Gorgi Pavlov
Gorgi Pavlov has a total of 2 co-authors. Here are the top co-authors by number of shared publications:
James T. Hsu
1 shared publication- James T. Hsua
Dept. of Chemical and Biomolecular Engineering Lehigh University Bethlehem PA 18015
1 shared publication
Latest Publications
- Open Access
- 6 August 2026
- Published by Elsevier in SSRN Electronic Journal
Excess Movement in High-frequency Prediction-market Quotes: The Entry-delay Frontier
High-frequency prediction-market studies often treat displayed quotes as prices one could trade at. We show that this can produce both excess movement and what looks like profit from mean reversion. In 7,118 Polymarket sports contracts, the aggregate ratio of realized variance to the binary variance budget is 1.53 at one-minute sampling, but 1.05 at sixty minutes. We regroup the known finite-grid identity into two parts, initial calibration and the cross-products of returns, and then split the second part by lag. Miscalibration is small. The excess sits at short lags, and it rejects a joint null: quote error with zero mean and no dependence over time, a martingale fair value, and error orthogonal to that fair value. Under that null the return cross-moment must vanish beyond lag one, whatever the error variance does. We measure it as non-zero at lag two. Serially correlated error remains viable. We then price a ten-point one-minute midpoint fade against 1,153,546 transaction records. Across 36 entry-delay and holding-horizon policies, applied to 3,000 signals in 396 markets, every estimated policy P&L is negative. The best is-0.327 points [-0.70, +0.04], and simultaneous inference does not exclude zero. Among tape-covered signals, 39% carry no transaction at all, and their midpoint reversion is 5.234 points against 1.069 when an entry print exists. High-frequency excess movement on thin books can therefore reflect quote dynamics alone. Those dynamics need not stand for transactions, nor for returns one could have earned at observed transaction prices.
- 6 August 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessExcess Movement in High-frequency Prediction-market Quotes: The Entry-delay Frontier
High-frequency prediction-market studies often treat displayed quotes as prices one could trade at. We show that this can produce both excess movement and what looks like profit from mean reversion. In 7,118 Polymarket sports contracts, the aggregate ratio of realized variance to the binary variance budget is 1.53 at one-minute sampling, but 1.05 at sixty minutes. We regroup the known finite-grid identity into two parts, initial calibration and the cross-products of returns, and then split the second part by lag. Miscalibration is small. The excess sits at short lags, and it rejects a joint null: quote error with zero mean and no dependence over time, a martingale fair value, and error orthogonal to that fair value. Under that null the return cross-moment must vanish beyond lag one, whatever the error variance does. We measure it as non-zero at lag two. Serially correlated error remains viable. We then price a ten-point one-minute midpoint fade against 1,153,546 transaction records. Across 36 entry-delay and holding-horizon policies, applied to 3,000 signals in 396 markets, every estimated policy P&L is negative. The best is-0.327 points [-0.70, +0.04], and simultaneous inference does not exclude zero. Among tape-covered signals, 39% carry no transaction at all, and their midpoint reversion is 5.234 points against 1.069 when an entry print exists. High-frequency excess movement on thin books can therefore reflect quote dynamics alone. Those dynamics need not stand for transactions, nor for returns one could have earned at observed transaction prices.
- Open Access
- 26 June 2026
- Published by Elsevier in SSRN Electronic Journal
Moving-Wall Fidelity in Lattice-Boltzmann Simulations:An Integratedness Gradient, and a Near-Wall Sign Set by Discretization
The fidelity framework of Papers~1--2 quantified how faithfully astatic lattice-Boltzmann (LBM) wall is represented; it stopped atthe wall that moves --- the impeller blade, the piston, the rotatingwall whose pumping a stirred-tank or pump simulation exists to predict.Extending it to the moving wall, we find a moving-wall observable'ssensitivity to the boundary operator is graded by how much of the wall itintegrates: an integratedness gradient. The wall source isunchanged --- a tangentially moving wall injects the Paper~1 ghost-energyinvariant $5/18$ per $\Ma^2$, a velocity-set invariant identical for everyLadd-form bounce-back operator, verified in OpenLB. The near-wallresponse carries the surprise: the sign of the near-wallvelocity defect --- the quantity read as ``slip'' --- is adiscretization property, not a kinetic signature of the operator. Itdoes not flip at the slip-free magic relaxation value, is set by thevoxelization effective wall radius, and rides a lattice-locked azimuthalmodulation exceeding its own mean; the apparent cross-code ``signreversal'' of a shared operator is a sampling artifact, withfloat/double and FP16/FP32 controls excluding precision. Theintegrated pump, by contrast, is operator-robust --- therotating-disk pump to within grid scatter, the cylinder torque by momentumexchange to an operator spread of only $\sim\!5\times$. But the sametorque extracted from the near-wall velocity gradient appears fragile by30x: an integrated observable's gradient rung can be set by itsextraction convention. We package the moving-wall channel as a pumpingfidelity Fpump; all rungs clear the F>100 gate, so the gate isnecessary but not sufficient. A symmetric-motion null lemma fixes why avelocity-free geometric advance cannot represent rotation. Practical rule:integrated throughput transfers across operators and codes, but thenear-wall channel must be calibrated per operator and perdiscretization.- 26 June 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessMoving-Wall Fidelity in Lattice-Boltzmann Simulations:An Integratedness Gradient, and a Near-Wall Sign Set by Discretization
The fidelity framework of Papers~1--2 quantified how faithfully astatic lattice-Boltzmann (LBM) wall is represented; it stopped atthe wall that moves --- the impeller blade, the piston, the rotatingwall whose pumping a stirred-tank or pump simulation exists to predict.Extending it to the moving wall, we find a moving-wall observable'ssensitivity to the boundary operator is graded by how much of the wall itintegrates: an integratedness gradient. The wall source isunchanged --- a tangentially moving wall injects the Paper~1 ghost-energyinvariant $5/18$ per $\Ma^2$, a velocity-set invariant identical for everyLadd-form bounce-back operator, verified in OpenLB. The near-wallresponse carries the surprise: the sign of the near-wallvelocity defect --- the quantity read as ``slip'' --- is adiscretization property, not a kinetic signature of the operator. Itdoes not flip at the slip-free magic relaxation value, is set by thevoxelization effective wall radius, and rides a lattice-locked azimuthalmodulation exceeding its own mean; the apparent cross-code ``signreversal'' of a shared operator is a sampling artifact, withfloat/double and FP16/FP32 controls excluding precision. Theintegrated pump, by contrast, is operator-robust --- therotating-disk pump to within grid scatter, the cylinder torque by momentumexchange to an operator spread of only $\sim\!5\times$. But the sametorque extracted from the near-wall velocity gradient appears fragile by30x: an integrated observable's gradient rung can be set by itsextraction convention. We package the moving-wall channel as a pumpingfidelity Fpump; all rungs clear the F>100 gate, so the gate isnecessary but not sufficient. A symmetric-motion null lemma fixes why avelocity-free geometric advance cannot represent rotation. Practical rule:integrated throughput transfers across operators and codes, but thenear-wall channel must be calibrated per operator and perdiscretization. - Open Access
- 10 June 2026
- Published by Elsevier in SSRN Electronic Journal
Fidelity Numbers for Lattice-Boltzmann Validity: A Signal-to-Noise Framework
Every lattice-Boltzmann simulation is a layered signal-to-noise problem with several distinct noise channels: lattice thermal fluctuation, modified-equation truncation bias, non-conserved "ghost" kinetic-mode contamination, and floating-point representation noise. We propose a hierarchy of four fidelity numbers, F = (F_th, F_tr, F_gh, F_fp), that diagnose these channels independently. Each is a dimensionless validity indicator constructed from a deterministic LBM signal and an independent noise-source amplitude at the target scale; we adopt a common engineering threshold F > 100 calibrated to roughly 10× amplitude separation per channel (the precise amplitude-to-power convention varies per channel). In the conservative engineering convention adopted here, a simulation is considered fidelity-clean only when all active channels satisfy F > 100. The framework's main results are: (i) a validated closed-form BGK bulk hyperviscosity law α_bulk^BGK(τ_+) = τ_+(1 - τ_+)/3, with sign flip at the BGK magic-τ (τ_+ = 1), confirmed via low-Mach Kolmogorov decay-rate fits on D2Q9 (the derivation applies uniformly to D2Q9/D3Q15/D3Q19/D3Q27); (ii) an empirical Λ_C-tunable TRT bulk-cancellation locus (slope ∂α/∂Λ_C ≈ -0.67), distinct from the Ginzburg–d'Humières wall-slip magic Λ_C = 3/16; (iii) a channel-activation framework (C_1, C_2, C_3) that classifies canonical LBM benchmarks by which ghost-injection pathways they exercise, showing that analytically clean benchmarks are typically null to at least one channel and that all-channel tests such as lid-driven cavity sacrifice channel isolation; (iv) a pre-run validity calculator that flags industrial BGK-LBM operating points (e.g. Co = 0.1 on Re ~ 10^4 flows) crossing the local kinetic wall-channel fidelity ceiling at typical production resolutions. The face-HBB invariant ⟨P²⟩_tan = 2 c_s^4 - c_s^6, derived in a companion paper, enters the local ghost fidelity F_gh and, through the hydrodynamic projection, the observable wall-slip fidelity F_slip; here it is used only as the imported C_2 wall-channel input. Calibrations include χ = 0.109 ± 0.003 for D3Q7 BGK scalar truncation, β_BC^(1/2) ≈ 1.3 × 10^-7 per √step for D2Q9/FP32 halfway-BB boundary residuals, and a three-regime F_fp classifier for 16-bit GPU-LBM formats.
- 10 June 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessFidelity Numbers for Lattice-Boltzmann Validity: A Signal-to-Noise Framework
Every lattice-Boltzmann simulation is a layered signal-to-noise problem with several distinct noise channels: lattice thermal fluctuation, modified-equation truncation bias, non-conserved "ghost" kinetic-mode contamination, and floating-point representation noise. We propose a hierarchy of four fidelity numbers, F = (F_th, F_tr, F_gh, F_fp), that diagnose these channels independently. Each is a dimensionless validity indicator constructed from a deterministic LBM signal and an independent noise-source amplitude at the target scale; we adopt a common engineering threshold F > 100 calibrated to roughly 10× amplitude separation per channel (the precise amplitude-to-power convention varies per channel). In the conservative engineering convention adopted here, a simulation is considered fidelity-clean only when all active channels satisfy F > 100. The framework's main results are: (i) a validated closed-form BGK bulk hyperviscosity law α_bulk^BGK(τ_+) = τ_+(1 - τ_+)/3, with sign flip at the BGK magic-τ (τ_+ = 1), confirmed via low-Mach Kolmogorov decay-rate fits on D2Q9 (the derivation applies uniformly to D2Q9/D3Q15/D3Q19/D3Q27); (ii) an empirical Λ_C-tunable TRT bulk-cancellation locus (slope ∂α/∂Λ_C ≈ -0.67), distinct from the Ginzburg–d'Humières wall-slip magic Λ_C = 3/16; (iii) a channel-activation framework (C_1, C_2, C_3) that classifies canonical LBM benchmarks by which ghost-injection pathways they exercise, showing that analytically clean benchmarks are typically null to at least one channel and that all-channel tests such as lid-driven cavity sacrifice channel isolation; (iv) a pre-run validity calculator that flags industrial BGK-LBM operating points (e.g. Co = 0.1 on Re ~ 10^4 flows) crossing the local kinetic wall-channel fidelity ceiling at typical production resolutions. The face-HBB invariant ⟨P²⟩_tan = 2 c_s^4 - c_s^6, derived in a companion paper, enters the local ghost fidelity F_gh and, through the hydrodynamic projection, the observable wall-slip fidelity F_slip; here it is used only as the imported C_2 wall-channel input. Calibrations include χ = 0.109 ± 0.003 for D3Q7 BGK scalar truncation, β_BC^(1/2) ≈ 1.3 × 10^-7 per √step for D2Q9/FP32 halfway-BB boundary residuals, and a three-regime F_fp classifier for 16-bit GPU-LBM formats.
- Open Access
- 12 May 2026
- Published by Elsevier in Computer Physics Communications
- Vol. 326
moljax: GPU-accelerated method of lines for stiff reaction-diffusion PDEs with FFT preconditioning
- 12 May 2026
- Published by Elsevier in Computer Physics Communications
- Vol. 326
Open Accessmoljax: GPU-accelerated method of lines for stiff reaction-diffusion PDEs with FFT preconditioning
- RetractedOpen Access
- 12 May 2026
- Published by Elsevier in Computers & Chemical Engineering
- Vol. 212
Extending FFT-based numerical inverse Laplace transform to nonlinear chromatography via iterative re-linearization
- 12 May 2026
- Published by Elsevier in Computers & Chemical Engineering
- Vol. 212
RetractedOpen AccessExtending FFT-based numerical inverse Laplace transform to nonlinear chromatography via iterative re-linearization
- Open Access
- 9 May 2026
- Published by Elsevier in SSRN Electronic Journal
Universal wall excitation and collision-model response of kinetic modes in lattice-Boltzmann flows
Solid boundaries in lattice-Boltzmann (LBM) simulations do not only impose hydrodynamic constraints; they also inject non-conserved kinetic population components. We derive a lattice-level invariant for the tangential ghost-mode energy injected by face-centered halfway-bounce-back walls. For every standard nearest-neighbor single-speed D3QN momentum lattice satisfying second- and fourth-moment isotropy and the velocity-component bound |c_{i,α}| ∈ {0,1}, the invariant is I_wall ≡ Σ_{k ∈ non-cons.} (P_k^α)^2 = 2 c_s^4 - c_s^6,giving exactly 5/27 for the standard c_s^2 = 1/3 family (Theorem 1). The invariant is collision-model independent; the same source feeds BGK and TRT through different kinetic-response operators R_g that damp the injected modes, and the decomposition extends naturally to MRT and regularized collisions in an orthonormal moment basis. This source/response decomposition yields the closed-form local ghost-energy formula ε_g = (Ma^2 / 2 c_s^2) ‖ R_g Q_g B_w ‖^2,which reduces to ε_g = r^2 Ma^2 I_wall / (2 c_s^2) when the ghost response is scalar (R_g = r I), and gives an explicit BGK refinement ceiling under fixed-(Ma, Re) acoustic scaling, N_crit^F = (Re c_s / Ma) [ (1/Ma) sqrt(2 / (F (2 - c_s^2))) - 1/2 ],exact within the leading-order analysis. TRT and MRT modify the local response by decoupling physical and non-conserved relaxation; for the standard D3QN family the wall source splits into universal symmetric and antisymmetric parity sectors of weights 3/27 and 2/27, so TRT improves the local kinetic ghost fidelity over BGK by a constant factor of 5/3 asymptotically rather than removing the τ_+^2 growth entirely. Operator-level calculations on D3Q15, D3Q19, and D3Q27 confirm the invariant and the parity split, and a full D3Q15/D3Q19/D3Q27 × BGK/TRT/MRT three-dimensional Poiseuille sweep shows that the same universal wall source projects into lattice- and operator-dependent hydrodynamic deficits. At N = 192, MRT reduces the BGK centerline deficit to approximately 50–55% of the BGK value on all three lattices. A BGK-equivalent MRT control collapses with the representative MRT schedule to within 0.30% across the full tested grid, indicating that this body-force observable is insensitive to the detailed higher-kinetic rate schedule over the tested range and instead reflects the full moment-space projection and forcing coupling. The framework thereby converts LBM wall fidelity into a diagnostic decomposition: a universal lattice-fixed source, a tunable collision response, and a geometry/observable projection.
- 9 May 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessUniversal wall excitation and collision-model response of kinetic modes in lattice-Boltzmann flows
Solid boundaries in lattice-Boltzmann (LBM) simulations do not only impose hydrodynamic constraints; they also inject non-conserved kinetic population components. We derive a lattice-level invariant for the tangential ghost-mode energy injected by face-centered halfway-bounce-back walls. For every standard nearest-neighbor single-speed D3QN momentum lattice satisfying second- and fourth-moment isotropy and the velocity-component bound |c_{i,α}| ∈ {0,1}, the invariant is I_wall ≡ Σ_{k ∈ non-cons.} (P_k^α)^2 = 2 c_s^4 - c_s^6,giving exactly 5/27 for the standard c_s^2 = 1/3 family (Theorem 1). The invariant is collision-model independent; the same source feeds BGK and TRT through different kinetic-response operators R_g that damp the injected modes, and the decomposition extends naturally to MRT and regularized collisions in an orthonormal moment basis. This source/response decomposition yields the closed-form local ghost-energy formula ε_g = (Ma^2 / 2 c_s^2) ‖ R_g Q_g B_w ‖^2,which reduces to ε_g = r^2 Ma^2 I_wall / (2 c_s^2) when the ghost response is scalar (R_g = r I), and gives an explicit BGK refinement ceiling under fixed-(Ma, Re) acoustic scaling, N_crit^F = (Re c_s / Ma) [ (1/Ma) sqrt(2 / (F (2 - c_s^2))) - 1/2 ],exact within the leading-order analysis. TRT and MRT modify the local response by decoupling physical and non-conserved relaxation; for the standard D3QN family the wall source splits into universal symmetric and antisymmetric parity sectors of weights 3/27 and 2/27, so TRT improves the local kinetic ghost fidelity over BGK by a constant factor of 5/3 asymptotically rather than removing the τ_+^2 growth entirely. Operator-level calculations on D3Q15, D3Q19, and D3Q27 confirm the invariant and the parity split, and a full D3Q15/D3Q19/D3Q27 × BGK/TRT/MRT three-dimensional Poiseuille sweep shows that the same universal wall source projects into lattice- and operator-dependent hydrodynamic deficits. At N = 192, MRT reduces the BGK centerline deficit to approximately 50–55% of the BGK value on all three lattices. A BGK-equivalent MRT control collapses with the representative MRT schedule to within 0.30% across the full tested grid, indicating that this body-force observable is insensitive to the detailed higher-kinetic rate schedule over the tested range and instead reflects the full moment-space projection and forcing coupling. The framework thereby converts LBM wall fidelity into a diagnostic decomposition: a universal lattice-fixed source, a tunable collision response, and a geometry/observable projection.
- Open Access
- 14 March 2026
- Published by Elsevier in Chemical Engineering Science
- Vol. 328
Systematic parameter selection for FFT-based numerical inverse laplace transform: CFL-informed tuning rules and quality diagnostics
- 14 March 2026
- Published by Elsevier in Chemical Engineering Science
- Vol. 328
Open AccessSystematic parameter selection for FFT-based numerical inverse laplace transform: CFL-informed tuning rules and quality diagnostics
- Open Access
- 26 February 2026
- Published by Elsevier in SSRN Electronic Journal
Extending FFT-based numerical inverse Laplace transform to nonlinear chromatography via adaptive re-linearization
The Fast Fourier Transform based Numerical Inverse Laplace Transform (FFT-NILT) provides significant computational speedup for linear chromatographic transport problems. However, extending FFT-NILT to nonlinear binding isotherms (Langmuir, SMA) has remained an open challenge because the convolution product c·q in the binding kinetics destroys the multiplicative structure required by the Laplace domain. We present an adaptive re-linearization scheme that overcomes this limitation by iteratively updating the effective binding parameters around the current operating concentration. At each step, the transfer function is re-linearized with effective parameters q_max,eff = q_max - q_0(c_0) and k_d,eff = k_d + k_a·c_0, which shifts the kinetic pole further into the left half-plane and improves CFL feasibility. Validation against the Thomas analytical solution for plug-flow adsorption demonstrates that a single re-linearization step reduces the root-mean-square error by a factor of 7 compared to the linear approximation, achieving lower global RMS error than finite-volume discretization with 512 cells on this mildly nonlinear benchmark. For the full General Rate Model with Langmuir and SMA binding, the method produces breakthrough curves within 0.7-7.6% relative L2 error of CADET reference solutions across dilute-to-overloaded conditions, with wall times of approximately 58 ms per evaluation. The method preserves the Laplace-domain transfer-function structure: NILT recovers the complete time-domain solution, while moments are available as analytical byproducts and the explicit parameter dependence of the transfer function enables frequency-domain sensitivity analysis. The approach is implemented as a first-class solver option within the CADET ecosystem, with reference-free mass-balance diagnostics providing convergence verification without external validation.
- 26 February 2026
- Published by Elsevier in SSRN Electronic Journal
Open AccessExtending FFT-based numerical inverse Laplace transform to nonlinear chromatography via adaptive re-linearization
The Fast Fourier Transform based Numerical Inverse Laplace Transform (FFT-NILT) provides significant computational speedup for linear chromatographic transport problems. However, extending FFT-NILT to nonlinear binding isotherms (Langmuir, SMA) has remained an open challenge because the convolution product c·q in the binding kinetics destroys the multiplicative structure required by the Laplace domain. We present an adaptive re-linearization scheme that overcomes this limitation by iteratively updating the effective binding parameters around the current operating concentration. At each step, the transfer function is re-linearized with effective parameters q_max,eff = q_max - q_0(c_0) and k_d,eff = k_d + k_a·c_0, which shifts the kinetic pole further into the left half-plane and improves CFL feasibility. Validation against the Thomas analytical solution for plug-flow adsorption demonstrates that a single re-linearization step reduces the root-mean-square error by a factor of 7 compared to the linear approximation, achieving lower global RMS error than finite-volume discretization with 512 cells on this mildly nonlinear benchmark. For the full General Rate Model with Langmuir and SMA binding, the method produces breakthrough curves within 0.7-7.6% relative L2 error of CADET reference solutions across dilute-to-overloaded conditions, with wall times of approximately 58 ms per evaluation. The method preserves the Laplace-domain transfer-function structure: NILT recovers the complete time-domain solution, while moments are available as analytical byproducts and the explicit parameter dependence of the transfer function enables frequency-domain sensitivity analysis. The approach is implemented as a first-class solver option within the CADET ecosystem, with reference-free mass-balance diagnostics providing convergence verification without external validation.
- 1 April 2018
- Published by Elsevier in Computers & Chemical Engineering
- Vol. 112, 304-315
Modelling the effect of temperature on the gel-filtration chromatographic protein separation
- 1 April 2018
- Published by Elsevier in Computers & Chemical Engineering
- Vol. 112, 304-315
Modelling the effect of temperature on the gel-filtration chromatographic protein separation
- 17 January 2018
- Published by Wiley in AIChE Journal
- Vol. 64 (6) , 1928-1937
The pH, temperature, and protein structure effect on β‐lactoglobulin A and B separation in anion‐exchange chromatography
The effect of pH and temperature on separating a mixture of similar proteins, namely β‐lactoglobulin A (LGA) and β‐lactoglobulin B (LGB) in anion‐exchange chromatography is explored. The proteins carry a slight difference in negative charge at basic pH, providing a separation basis on an Q Sepharose Fast Flow anion‐exchange resin. They were separated at different temperatures and pH values, and the separation factor was evaluated. The experimental results were matched to a theoretical model to compute the equilibrium constant KA. The data shows that an increase in temperature and pH leads to an increase in the retention time of the proteins. The results were correlated with the net charge of the molecule for the separation so that the elution can be simulated for any condition that was studied. The tertiary structures of LGA and LGB are analyzed to illustrate the structure effect on the separation. © 2018 American Institute of Chemical Engineers AIChE J, 64: 1928–1937, 2018
- 17 January 2018
- Published by Wiley in AIChE Journal
- Vol. 64 (6) , 1928-1937
The pH, temperature, and protein structure effect on β‐lactoglobulin A and B separation in anion‐exchange chromatography
The effect of pH and temperature on separating a mixture of similar proteins, namely β‐lactoglobulin A (LGA) and β‐lactoglobulin B (LGB) in anion‐exchange chromatography is explored. The proteins carry a slight difference in negative charge at basic pH, providing a separation basis on an Q Sepharose Fast Flow anion‐exchange resin. They were separated at different temperatures and pH values, and the separation factor was evaluated. The experimental results were matched to a theoretical model to compute the equilibrium constant KA. The data shows that an increase in temperature and pH leads to an increase in the retention time of the proteins. The results were correlated with the net charge of the molecule for the separation so that the elution can be simulated for any condition that was studied. The tertiary structures of LGA and LGB are analyzed to illustrate the structure effect on the separation. © 2018 American Institute of Chemical Engineers AIChE J, 64: 1928–1937, 2018