, we directly compute the update step ($Δ\boldsymbol{w}$) without computing and storing the full Hessian or its inverse. We then integrate our approximated Hessian with stochastic gradient descent and stochastic variance-reduced gradient methods. The results of numerical experiments on both convex and non-convex functions show that the proposed approach was able to obtain a better approximation of Newton\textquotesingle s method, exhibiting performance competitive with that of state-of-the-art first-order and stochastic quasi-Newton methods. Furthermore, we provide a theoretical convergence analysis for convex functions.
The classical inequality of Bohr concerning Taylor coeficients of bounded holomorphic functions on the unit disk, has proved to be of significance in answering in the negative the conjecture that if the non-unital von Neumann inequality held for a Banach algebra then it was necessarily an operator algebra. Here we provide a rather short and easy proof of the inequality.
This paper deals with representing in concrete fashion those Hilbert spaces that are vector subspaces of the Hardy spaces $H^p(\bb D^n) \ (1\le p\le \infty)$ that remain invariant under the action of coordinate wise multiplication by an $n$-tuple $(T_{B_1},\dots, T_{B_n})$ of operators where each $B_i, \ 1\le i\le n,$ is a finite Blaschke factor on the open unit disc. The critical point to be noted is that these $T_{B_i}$ are assumed to be weaker than isometries as operators. Thus our main theorem extends the principal result of \cite{LS} in the following three directions: $(i)$ from one to several variables; $(ii)$ from multiplication with the coordinate function $z$ to an $n$-tuple of multiplication by finite Blaschke factors $B_i, \ 1\le i\le n;$ $(iii)$ from vector subspaces of $H^2(\bb D)$ to the case of vector subspaces of $H^p(\bb D^n), \ 1\le p\le \infty.$ We further derive a generalization of Slocinski's well known Wold type decomposition of a pair of doubly commuting isometries to the case of $n$-tuple of doubly commuting operators whose actions are weaker than isometries.
The orbital dynamics of fast spinning neutron stars encountering a massive Black Hole (BH) with unbounded orbits are investigated using the quadratic-in-spin Mathisson-Papapetrou- Dixon (MPD) formulation. We consider the motion of the spinning neutron stars with astrophysically relevant speed in the gravity field of the BH. For such slow-speed scattering, the hyperbolic orbits followed by these neutron stars all have near the e = 1 eccentricity, and have distinct properties compared with those of e >> 1. We have found that compared with geodesic motion, the spin-orbit and spin-spin coupling will lead to a variation of scattering angles at spatial infinity, and this variation is more prominent for slow-speed scattering than fast-speed scattering. Such a variation leads to an observable difference in pulse-arrival-time within a few hours of observation, and up to a few days or months for larger BH masses or longer spinning periods. Such a relativistic pulsar-BH system also emits a burst of gravitational waves (GWs) in the sensitivity band of LISA, and for optimal settings, can be seen up to 100 Mpc away. A radio follow up of such a GW burst with SKA or FAST will allow for measuring the orbital parameters with high accuracy and testing the predictions of General Relativity (GR).
In this paper, we extend the classical Bohr's inequality to the setting of the non-commutative Hardy space $H^1$ associated with a semifinite von Neumann algebra. As a consequence, we obtain Bohr's inequality for operators in the von Neumann-Schatten class $\cl C_1$ and square matrices of any finite order. Interestingly, we establish that the optimal bound for $r$ in the above mentioned Bohr's inequality concerning von Neumann-Shcatten class is 1/3 whereas it is 1/2 in the case of $2\times 2$ matrices and reduces to $\sqrt{2}-1$ for the case of $3\times 3$ matrices. We also obtain a generalization of our above-mentioned Bohr's inequality for finite matrices where we show that the optimal bound for $r$, unlike above, remains 1/3 for every fixed order $n\times n,\ n\ge 2$.
Biological data including gene expression data are generally high-dimensional and require efficient, generalizable, and scalable machine-learning methods to discover their complex nonlinear patterns. The recent advances in machine learning can be attributed to deep neural networks (DNNs), which excel in various tasks in terms of computer vision and natural language processing. However, standard DNNs are not appropriate for high-dimensional datasets generated in biology because they have many parameters, which in turn require many samples. In this paper, we propose a DNN-based, nonlinear feature selection method, called the feature selection network (FsNet), for high-dimensional and small number of sample data. Specifically, FsNet comprises a selection layer that selects features and a reconstruction layer that stabilizes the training. Because a large number of parameters in the selection and reconstruction layers can easily result in overfitting under a limited number of samples, we use two tiny networks to predict the large, virtual weight matrices of the selection and reconstruction layers. Experimental results on several real-world, high-dimensional biological datasets demonstrate the efficacy of the proposed method.
Graph structured data has wide applicability in various domains such as physics, chemistry, biology, computer vision, and social networks, to name a few. Recently, graph neural networks (GNN) were shown to be successful in effectively representing graph structured data because of their good performance and generalization ability. GNN is a deep learning based method that learns a node representation by combining specific nodes and the structural/topological information of a graph. However, like other deep models, explaining the effectiveness of GNN models is a challenging task because of the complex nonlinear transformations made over the iterations. In this paper, we propose GraphLIME, a local interpretable model explanation for graphs using the Hilbert-Schmidt Independence Criterion (HSIC) Lasso, which is a nonlinear feature selection method. GraphLIME is a generic GNN-model explanation framework that learns a nonlinear interpretable model locally in the subgraph of the node being explained. More specifically, to explain a node, we generate a nonlinear interpretable model from its $N$-hop neighborhood and then compute the K most representative features as the explanations of its prediction using HSIC Lasso. Through experiments on two real-world datasets, the explanations of GraphLIME are found to be of extraordinary degree and more descriptive in comparison to the existing explanation methods.
Being one of the strongest materials, ternary TiSiN exhibits a very interesting family of binary transition metal nitride and silicide systems. A novel technique to fabricate morphologically fascinating nano and micro structures of TiSiN is reported here. The referred TiSiN films, majorly constituted with cubic TiN phase, are enriched with crystalline nanoparticles, micro-flowers and faceted micro-crystals which possess attractive functionalities towards plasmon mediated optoelectronic applications. Reactivity of titanium to silicon nitride based dielectric topping on the substrate at high temperature plays the key role in nitride formation for the demonstrated protocol. The optoelectronic response for these morphologically enriched composite films indicates an influential role of photo-induced surface plasmon polaritons on their dc transport properties. A plasmonically tuned resistive switching, controlled by the surface morphology in association with the film thickness, is observed under light illumination. Using Drudes modified frequency dependent bulk electron scattering rates and surface mediated SPPs-electron scattering rates, a generic model is proposed for addressing unambiguously the increased device resistance in response to light. The featured synthesis process opens a new direction towards the growth of transition metal nitrides while the proposed model serves as a basic platform to understand photo-induced electron scattering mechanisms in metal.
We investigate the spin dynamics of a millisecond pulsar (MSP) in a tightly bounded orbit around a massive black hole. These binaries are progenitors of the extreme-mass-ratio-inspirals (EMRIs) and intermediate-mass-ratio-inspirals (IMRIs) gravitational wave events. The Mathisson-Papapetrou-Dixon (MPD) formulation is used to determine the orbital motion and spin modulation and evolution. We show that the MSP will not be confined in a planar Keplerian orbit and its spin will exhibit precession and nutation induced by spin-orbit coupling and spin-curvature interaction. These spin and orbital behaviours will manifest observationally in the temporal variations in the MSP's pulsed emission and, with certain geometries, in the self-occultation of the pulsar's emitting poles. Radio pulsar timing observations will be able to detect such signatures. These extreme-mass-ratio binaries (EMRBs) and intermediate-mass-ratio binaries (IMRBs) are also strong gravitational wave sources. Combining radio pulsar timing and gravitational wave observations will allow us to determine the dynamics of these systems in high precision and hence the subtle behaviours of spinning masses in strong gravity.
In this note we study sub-Hardy Hilbert spaces on which the the action of the operator of multiplication by the coordinate function z is assumed to be weaker than that of an isometry. We identify such operators with a class of weighted shifts. The well known results of de Branges and Beurling are deduced as corollaries .
In a recent paper, M. Raghupathi has extended the famous theorem of Beurling to the context of subspaces that are invariant under the class of subalgebras of $H^\infty$ of the form $IH^\infty$, where $I$ is an inner function. In this paper, we provide analouges of the above mentioned $IH^\infty$ related extension of Beurling's theorem to the context of uniform algebras, on compact abelian groups with ordered duals, the Lebesgue space on the real line and in the setting of the space $BMOA$. We also provide a significant simplification of the proof of the Beurling's theorem in the setting of uniform algebras and a new proof of the Helson-Lowdenslager theorem that generalizes Beurling's theorem in the context of compact abelian groups with ordered duals.
In this note we produce generalized versions of the classical inequalities of Hardy and of Hilbert and we establish their equivalence. Our methods rely on the H^1-BMOA duality. We produce a class of examples to establish that the generalizations are non-trivial.
Dwarf galaxies and globular clusters may contain intermediate mass black holes ($10^{3}$ to $10^{5}$ solar masses) in their cores. Estimates of ~$10^{3}$ neutron stars in the central parsec of the Galaxy and similar numbers in small elliptical galaxies and globular clusters along with an estimated high probability of ms-pulsar formation in those environments has led many workers to propose the use of ms-pulsar timing to measure the mass and spin of intermediate mass black holes. Models of pulsar motion around a rotating black hole generally assume geodesic motion of a "test" particle in the Kerr metric. These approaches account for well-known effects like de Sitter precession and the Lense-Thirring effect but they do not account for the non-linear effect of the pulsar's stress-energy tensor on the space-time metric. Here we model the motion of a pulsar near a black hole with the Mathisson-Papapetrou-Dixon (MPD) equations. Numerical integration of the MPD equations for black holes of mass 2 X $10^{6}$, $10^{5}$ and $10^{3}$ solar masses shows that the pulsar will not remain in an orbital plane with motion vertical to the plane being largest relative to the orbit's radial dimensions for the lower mass black holes. The pulsar's out of plane motion will lead to timing variations that are up to ~10 microseconds different from those predicted by planar orbit models. Such variations might be detectable in long term observations of millisecond pulsars. If pulsar signals are used to measure the mass and spin of intermediate mass black holes on the basis of dynamical models of the received pulsar signal then the out of plane motion of the pulsar should be part of that model.
Sahni and Singh settled a problem posed by Yousefi & Hesameddini by generalizing their main result with a simple proof in the setting of sub-Hilbert spaces in $H^2(T)$. In this paper, we extend the main result of Sahni and Singh to the setting of the Banach spaces $H^p$, $1\le p\le\infty$ on the circle and the torus.
In this paper we characterize for 0 < p \leq \infty, the closed subspaces of Hp that are invariant under multiplication by all powers of a finite Blaschke factor B, except the first power. Our result clearly generalizes the invariant subspace theorem obtained by Paulsen and Singh [9] which has proved to be the starting point of important work on constrained Nevanlinna-Pick interpolation. Our method of proof can also be readily adapted to the case where the subspace is invariant under all positive powers of B (z). The two results are in the mould of the classical Lax-Halmos Theorem and can be said to be Lax-Halmos type results in the finitre multiplicity case for two commuting shifts and for a single shift respectively.
A fundamentally different approach to path integral quantum mechanics in curved space-time is presented, as compared to the standard approaches currently available in the literature. Within the context of scalar particle propagation in a locally curved background, such as described by Fermi or Riemann normal co-ordinates, this approach requires use of a constructed operator to rotate the initial, intermediate, and final position ket vectors onto their respective local tangent spaces, defined at each local time step along some arbitrary classical reference worldline. Local time translation is described using a quantum mechanical representation of Lie transport, that while strictly non-unitary in operator form, nevertheless correctly recovers the free-particle Lagrangian in curved space-time, along with new contributions. This propagator yields the prediction that all probability violating terms due to curvature contribute to a quantum violation of the weak equivalence principle, while the remaining terms that conserve probability also correspondingly satisfy the weak equivalence principle, at least to leading-order in the particle's Compton wavelength. Furthermore, this propagator possesses an overall curvature-dependent and gauge-invariant phase factor that can be interpreted as the gravitational Aharonov-Bohm effect and Berry's phase.
This paper claims that local space-time curvature can non-trivially contribute to the properties of orbital angular momentum in quantum mechanics. Of key importance is the demonstration that an extended orbital angular momentum operator due to gravitation can identify the existence of orbital states with half-integer projection quantum numbers "m" along the axis of quantization, while still preserving integer-valued orbital quantum numbers "l" for a simply connected topology. The consequences of this possibility are explored in depth, noting that the half-integer "m" states vanish as required when the locally curved space-time reduces to flat space-time, fully recovering all established properties of orbital angular momentum in this limit. In particular, it is shown that a minimum orbital number of "l = 2" is necessary for the gravitational interaction to appear within this context, in perfect correspondence with the spin-2 nature of linearized general relativity.