Searcharxiv⌕ Search

arXiv subjects

Moinul Hossain Rahat

Publications and source records attributed to Moinul Hossain Rahat.

At least 19 recordsLinked to original sources
arXiv2026

Litespark Inference For CPUs: Ultra-Fast SIMD Framework for Ternary (1.58-bit) Language Models

Large language models (LLMs) have transformed artificial intelligence, but their computational requirements remain prohibitive for most users. Standard inference demands expensive datacenter GPUs or cloud API access, leaving over one billion personal computers underutilized for AI workloads. Ternary models offer a path forward: their weights are constrained to {-1, 0, +1}, theoretically eliminating the need for floating-point multiplication. However, existing frameworks fail to exploit this structure, treating ternary models as dense floating-point networks. We address this gap with custom SIMD kernels that replace matrix multiplication with simple addition and subtraction operations, targeting the integer dot product instructions available on modern CPUs. Our implementation, Litespark-Inference, is pip-installable and integrates directly with Hugging-Face, achieving 18.15x higher throughput, 7.15x faster time-to-first-token and 6.03x memory reduction compared to standard PyTorch inference on Apple Silicon, with comparable or higher throughput speedups up to 95.81x on Intel and AMD processors.

cs.CL↗
arXiv2026

SpinGQE: A Generative Quantum Eigensolver for Spin Hamiltonians

The ground state search problem is central to quantum computing, with applications spanning quantum chemistry, condensed matter physics, and optimization. The Variational Quantum Eigensolver (VQE) has shown promise for small systems but faces significant limitations. These include barren plateaus, restricted ansatz expressivity, and reliance on domain-specific structure. We present SpinGQE, an extension of the Generative Quantum Eigensolver (GQE) framework to spin Hamiltonians. Our approach reframes circuit design as a generative modeling task. We employ a transformer-based decoder to learn distributions over quantum circuits that produce low-energy states. Training is guided by a weighted mean-squared error loss between model logits and circuit energies evaluated at each gate subsequence. We validate our method on the four-qubit Heisenberg model, demonstrating successfulconvergencetonear-groundstates. Throughsystematichyperparameterexploration, we identify optimal configurations: smaller model architectures (12 layers, 8 attention heads), longer sequence lengths (12 gates), and carefully chosen operator pools yield the most reliable convergence. Our results show that generative approaches can effectively navigate complex energy landscapes without relying on problem-specific symmetries or structure. This provides a scalable alternative to traditional variational methods for general quantum systems. An open-source implementation is available at https://github.com/Mindbeam-AI/SpinGQE.

quant-ph↗
arXiv2025

Litespark Technical Report: High-Throughput, Energy-Efficient LLM Training Framework

Training Large Language Models (LLMs) is plagued by long training times and massive energy consumption, with modern models requiring months of computation and gigawatt-hours of electricity. In light of these challenges,we introduce Litespark, a novel pre-training framework that addresses these inefficiencies through targeted optimizations to transformer attention and MLP layers. Our approach combines architectural improvements with algorithmic enhancements to maximize Model FLOPs Utilization (MFU) while maintaining compatibility with standard transformer implementations. Comprehensive benchmarking on 3B and 30B parameter Llama models using the SlimPajama-627B dataset demonstrates substantial performance gains: 2x-6x training throughput improvement and $55\%-83$% energy consumption reduction across multi-node H200 GPU clusters. These optimizations are model- and hardware-agnostic, enabling broad applicability across transformer architectures and extending to post-training phases including supervised fine-tuning and direct preference optimization.

cs.LG↗
arXiv2024

Type-I two-Higgs-doublet model and gravitational waves from domain walls bounded by strings

The spontaneous breaking of a $U(1)$ symmetry via an intermediate discrete symmetry may yield a hybrid topological defect of \emph{domain walls bounded by cosmic strings}. The decay of this defect network leads to a unique gravitational wave signal spanning many orders in observable frequencies, that can be distinguished from signals generated by other sources. We investigate the production of gravitational waves from this mechanism in the context of the type-I two-Higgs-doublet model extended by a $U(1)_R$ symmetry, that simultaneously accommodates the seesaw mechanism, anomaly cancellation, and eliminates flavour-changing neutral currents. The gravitational wave spectrum produced by the string-bounded-wall network can be detected for $U(1)_R$ breaking scale from $10^{12}$ to $10^{15}$ GeV in forthcoming interferometers including LISA and Einstein Telescope, with a distinctive $f^{3}$ slope and inflexion in the frequency range between microhertz and hertz.

hep-ph↗
arXiv2024

The split majoron model confronts the NANOGrav signal and cosmological tensions

In the light of the evidence of a gravitational wave background from the NANOGrav 15yr data set, we reconsider the split majoron model as a new physics extension of the standard model able to generate a needed contribution to solve the current tension between the data and the standard interpretation in terms of inspiraling supermassive black hole massive binaries. In the split majoron model the seesaw right-handed neutrinos acquire Majorana masses from spontaneous symmetry breaking of global $U(1)_{B-L}$ in a strong first order phase transition of a complex scalar field occurring above the electroweak scale. The final vacuum expectation value couples to a second complex scalar field undergoing a low scale phase transition occurring after neutrino decoupling. Such a coupling enhances the strength of this second low scale first order phase transition and can generate a sizeable primordial gravitational wave background contributing to the NANOGrav 15yr signal. Some amount of extra-radiation is generated after neutron-to-proton ration freeze-out but prior to nucleosynthesis. This can be either made compatible with current upper bound from primordial deuterium measurements or even be used to solve a potential deuterium problem. Moreover, the free streaming length of light neutrinos can be suppressed by their interactions with the resulting majoron background and this mildly ameliorates existing cosmological tensions. Thus cosmological observations nicely provide independent motivations for the model.

hep-ph↗
arXiv2024

Electric Dipole Moments as indirect probes of Dark Sectors

Dark sectors provide beyond Standard Model scenarios which can address unresolved puzzles, such as the observed dark matter abundance or the baryon asymmetry of the Universe. A naturally small portal to the dark sector is obtained if dark-sector interactions stem from a non-Abelian hidden gauge group that couples through kinetic mixing with the hypercharge boson. In this work, we investigate the phenomenology of such a portal of dimension five in the presence of CP violation, focusing on its signatures in fermion electric dipole moments. We show that, currently unbounded regions of the parameter space from dark photon searches can be indirectly probed with upcoming electron dipole moment experiments for dark boson masses in the range $1-100$ GeV. We also discuss two particular scenarios where a $SU(2)_D$ dark gauge group spontaneously breaks into either an Abelian $U(1)_D$ or nothing. In both cases, we show that potentially observable electron dipole moments can be produced in vast regions of the parameter space compatible with current experimental constraints and observed dark matter abundance.

hep-ph↗
arXiv2024

Gravitational waves from phase transitions and cosmic strings in neutrino mass models with multiple majorons

We explore the origin of Majorana masses within the majoron model and how this can lead to the generation of a distinguishable primordial stochastic background of gravitational waves. We first show how in the simplest majoron model only a contribution from cosmic string can be within the reach of planned experiments. We then consider extensions containing multiple complex scalars, demonstrating how in this case a spectrum comprising contributions from both a strong first order phase transition and cosmic strings can naturally emerge. We show that the interplay between multiple scalar fields can amplify the phase transition signal, potentially leading to double peaks over the wideband sloped spectrum from cosmic strings. We also underscore the possibility of observing such a gravitational wave background to provide insights into the reheating temperature of the universe. We conclude highlighting how the model can be naturally combined with scenarios addressing the origin of matter of the universe, where baryogenesis occurs via leptogenesis and a right-handed neutrino plays the role of dark matter.

hep-ph↗
arXiv2024

Towards distinguishing Dirac from Majorana neutrino mass with gravitational waves

We propose a new method towards distinguishing the Dirac versus Majorana nature of neutrino masses from the spectrum of gravitational waves (GWs) associated with neutrino mass genesis. Motivated by the principle of generating small neutrino masses without tiny Yukawa couplings, we assume generic seesaw mechanisms for both Majorana and Dirac neutrino masses. For Majorana neutrinos, we further assume a spontaneously broken gauged $U(1)_{B-L}$ symmetry, independently of the type of Majorana seesaw mechanism, which gives a cosmic string induced GW signal flat over a wide range of frequencies. For Dirac neutrinos, we assume the spontaneous breaking of a $Z_2$ symmetry, the minimal symmetry choice associated with all Dirac seesaw mechanisms, which is softly broken, generating a peaked GW spectrum from the annihilation of the resulting domain walls. In fact, the GW spectra for all types of Dirac seesaws with such a broken $Z_2$ symmetry are identical, subject to a mild caveat. As an illustrative example, we study the simplest respective type-I seesaw mechanisms, and show that the striking difference in the shapes of the GW spectra can help differentiate between these Dirac and Majorana seesaws, complementing results of neutrinoless double beta decay experiments. We also discuss detailed implications of the recent NANOGrav data for Majorana and Dirac seesaw models.

hep-ph↗
arXiv2023

NANOGrav signal from axion inflation

Several pulsar timing arrays including NANOGrav, EPTA, PPTA, and CPTA have recently reported the observation of a stochastic background of gravitational wave spectrum in the nano-Hz frequencies. An inflationary interpretation of this observation is challenging from various aspects. We report that such a signal can arise from the Chern-Simons coupling in axion inflation, where a pseudoscalar inflaton couples to a (massive) $U(1)$ gauge field, leading to efficient production of a transverse gauge mode. Such tachyonic particle production during inflation exponentially enhances the primordial perturbations and leads to a unique parity-violating gravitational wave spectrum, that remains flat near the CMB scales but becomes blue-tilted at smaller scales. We identify the parameter space consistent with various cosmological constraints and show that the resultant gravitational wave signals can provide extra contribution on top of the standard astrophysical contribution from inspiraling supermassive black hole binaries towards explaining the observed excess at NANOGrav. The parity-violating nature of the signal can be probed in future interferometers, distinguishing it from most other new physics signals attempting to explain the NANOGrav result.

hep-ph↗
arXiv2023

Flavon vacuum alignment beyond SUSY

In flavor models the vacuum alignment of flavons is typically achieved via the $F$-terms of certain fields in the supersymmetric limit. We propose a method for preserving such alignments, up to a rescaling of the vacuum expectation values, even after supersymmetry (and the flavor symmetry) are softly broken, facilitating the vacuum alignment in models which are non-supersymmetric at low energies. Examples of models with different flavor groups, namely $A_4$, $T_7$, $S_4$ and $Δ(27)$, are discussed.

hep-ph↗
arXiv2023

Primordial non-Gaussianity as a probe of seesaw and leptogenesis

We present the possibility that the seesaw mechanism and nonthermal leptogenesis can be {investigated} via primordial non-Gaussianities in the context of a majoron curvaton model. Originating as a massless Nambu-Goldstone boson from the spontaneous breaking of the global baryon ($B$) minus lepton ($L$) number symmetry at a scale $v_{B-L}$, majoron becomes massive when it couples to a new confining sector through anomaly. Acting as a curvaton, majoron produces the observed red-tilted curvature power spectrum without relying on any inflaton contribution, and its decay in the post-inflationary era gives rise to a nonthermal population of right-handed neutrinos that participate in leptogenesis. A distinctive feature of the mechanism is the generation of observable non-Gaussianity, {in the parameter space where the red-tilted power spectrum and sufficient baryon asymmetry are produced.} We {find} that the non-Gaussianity parameter $f_{\rm NL} \gtrsim \mathcal{O} (0.1)$ is produced for high-scale seesaw ($v_{B-L}$ at $\mathcal{O}(10^{14-17})$ GeV) and leptogenesis ($M_1 \gtrsim \mathcal{O}(10^6)$ GeV) where the latter represents the lightest right-handed neutrino mass. While the current bounds on local non-Gaussianity excludes some part of parameter space, the rest can be fully probed by future experiments like CMB-S4, LSST, and 21 cm tomography.

hep-ph↗
arXiv2022

Parity-Odd and Even Trispectrum from Axion Inflation

The four-point correlation function of primordial scalar perturbations has parity-even and parity-odd contributions and the parity-odd signal in cosmological observations is opening a novel window to look for new physics in the inflationary epoch. We study the distinct parity-odd and even prediction from the axion inflation model, in which the inflaton couples to a vector field via a Chern-Simons interaction, and the vector field is considered to be either approximately massless ($m_A \ll $ Hubble scale $H$) or very massive ($m_A \sim H $). The parity-odd signal arises due to one transverse mode of the vector field being predominantly produced during inflation. We adopt the in-in formalism to evaluate the correlation functions. Considering the vector field mode function to be dominated by its real part up to a constant phase, we simplify the formulas for numerical computations. The numerical studies show that the massive and massless vector fields give significant parity-even signals, while the parity-odd contribution is about one to two orders of magnitude smaller.

hep-ph↗
arXiv2022

Gravitational Wave Probes of Massive Gauge Bosons at the Cosmological Collider

We extend the reach of the ``cosmological collider'' for massive gauge boson production during inflation from the CMB scales to the interferometer scales. Considering a Chern-Simons coupling between the gauge bosons and the pseudoscalar inflaton, one of the transverse gauge modes is efficiently produced and its inverse decay leaves an imprint in the primordial scalar and tensor perturbations. We study the correlation functions of these perturbations and derive the updated constraints on the parameter space from CMB observables. We then extrapolate the tensor power spectrum to smaller scales consistently taking into account the impact of the gauge field on inflationary dynamics. Our results show that the presence of massive gauge fields during inflation can be detected from characteristic gravitational wave signals encompassing the whole range of current and planned interferometers.

hep-ph↗
arXiv2022

BBN Photodisintegration Constraints on Gravitationally Produced Vector Bosons

Gravitational production of massive particles due to cosmic expansion can be significant during the inflationary and reheating period of the Universe. If the particle also has non-gravitational interactions that do not significantly affect its production, numerous observational probes open up, including cosmological probes. In this work, we focus on the gravitational production of light vector bosons that couple feebly to the Standard Model (SM) particles. Due to the very feeble coupling, the light vector bosons never reach thermal equilibrium, and if the Hubble scale at the end of inflation is above $10^8$ GeV, the gravitational production can overwhelm the thermal production via the freeze-in mechanism by many orders of magnitude. As a result, much stronger constraints from the Big Bang Nucleosynthesis (BBN) can be placed on the lifetime and mass of the vector bosons compared to the scenario where only thermal production is considered. As an example, we study the sub-GeV scale dark photons, which couple to the SM only through kinetic mixing, and derive constraints on the mass and kinetic mixing parameter of the dark photon from the photodisintegration effects on the light element abundances relevant at the end of the BBN when the cosmic age was around $10^4$ s.

hep-ph↗
arXiv2022

Constraining Low-Scale Flavor Models with $\boldsymbol{(g-2)_μ}$ and Lepton Flavor Violation

We present here two concrete examples of models where a sub-TeV scale breaking of their respective $\mathcal{T}_{13}$ and $A_5$ flavor symmetries is able to account for the recently observed discrepancy in the muon anomalous magnetic moment, $(g-2)_μ$. Similarities in the flavor structures of the charged-lepton Yukawa matrix and dipole matrix yielding $(g-2)_μ$ give rise to strong constraints on low-scale flavor models when bounds from lepton flavor violation (LFV) are imposed. These constraints place stringent limits on the off-diagonal Yukawa structure, suggesting a mostly (quasi-)diagonal texture for models with a low flavor breaking scale $Λ_f$. We argue that many of the popular flavor models in the literature designed to explain the fermion masses and mixings are not suitable for reproducing the observed discrepancy in $(g-2)_μ$, which requires a delicate balance of maintaining a low flavor scale while simultaneously satisfying strong LFV constraints.

hep-ph↗
arXiv2021

Low-scale Resonant Leptogenesis in $SU(5)$ GUT with $\mathcal{T}_{13}$ Family Symmetry

We investigate low-scale resonant leptogenesis in an $SU(5) \times \mathcal{T}_{13}$ model where a single high energy phase in the complex Tribimaximal seesaw mixing produces the yet-to-be-observed low energy Dirac and Majorana ${CP}$ phases. A fourth right-handed neutrino, required to generate viable light neutrino masses within this scenario, also turns out to be necessary for successful resonant leptogenesis where $CP$ asymmetry is produced by the same high energy phase. We derive a lower bound on the right-handed neutrino mass spectrum in the $\mathrm{GeV}$ range, where part of the parameter space, although in tension with Big Bang Nucleosynthesis constrains, can be probed in planned high intensity experiments like DUNE. We also find the existence of a curious upper bound ($\text{TeV}$-scale) on the right-handed neutrino mass spectrum in majority of the parameter space due to significant washout of the resonant asymmetry by lighter right-handed neutrinos. While in most of the parameter space of our model classical Boltzmann equations are sufficient, when right-handed neutrino masses are below the electroweak sphalerons freeze-out temperature we resort to the more general density matrix formalism to take into account decays and oscillations of the right-handed neutrinos as well as their helicities which are relevant when they are relativistic.

hep-ph↗
arXiv2021

Leptogenesis from the Asymmetric Texture

We investigate non-resonant thermal leptogenesis in the context of the $SU(5) \times \mathcal{T}_{13}$ "asymmetric texture", where both Dirac and Majorana ${CP}$ violation arise from a single phase in the tribimaximal seesaw mixing matrix. We show that the baryon asymmetry of the universe can be explained in this model only when flavor effects are considered for right-handed neutrino masses of $\mathcal{O}(10^9 - 10^{12})$ $\text{GeV}$. The sign of the baryon asymmetry also determines the sign of the previously predicted Dirac $\require{cancel}\cancel{CP}$ phase $|δ_{CP}| = 1.32π$, consistent with the latest global fit $δ_{CP}^{PDG} = 1.37 \pm 0.17π$.

hep-ph↗
arXiv2020

Tribimaximal Mixing in the $SU(5) \times \mathcal{T}_{13}$ Texture

We extend the recently proposed $SU(5) \times \mathcal{T}_{13}$ model for the asymmetric texture to the up-type quark and seesaw sectors. The hierarchical up-type quark masses are generated from higher-dimensional operators involving family-singlet Higgses, gauge-singlet familons, and vectorlike messengers. The complex-tribimaximal (TBM) seesaw mixing arises from the vacuum structure of a minimal number of familons, resulting in an alignment between the Yukawa and Majorana matrices of the seesaw formula. Introducing four right-handed neutrinos, normal ordering of the light neutrino masses is obtained, with $m_{ν_1} = 27.6\ \mathrm{meV}$, $m_{ν_2} = 28.9\ \mathrm{meV}$ and $m_{ν_3} = 57.8\ \mathrm{meV}$. Their sum almost saturates Planck's cosmological upper bound ($120$ $\text{meV}$). The right-handed neutrino masses are expressed in terms of two parameters for a particular choice of familon vacuum alignment. We predict the $\require{cancel}\cancel{CP}$ Jarlskog-Greenberg invariant to be $|\mathcal{J}| = 0.028$, consistent with the current PDG estimate, and Majorana invariants $|\mathcal{I}_1| = 0.106$ and $|\mathcal{I}_2| = 0.011$. A sign ambiguity in the model parameters leads to two possibilities for the invariant mass parameter $|m_{ββ}|$: $13.02$ or $25.21$ $\text{meV}$, both within an order of magnitude of the most rigorous experimental upper limit ($61$--$165$ $\text{meV}$).

hep-ph↗