Modeling Environmental Tobacco Smoke (ETS) Infiltration in Low-Income Multifamily Housing before and after Building Energy Retrofits

Secondhand exposure to environmental tobacco smoke (ETS) in multifamily housing remains a health concern despite strong recommendations to implement non-smoking policies. Multiple studies have documented exposure to ETS in non-smoking units located in buildings with smoking units. However, characterizing the magnitude of ETS infiltration or measuring the impact of building interventions or resident behavior on ETS is challenging due to the complexities of multifamily buildings, which include variable resident behaviors and complex airflows between numerous shared compartments (e.g., adjacent apartments, common hallways, elevators, heating, ventilating and air conditioning (HVAC) systems, stack effect). In this study, building simulation models were used to characterize changes in ETS infiltration in a low income, multifamily apartment building in Boston which underwent extensive building renovations targeting energy savings. Results suggest that exterior wall air sealing can lead to increases in ETS infiltration across apartments, while compartmentalization can reduce infiltration. The magnitude and direction of ETS infiltration depends on apartment characteristics, including construction (i.e., level and number of exterior walls), resident behavior (e.g., window opening, operation of localized exhaust fans), and seasonality. Although overall ETS concentrations and infiltration were reduced post energy-related building retrofits, these trends were not generalizable to all building units. Whole building smoke-free policies are the best approach to eliminate exposure to ETS in multifamily housing.

reproductive outcomes, such as low fertility, low birth weight and premature birth, sudden infant death syndrome, bone density loss, and tooth decay [3].
The home environment has been identified as the main source of ETS exposure [4,5]. In multifamily housing, ETS sources can include the resident smoker(s) as well as smokers living in other units within the building. In the US, it is estimated that of the 62.7 million people living in multifamily housing with smoke-free home rules, 44%-46% have experienced exposure to ETS infiltration [6]. A study in a Boston multifamily housing complex detected nicotine in up to 89% of nonsmoker units [7]. Other studies have documented reports of smoke intrusion or tobacco smoke smells in non-smoking apartments [7][8][9], and real-time instances of ETS transfer have been documented in smoke-free apartments from adjacent smoker occupied households [10]. King et al. found evidence of ETS PM 2.5 movement from smoker to non-smoker units and shared hallway areas, with PM 2.5 concentrations in smoker and non-smoker units ranging from 4.2 to 229.6 µg/m 3 and from 1.6 to 15.7 µg/m 3 , respectively [8]. Identifying interventions to reduce exposure to PM 2.5 from ETS in multifamily housing is critical to reduce the burden of smoking-related morbidity and mortality. However, it is challenging to characterize both the magnitude of ETS infiltration as well as the impact of interventions because of the complexity of airflow paths in multifamily buildings. ETS can move between apartment units through natural ventilation pathways (e.g., windows), mechanical ventilation systems, ductwork, shared walls, doors, and building cracks [11,12]. Building simulation models can be useful tools to study this complex behavior, and have increasingly been used to tease apart the effects of building interventions on indoor air quality [13][14][15].
In this study, we used a building simulation model to estimate the impact of building retrofits on the migration and infiltration of ETS across apartments in a 1960s-era multifamily housing complex located in Boston, MA, USA. In 2010, the housing complex underwent extensive energy-related building renovations, providing an opportunity to study changes in ETS migration in a real setting. Using building simulation models, infiltration of ETS PM 2.5 was quantified across different units in a multifamily building and determined how ETS concentrations indoors were influenced by building interventions (e.g., weatherization, energy retrofits), human behavior (e.g., window opening, operation of localized exhaust fans), and seasonality.

Materials and Methods
Site. Castle Square Apartments (CSA) is a 500-unit 100% low-income apartment complex located in Boston's South End, which includes nineteen 4-story low rise buildings with 308 units (townhouses). The complex was constructed in the 1960s, when energy conservation was not considered in construction. In 2010, the property underwent extensive energy and green design improvements as part of a comprehensive 4% Low Income Housing Tax Exempt Bond rehabilitation, resulting in energy consumption reductions of 48% in the townhouse buildings. Building retrofits included air sealing, high efficiency mechanical improvements, window and door replacements, lighting and appliance replacements, and installation of bathroom and kitchen exhaust fans/systems. See Table A1 for renovation details.
Building simulation. CONTAM 3.1.0.3 (NIST, Gaithersburg, MD, USA) was used to simulate a typical CSA townhouse. CONTAM is a well validated multi-zone airflow and contaminant transport analysis model [16] that provides a framework for simulating building model construction and estimates airflows and contaminant concentrations of specified pollutants. The townhouse model template was developed based on architectural drawings, mechanical blueprints and construction specifications of a CSA townhouse building pre and post deep energy building retrofit. The units are organized as stacked duplexes, with the lower units on floors 1 and 2, and the upper units on floors 3 and 4. The most typical building configuration has 8 units: 4 units on floors 1 and 2, and 4 units on floors 3 and 4 ( Figure 1). Each unit has a living room, kitchen, and mechanical room located on the lower level, and a bathroom and bedrooms (2 or 3) located on the upper level. Building dimensions, building elements (windows, doors), and mechanical ventilation systems were obtained from townhouse plans and blueprints, and used to assign room and building element dimensions. Leakage values were assigned to ceilings and walls (exterior, interior, and firewalls), wall joints (wall-ceiling, wall-floor), and window and door frames, kitchen dampers, and window air conditioning units based on effective leakage areas (ELAs) published by ASHRAE [17] or previous modeling studies. The ASHRAE tables publish minimum, average and maximum ELAs for building elements, and values for pre and post energy retrofit conditions were selected from this range. For elements not available from ASHRAE, leakage coefficients were selected from the building element with the closest matching description. No measurements were performed to verify the leakage values at CSA townhouses, thus the modeling results are representative of typical housing with CSA townhouse layout characteristics, but not the actual buildings. No information was available to validate the leakage values in the CSA townhouses, thus these leakage values are representative only of average building components. Table A2 shows the leakage values assigned per building element, by item, linear foot, or area. The building is heated via a forced air mechanical ventilation system which operated only during the heating season.  Table A2 shows the leakage values assigned per building element, by item, linear foot, or area. The building is heated via a forced air mechanical ventilation system which operated only during the heating season. Appendix Table A3 details the airflow rates and filter characteristics of the ventilation system. The stairwells were open to the outdoors and thus did not have a heating or ventilation system.  Indoor temperature profiles were assigned based on field measurements collected in a study conducted at 15 apartments in CSA (data unpublished). Hourly indoor temperatures were averaged across apartments to construct a 24-h indoor temperature schedule. On average, 24-h average indoor temperatures were 27˝C and 21.9˝C in the summer and winter, respectively. For outdoor meteorology, hourly values of solar radiation and weather conditions (e.g., temperature, relative humidity and direction) for a typical year were downloaded from the TMY2 dataset (Boston station 14739, 1961-1990: Typical Meteorological Year 2, National Solar Radiation Data Base, National Renewable Energy Laboratory). One January week and one July week were selected to represent typical winter and summer weather in the Northeast, respectively. In the winter, average outdoor temperature waś 0.2˝C (range´9.4 to 13.9˝C), average relative humidity was 55.6% (range 25% to 96%), and average wind speed was 7.2 m/s (range 1.5 to 17.5 m/s). In the summer, average outdoor temperature was 23.7˝C (range 15.6 to 33.3˝C), average relative humidity was 63.8% (range 30% to 97%), and average wind speed was 5.5 m/s (range 0.5 to 9.8 m/s).
Smoking parameterization. One smoker was simulated in each townhouse unit in the building, and located in the living room. An emission rate of 10 mg/cigarette [18] and a deposition rate of 0.1/h [19] were assumed in the model. The smoker was assumed to be active from 7 a.m. to 7 p.m., and smoked a cigarette every half hour. Only PM 2.5 from smoking was tracked and modeled in this analysis. PM 2.5 from other sources such as cooking and outdoors were also modeled but are not discussed herein, they will be reported in a future publication. Predicted ETS PM 2.5 from each smoker was tracked independently in order to differentiate between ETS PM 2.5 from an in-unit smoker versus ETS PM 2.5 infiltrating from other apartments.
Multifactorial simulation. Twelve multifactorial scenarios were simulated (Table 1) defined by building conditions before and after energy retrofit renovations, season (winter and summer), window opening (closed or open during cooking and shower events), and exhaust fan use (non-existent pre renovation, on/off post renovation). Windows were opened and kitchen exhaust fans were operated during stove operation periods only-10 min for breakfast and lunch, and 40 min for dinner. Bathroom exhaust fans were operated during showering events only. Flow rates for kitchen and bathroom exhaust fans were 160 cfm and 70 cfm, respectively, based on mechanical plan specifications. In all units interior doors between rooms were modeled as open all day, with the exception of the bedroom doors, which were closed during the night. Doors to the outside were opened for 1 min at 7 a.m., 9 a.m., 1 p.m., 5 p.m., and 7 p.m. every day. See Table A3 for details. All simulations were run during a typical winter and summer week (in January and July, respectively). In the winter, heating was regulated via a centralized mechanical ventilation system, which contained a Minimum Efficiency Reporting Value (MERV) filter, based on standards developed by ASHRAE [20,21]. The pre and post retrofit filters were assigned MERV ratings of 4 and 7, which approximately correspond to PM 2.5 removal efficiencies of 10% and 40%, respectively. In the summer the centralized mechanical ventilation system was not operated, but window-mounted air conditioning units regulated indoor temperatures. The AC units are 100% recirculating and did not remove ETS PM 2.5 . Exhaust fans in the kitchen and bathroom were installed and operated post retrofit only, as no exhaust fans existed pre-retrofit. In order to isolate the impact of exhaust fans on infiltration, fans were only operated in units A and E for model simulations.
Statistical Analysis. SimReadW (NIST, Gaithersburg, MD, USA) was used to convert CONTAM output files to text and Microsoft Excel files. SAS software (Version 9.2; SAS Institute Inc., Cary, NC, USA) was then used to calculate 24-h average ETS PM 2.5 concentrations over 7 days each season, by apartment. Room volume was used to weight ETS PM 2.5 concentrations and obtain an apartment average. Total infiltrated ETS PM 2.5 was estimated by summing ETS PM 2.5 from all other apartments measured in a specific unit, assuming there was a smoker in each unit of the building. 24-h average air exchange rates were also estimated for each apartment unit by summarizing the airflow across all paths (i.e., windows, doors, cracks), and dividing by apartment volume.

Results
Across all modeled units, seasons, window opening, and exhaust fan scenarios, 24-h average predicted concentrations of ETS PM 2.5 generated from in-unit smokers present in all units ranged between 16 and 152 µg/m 3 pre retrofit, and between 16 and 185 µg/m 3 post retrofit. Predicted 24-haverage ETS PM 2.5 concentrations due to infiltration from other apartments in the building ranged from 0.2 to 34 µg/m 3 pre retrofit and 0.2 to 8.2 µg/m 3 post retrofit. Note that these predictions assume there is one smoker in each apartment across the building. On average across all apartments and scenarios, 0.7% to 41% of the ETS PM 2.5 concentration in a unit was due to infiltration from other apartments, and over 99% of this infiltration originated from directly adjacent apartments. In the winter, mean daily air exchange rates were 0.99/h (0.62-1.43/h) pre retrofit, and 0.79/h (0.49-1.18/h) post retrofit. In the summer, mean daily air exchange rates across all apartments and scenarios were 0.28/h (range: 0.16-0.49/h) pre retrofit, and 0.24/h (range: 0.13-0.50/h) post retrofit. Figure 2 is a compilation of annotated CONTAM output that illustrates the spatial distribution of predicted ETS PM 2.5 generated by a smoker in apartment A across apartments A, B, E, and F, over time and in relation to occupant activities (e.g., window opening, kitchen and bathroom exhaust fans on). The figure is based on scenario 6 (post retrofit, winter, windows open, and kitchen and bathroom exhaust fan used in units A and E). The y-axis maximum is different for each graph, highlighting the magnitude differences (1-2 orders of magnitude lower) in ETS PM 2.5 infiltration compared to primary generation, as expected. Table 2 summarizes the predicted 24-h average ETS PM 2.5 concentration per apartment, averaged over a week, for ETS PM 2.5 generated by the smoker in each unit as well as infiltrating from apartments located adjacent horizontally (right or left) and vertically (above or below). Results are stratified across all factors modeled. Pre retrofit, average ETS PM 2.5 concentrations ranged between 22 and 33 µg/m 3 in the winter and between 92 and 186 µg/m 3 in the summer. Post retrofit average ETS PM 2.5 concentrations ranged between 20-28 µg/m 3 in the winter and 97-232 µg/m 3 in the summer. Predicted concentrations were significantly lower in the winter compared to the summer, due to regular filtration by the mechanical heating system, fresh air introduced through the mechanical ventilation system, and the higher natural ventilation driven by indoor-outdoor temperature differentials in the winter. In the winter, filtration and increased fresh air supply reduced particle concentrations 40%. When the air handling system was off (and windows closed), particle concentrations were 64% lower in the winter compared to the summer due to the increased natural ventilation.
In the winter, retrofitting resulted in a decrease (~7%) in total predicted ETS PM 2.5 concentrations in the downstairs units (A and B) but an increase (~4%) in the upstairs units (E & F). In the summer, retrofitting resulted in a concentration increase in all apartments (~22% on average) due to the increased air tightness with minimal changes in ventilation (i.e., no change in fresh air supply and no filtration through the mechanical ventilation system). In the winter, vertical infiltration dominated due to the stack effect. Horizontal contributions of ETS PM 2.5 were relatively small for all units, representing approximately 1 µg/m 3 or less as an overall average. In the summer, vertical infiltration was observed in both the upper and lower level units, given daytime/nighttime temperature differentials that can change the direction of the stack effect. ETS PM 2.5 concentrations and vertical infiltration were greater in interior units (B and F) compared to end units (A and E). Across both seasons, Unit B had consistently higher average weekly concentrations of ETS PM 2.5 from the in-unit smoker, compared to the adjacent unit (A) and units located above (E and F), reflecting the influence of fewer outside walls/ceilings in reducing concentrations. Horizontal infiltration of PM 2.5 was also greater in unit B compared to unit A, reflecting the influence of multiple adjacent units compared to other apartments.  In total, focusing just on infiltration (representing ETS PM2.5 exposure for a non-smoking household vertically and horizontally adjacent to units with smokers), retrofits led to a decrease in predicted ETS PM2.5 concentrations in the winter in units A, E, and F, but a small increase in unit B. As the ETS PM2.5 concentrations were greatest in the upper units due to the stack effect, the benefits in units E and F were far greater than the disbenefit in unit B. In contrast, in the summer, retrofits led to an increase in predicted ETS PM2.5 concentrations in all units, which were generally larger in magnitude than the concentration decreases estimated in the winter.
For predicted ETS PM2.5 infiltration from other units, examining the relative contributions by unit provides insight about key infiltration pathways before and after retrofits (Figure 3). In the winter (Figure 3a), retrofits had a limited influence on all apartments except unit B, where the percentage infiltration from unit A increased and the percentage infiltration from unit C decreased. This was likely due to the air sealing that occurred between units B and C which did not occur between units A and B. Comparing pre and post retrofit infiltration in the summer (Figure 3b), changes in the origin of vertical and horizontal infiltration were observed, although with a strong dependence on window opening and exhaust fan use. When the exhaust fans were operated in units A and E, horizontal infiltration increased in both apartments, likely due to the change in pressure between apartments. In total, focusing just on infiltration (representing ETS PM 2.5 exposure for a non-smoking household vertically and horizontally adjacent to units with smokers), retrofits led to a decrease in predicted ETS PM 2.5 concentrations in the winter in units A, E, and F, but a small increase in unit B. As the ETS PM 2.5 concentrations were greatest in the upper units due to the stack effect, the benefits in units E and F were far greater than the disbenefit in unit B. In contrast, in the summer, retrofits led to an increase in predicted ETS PM 2.5 concentrations in all units, which were generally larger in magnitude than the concentration decreases estimated in the winter. For predicted ETS PM 2.5 infiltration from other units, examining the relative contributions by unit provides insight about key infiltration pathways before and after retrofits (Figure 3). In the winter (Figure 3a), retrofits had a limited influence on all apartments except unit B, where the percentage infiltration from unit A increased and the percentage infiltration from unit C decreased. This was likely due to the air sealing that occurred between units B and C which did not occur between units A and B. Comparing pre and post retrofit infiltration in the summer (Figure 3b), changes in the origin of vertical and horizontal infiltration were observed, although with a strong dependence on window opening and exhaust fan use. When the exhaust fans were operated in units A and E, horizontal infiltration increased in both apartments, likely due to the change in pressure between apartments.  Infiltration patterns are strongly influenced by daily ambient temperature, especially during the summer when ambient temperatures may fluctuate above and below indoor temperatures (Figure 4). On summer days when the average daily temperature outdoors was lower than indoors (days 1, 4, 5, 6), the infiltration magnitude was low (<10 µg/m 3 ), with vertical infiltration contributing most to ETS PM2.5 concentrations in the upper units and horizontal infiltration contributing most in the lower Infiltration patterns are strongly influenced by daily ambient temperature, especially during the summer when ambient temperatures may fluctuate above and below indoor temperatures (Figure 4). On summer days when the average daily temperature outdoors was lower than indoors (days 1, 4, 5, 6), the infiltration magnitude was low (<10 µg/m 3 ), with vertical infiltration contributing most to ETS PM 2.5 concentrations in the upper units and horizontal infiltration contributing most in the lower units. On days 2, 3, and 7, a reverse stack effect occurred, and vertical infiltration increased by an order of magnitude in the lower apartments. In the winter, patterns were similar to the lower-temperature summer days, with vertical infiltration dominating, and outdoor air entering the downstairs apartment and diluting ETS PM 2.5 . In addition, in the upper units, increased outdoor/ambient temperature differentials led to increased vertical infiltration.

Discussion
Across all units, seasons, window opening, and exhaust fan scenarios, the 24-hour average predicted ETS PM2.5 concentration inside a smoking apartment across all apartments and scenarios was 73 µg/m 3 , ranging between 16 and 185 µg/m 3 . This estimate was for a smoker who smoked 24 cigarettes per day, with smokers in all adjacent units. These numbers are comparable to what has been modeled previously. In modeling studies, Chahine et al. estimated a mean concentration of 16.3 µg/m 3 among households with smokers (average number of cigarettes smoked = 14.9) and a 99th percentile concentration of 67 µg/m 3 [22], considering a range of home sizes and types, while Myatt et al.

Discussion
Across all units, seasons, window opening, and exhaust fan scenarios, the 24-h average predicted ETS PM 2.5 concentration inside a smoking apartment across all apartments and scenarios was 73 µg/m 3 , ranging between 16 and 185 µg/m 3 . This estimate was for a smoker who smoked 24 cigarettes per day, with smokers in all adjacent units. These numbers are comparable to what has been modeled previously. In modeling studies, Chahine et al. estimated a mean concentration of 16.3 µg/m 3 among households with smokers (average number of cigarettes smoked = 14.9) and a 99th percentile concentration of 67 µg/m 3 [22], considering a range of home sizes and types, while Myatt et al. reported a mean concentration of 18 µg/m 3 for households smoking eight cigarettes/day [23]. Comparing our predicted concentrations to those measured in field studies, a study of New York single family homes and efficiency apartments with indoor smokers, average ETS PM 2.5 concentrations were 84 µg/m 3 (range: 23-285 µg/m 3 ) [24]. Dockery et al. found that one smoker adds 20 µg/m 3 to PM 2.5 personal exposure [25], and Ozkaynak found 30 µg/m 3 higher concentrations in homes with smokers compared with non-smokers [26]. The lower concentrations reported in these last two studies compared to our models are likely due to differences in home volumes (larger), and smoking and ventilation rates.
Predicted mean 24 hour average ETS PM 2.5 concentrations infiltrating from other apartments in the building ranged from 0.2 to 34 µg/m 3 across all scenarios. This infiltration concentration included ETS PM 2.5 from all apartments, as a smoker was modeled in each unit in the building. However, over 99% of ETS PM 2.5 originated from the apartments directly adjacent. Few published field studies have directly quantified the transport of ETS-related fine particles between smoking and non-smoking units. King et al. showed that median PM 2.5 concentrations differed between non-smoking units, hallways and smoking units (8.3 vs. 16.6 vs. 20.2 µg/m 3 ) [8]. In their study, PM 2.5 concentrations in non-smoking units were similar to outdoor concentrations. Russo et al. enrolled smoking and non-smoking multifamily households in a study that collected real-time PM 2.5 measurements (one-minute intervals), as well as diaries of smoking activity in the smoking households [10]. While this study was not designed to specifically estimate PM 2.5 transfer, rough estimates can be made by comparing distributions of these data. For example, the 75th and 90th percentiles of PM 2.5 concentrations in non-smoking units adjacent to smoking units were elevated during hours when smoking was reported, when compared to other hours (24.2 vs. 8.8 µg/m 3 and 64.8 vs. 27.1 µg/m 3 ). From these data, it could be inferred that these transfers are observable, and can easily exceed 30 µg/m 3 over short time periods. Based on these studies, our modeled data are consistent with field-based measurements.
Air exchange rates in our study were much lower in the summer compared to the winter due to two mechanisms-the operation of the mechanical ventilation system with a fraction of fresh air supplied (fresh air was 5% of the total flow rate of 800 CFM), and the greater indoor-outdoor temperature differential which induces higher outdoor air infiltration. In the summer, the apartment units had a 100% recirculating air conditioning unit, which modified the temperature but didn't increase fresh air. Seasonal concentration patterns were also related to other model assumptions. For example, the window open schedule was based on cooking and showering activities, which were presumed to be equal in the summer and the winter. As has been published, window opening behavior is complex and can depend on many factors, including physical environmental, contextual, psychological, and physiological [27]. In reality residents open their windows more frequently in the summer, which results in higher summer air exchange rates [28]. In addition, there may be greater indoor smoking activity in the winter versus the summer. Because the same window opening schedule was maintained in the winter and summer for comparison purposes, winter air exchange rate in our study was higher as it was driven by greater temperature differentials. In Boston public housing units, higher PM 2.5 concentrations in the winter were attributed to reduced winter air exchange rates and increased indoor smoking activity [29]. Our simulations were not meant to accurately characterize behaviors in each season, but to illustrate the joint influence of multiple factors (i.e., indoor-outdoor temperature, pressurization, window opening) in a controlled experimental setting. Future work could include studying the impact of different window-opening schedules and capturing occupant behavior patterns that vary seasonally.
One of the most important insights from our models is the strong dependence of ETS PM 2.5 infiltration on apartment location within the building (i.e., upper/lower level, interior/exterior status) and season. In the winter, ETS PM 2.5 infiltration in upper level apartments (units E and F) was dominated by contributions from lower level apartments (units A and B). In contrast, lower level apartments (units A and B) had higher concentrations of ETS PM 2.5 infiltrating from neighboring adjacent units. These findings are supported by Arku et al., who also found higher PM 2.5 concentrations on upper level areas during the winter [29]. In contrast, in the summer, ETS PM 2.5 for units A and B predominantly originates from the units above (E and F, respectively), and a reverse stack effect leads to decreases in vertical infiltration to units E and F from units A and B, respectively. Coupled with our analyses of patterns in relation to daily average temperatures (Figure 4), this suggests that the stack effect (or reverse stack effect) plays a strong role in driving ETS PM 2.5 movement in multifamily housing, even after air sealing retrofits. Results also reflect the importance of tracking weather conditions when conducting ETS PM 2.5 field studies, as infiltration varies greatly depending on temperature differentials.
In the winter months, infiltration was reduced in units A, E, and F (with a greater reduction when kitchen and bathroom exhaust fans were operated in units A and E), with a modest increase in infiltration in unit B. Units A, E, and F each have at least three exterior surfaces that are exposed to the outdoors. In contrast, unit B has only two exterior walls, with units adjacent on two sides and above. The smaller number of exterior facing surfaces likely insulates unit B and decreases pressurization that occurs with differences between indoor-outdoor temperatures. Reduced exterior surfaces, reduce infiltration of cool air from the outside to the inside. While ETS PM 2.5 is pressurized from unit A into unit B, air sealing reduces upwards movement of ETS PM 2.5 that infiltrates from adjacent horizontal units, resulting in the increases of ETS PM 2.5 observed. Although the goal of the energy retrofit was not unit compartmentalization, our results suggest that the minimal compartmentalization that was implemented (i.e., between-unit floor and ceiling caulking) resulted in modest reductions in ETS infiltration, although apartment location and season strongly influenced these effects.
Although previous studies have found moderate reductions (~30%) in inter-unit transfer of ETS from smoker to non-smoker units after increasing ventilation rates, balancing the mechanical ventilation system, and air sealing [30], our study results showed more limited changes with increases in the summer. The CSA retrofit included some attention to improved compartmentalization between apartments, primarily in the adjoining kitchen walls (between B and C, and F and G), but given the limited scope of this work in the retrofit we did not expect, and did not observe, much change in infiltration between A and B, and E and F.
There are some limitations to our approach. The CONTAM simulation was constrained to model a specific indoor temperature profile rather than dynamically interacting with the outdoor meteorology. CONTAM does not have a direct way of dynamically modeling indoor temperature, although it can link to a separate program with that capability. Because of the complexity of linking the two programs, the temperature profiles were applied based on field data collected from a previous field study at CSA. In addition, it can be difficult to model every exposure pathway for ETS transfer within occupied multifamily buildings. Door opening, thermal gradients, mechanical equipment operation, window opening and the location and timing of smoking activity are all relevant to these exposures. We have attempted to capture most of these drivers of exposure variability, and have captured random human activities such as door and window opening with a fixed schedule, which may not be representative of more random human behavior, which could contribute significantly to dilution and ETS transfer between units. As mentioned above, window opening was standardized to be the same in the summer and winter. Studies have shown that probabilistic modeling is a better way to model window opening behavior compared to standardized occupant behavior profiles [31]. However, since our objective was to compare infiltration pre and post retrofits, keeping a standardized schedule allowed us to make direct comparisons. Given our focus on multifamily housing, the small apartment volumes typical of subsidized housing, and the specificity of the townhouse building design based on details from CSA, the quantitative findings regarding ETS infiltration may not generalize to other types of homes. However, the general trends and drivers of infiltration are generalizable-for example the relative importance of stack effect versus horizontal infiltration. Another limitation is that the air leakage values are all uniform (e.g., only based on surface area) and are not based on measured values. Future studies should include detailed air leakage measurements to verify interior, exterior, and total air leakages pre-and post-retrofits. Although building a specific building template is time consuming, the analytical approach could be readily extended to other types of homes. Our results should also be interpreted with caution with respect to CSA itself, given that we did not try to capture the prevalence of different behaviors at CSA, but instead used our multi-factorial design to represent different types of units. Whether the deep energy retrofit led to net benefits or disbenefits depends greatly on the percent of smokers at CSA, where they are located in the building, and their window opening and other behaviors. That said, our model provides an indication of the types of units and behaviors where increases and decreases in ETS PM 2.5 would be observed subsequent to an energy retrofit, and realistic occupant data can easily be integrated with our modeling platform and simulation outputs.
This study illustrates the strength of building simulation tools to quantify indoor pollutant migration and track pollutant source contributions in complex buildings such as multifamily housing. Although retrofit and building modifications-including ventilation upgrades and air sealing-are common strategies to improve indoor environmental conditions, these strategies have limited usefulness in preventing exposure if smoke-free units are located in the same building as smoking units. In geographical areas with large seasonal temperature fluctuations, policies that sequester smoker units on a specific level may find increased or decreased ETS PM 2.5 movement during specific seasons, due to the stack or reverse-stack effect.

Conclusions
Model results from a multifamily apartment complex in Boston showed that although ETS PM 2.5 infiltration from neighboring units is reduced after air sealing and ventilation improvements, pollutants generated from any given apartment can be detected in other units. Results provide evidence of the relative impact of housing characteristics (i.e., apartment location), energy-related building retrofits, resident behavior (i.e., window opening, exhaust fan use), and seasonality on ETS PM 2.5 concentrations and infiltration patterns. Results support the body of literature suggesting whole building smoke-free policies are the only intervention that will prevent exposure to ETS for all residents in multifamily housing. Our study demonstrates the usefulness of building simulation models in quantifying the impact of building retrofits on pollutant infiltration in an existing apartment complex.    1 10 min at 6:00, 6:30, and 20:00 (weekdays) 10 min at 9:00, 9:30, and 20:00 (weekends) (same as showering schedule) exhaust fan on 2 flow rate: 70 cfm Air Handling Unit supply and return fans on 3,4 flow rate: 800 cfm fresh air: 40 cfm (5%) recirculating air: 760 cfm 15 min per hour (all day) Open Doors front and back 1 1 min at 7:00, 9:00, 13:00, 17:00, 19:00 mechanical room 1 1 min at 9:00 1 Air flow determined by two-way flow model in CONTAM; 2 Exhaust fans only active in post-retrofit scenarios; 3 Air handing unit supply and return fans only on in heating season (winter and spring); 4 Air handling unit uses MERV-4 and MERV-7 particulate filters in pre-and post-retrofit scenarios, respectively.