Search for: All records

We propose the generalized initial velocity sampling algorithm at a transition state, in which the total initial kinetic energy and extra positive initial velocity along the reaction coordinate have been introduced to improve the accuracy and efficiency of nonadiabatic dynamics simulation. This sampling algorithm is very useful for chemical reactions with multiple transition states, e.g., two transition states in the thermolysis of four-membered heterocyclic peroxides. The dependence of the chemiexcitation yields, dissociation times, and other quantities on the total initial kinetic energy and extra positive initial velocity has been investigated. By taking different CASPT2 corrections and additional positive initial velocities into account, we found that the resulting triplet quantum yield matches the experimental result perfectly. In most ensembles, the secondary primary intersystem crossing, i.e., the first singlet excited state to the first triplet state, is a major direct production channel for the first triplet state product trajectories. 

The energy gaps, spin-orbit coupling (SOC) and admixture coefficients over a series of the configurations are evaluated by the SA-CASSCF/6-31G, SA-CASSCF/6-31G*, SA-CASSCF/ANO-RCC-VDZP and MS-CASPT2/ANO-RCC-VDZP to reveal the extent of the inaccuracy of the SA-CASSCF. By comparing the mean absolute errors for the energy gaps and the admixture coefficient magnitudes (ACMs) measured between the SA-CASSCF/6-31G, SA-CASSCF/6-31G* or SA-CASSCF/ANO-RCC-VDZP and the MS-CASPT2/ANO-RCC-VDZP, the SA-CASSCF/6-31G is selected as the electronic structure method in the nonadiabatic molecular dynamics simulation. The major components of the ACMs of the SA-CASSCF/6-31G and MS-CASPT2/ANO-RCC-VDZP are identified and compared, we find that the ACMs are underestimated by the SA-CASSCF/6-31G, which is verified by the reasonable triplet quantum yield simulated by the trajectory surface hopping and the calibrated SA-CASSCF/6-31G. The magnitude of the singlet-triplet mixing positively correlates to the hopping probability between the mixed singlet and triplet states, which is confirmed by the computed S-T transition probability. 

The thermolysis of trans-3,4-dimethyl-1,2-dioxetane is studied by trajectory surface hopping. The significant difference between long and short dissociation times is rationalized by frustrated dissociations and the time spent in triplet states. If the C−C bond breaks through an excited state channel, then the trajectory passes over a ridge of the potential energy surface of that state. The calculated triplet quantum yields match the experimental results. The dissociation half-times and quantum yields follow the same ascending order as per the product states, justifying the conjecture that the longer dissociation time leads to a higher quantum yield, proposed in the context of the methylation effect. The populations of the molecular Coulomb Hamiltonian and diagonal states reach equilibrium, but the triplet populations with different Sz components fluctuate indefinitely. Certain initial velocities, leading the trajectories to given product states, can be identified as the most characteristic features for sorting trajectories according to their product states. 

Deep learning (DL) models have demonstrated state-of-the-art performance in the classification of diagnostic imaging in oncology. However, DL models for medical images can be compromised by adversarial images, where pixel values of input images are manipulated to deceive the DL model. To address this limitation, our study investigates the detectability of adversarial images in oncology using multiple detection schemes. Experiments were conducted on thoracic computed tomography (CT) scans, mammography, and brain magnetic resonance imaging (MRI). For each dataset we trained a convolutional neural network to classify the presence or absence of malignancy. We trained five DL and machine learning (ML)-based detection models and tested their performance in detecting adversarial images. Adversarial images generated using projected gradient descent (PGD) with a perturbation size of 0.004 were detected by the ResNet detection model with an accuracy of 100% for CT, 100% for mammogram, and 90.0% for MRI. Overall, adversarial images were detected with high accuracy in settings where adversarial perturbation was above set thresholds. Adversarial detection should be considered alongside adversarial training as a defense technique to protect DL models for cancer imaging classification from the threat of adversarial images.