The zero-dimensional (0D) multi-reactor Modified Spark-Ignition (SI) Model (MSIM) developed in Part I was used to quantify how equivalence ratio (phi) affects in-cylinder nitrogen chemistry and engine-out nitrogen-based emissions during neat ammonia (NH3) SI operation. Experiments were performed on a heavy-duty compression-ignition internal combustion (IC) engine converted to SI and operated at phi = 0.7-1 and an engine speed of 1,000 rpm, with constant fuel energy per cycle and similar combustion phasing irrespective of phi. After model calibration that matched experimental in-cylinder pressure and combustion phasing, MSIM resolved NH3, NO, NO2, and N2O formation and consumption across main combustion, crevice release, and exhaust blowdown stages. Results show that compared to conventional two-zone SI models developed and validated primarily for hydrocarbon fuels, MSIM captured neat-NH3 chemistry and NO formation pathways correctly. NO formation during neat NH3 SI combustion was governed mainly by HNO and NHx chemistry. In contrast, the Zeldovich subset that is usually used to model NO formation in IC engines did not act as a universally NO-producing mechanism, with the reaction O + N2 <-> NO + N becoming a net NO-consuming route during main combustion. Further, results show that crevice chemistry was key to capturing the correct N2O magnitude, with N2O formation being dominated by NH2-NO2 pathways and its sensitivity to phi controlled mainly by the effective reacting crevice fraction. Finally, the inclusion of an exhaust blowdown reactor activated NH2-driven DeNOx routes to N2, helping to improve final NO predictions and changing the balance between NO2 and N2O in a phi-dependent way. Overall, MSIM was shown as a computationally efficient framework for evaluating neat-NH3 SI operating conditions and delivered pathway-resolved insights into nitrogen chemistry needed for efficient engine design.
Post-training alignment often reduces LLM diversity, leading to a phenomenon known as mode collapse. Unlike prior work that attributes this effect to algorithmic limitations, we identify a fundamental, pervasive data-level driver: typicality bias in preference data, whereby annotators systematically favor familiar text as a result of well-established findings in cognitive psychology. We formalize this bias theoretically, verify it on preference datasets empirically, and show that it plays a central role in mode collapse. Motivated by this analysis, we introduce Verbalized Sampling (VS), a simple, training-free prompting strategy to circumvent mode collapse. VS prompts the model to verbalize a probability distribution over a set of responses (e.g., "Generate 5 jokes about coffee and their corresponding probabilities"), which relieves the pressure to produce a single "typical" answer. Experiments show that VS significantly improves performance across creative writing (poems, stories, jokes), social dialogue simulation, synthetic data generation, and open-ended QA, without sacrificing safety and factual accuracy. For instance, in creative writing, VS increases diversity by 1.6-2.1x compared to direct prompting. We further observe an emergent trend that more capable models benefit more from VS. In sum, our work provides a new data-centric perspective on mode collapse and a practical inference-time remedy that helps unlock pre-trained generative diversity.
Digitization reforms have been hailed as an effective way of strengthening state capacity. However, digitization can also fundamentally reshape the organization of bureaucracies. Using a unique administrative dataset on agricultural taxation and surveys of local bureaucrats from Punjab, Pakistan, we show that digitization reforms can have unintended consequences for state capacity. We exploit the staggered rollout of the digitization of land records in Punjab to show that digitization had a negative effect on tax collection. The fall in taxes was not due to a decrease in the tax base. Instead, digitization affected the bureaucrats' capacity to collect taxes. (JEL D73, H11, H71, O12, O17)
The state of Mars' induced magnetosphere is highly dependent on upstream solar wind conditions. Changes in dynamic pressure alter the locations of magnetospheric boundaries such as the bow shock and induced magnetosphere boundary, while the orientation of the interplanetary magnetic field (IMF) impacts the formation of the bow shock. We present Mars Atmosphere and Volatile EvolutioN observations from a period when the solar wind dynamic pressure was low and the IMF was nearly radial. The quasi-radial IMF appeared to prevent the formation of the usual quasi-perpendicular bow shock, instead producing a spatially extended quasi-parallel shock in the observed region. The ambipolar potential in the magnetosheath/shock region enabled planetary heavy ions to leak through the shock and flow upstream. These ions were first accelerated by the solar wind motional electric field, which remained an important ion energization mechanism even though the solar wind flow and IMF were roughly aligned throughout most of the system. The ionosphere had expanded out to very high altitudes as the result of low dynamic pressure. The ionosphere was also unmagnetized, perhaps because of the formation of an ionopause-like boundary. Finally, we compare two adjacent orbits and suggest that Mars' remanent crustal fields may fundamentally alter bow shock formation during radial IMF conditions. Studying unusual combinations of solar wind conditions like the case presented here is critical for understanding the fundamental physics driving induced magnetospheres.
Doxorubicin (DOX), is an indispensable first-line chemotherapeutic. Despite this first-line indication, clinical use of DOX is limited by severe, off-target, and often irreversible cardiotoxicity. DOX induces cytotoxicity in rapidly dividing cancer cells via inhibition of Topoisomerase IIα. However, the underlying mechanisms by which DOX causes cell death in non-replicative, terminally differentiated cardiomyocytes remain poorly understood. Emerging evidence suggests that mitochondrial uptake of DOX is contributory to cardiotoxicity. Whether mitochondrial stress pathways, including the mitochondrial unfolded protein response (UPRmt), are activated and critical for mediating DOX cardiotoxicity is poorly understood. Moreover, whether phosphorylation of eukaryotic translation initiation factor 2α (eIF2α), a mediator of the Integrated Stress Response, regulates potential UPRmt signaling during DOX treatment is also unknown. Here, using human AC-16 cardiac cells, we examined the role of eIF2α phosphorylation during DOX treatment. Our data suggest that DOX triggers a transient increase in eIF2α phosphorylation, followed by a progressive decline. Further, knockdown of eIF2α decreased key transcriptional regulators of UPRmt signaling such as C/EBP Homologous Protein and ATF5, blunted the induction of UPRmt genes (AFG3L2, CLPP, HSPA9, HSPD1, LONP1, SPG7), and aggravated DOX induced cytotoxicity. Together, these findings identify eIF2α as a critical upstream regulator of UPRmt signaling, and suggest that activation of the UPRmt may confer cardio-protection against DOX-induced mitochondrial stress in human cardiac cells.
