This is related to the different collection methods of the 2 tools.
PA's way of collection of Sessions rate is based on "Active Time" of each session as collected by PA collector. Then the sessions rate (of Active Time) is displayed as Sessions/ Sec.
In PI, the sessions rate is based on "Elapsed Time" of each session; there is a difference between "Active Time" and "Elapsed Time" metrics. For example, a given session was active for short period of times in each minute within the hour presented. It appears the session was active the whole hour but really it did not. PI attempts to get the time the session was active in the given each minute by aggregating the "Elapsed Time" metric specific to that session. "Elapsed Time" at session level does not count how much time the session was active but how much time the session spent in running statements, basically aggregation of active time of each SQL statement run by that session in the given unit time frame.
PA uses memory based collection. It can sample 20 or more times per second on platform.
PI collects metrics of top 500 SQL statements each second, by querying data dictionary view. This implies that it cannot collect all SQL statements on a busy server, run in the given 1 second. Not all sessions are collected or maintained. PI is a performance measuring tool and as such the statements/sessions that completed their activity under a second are not performance bottlenecks. The goal is to collect metrics that are relevant to long running or resource consuming SQL statements.
These differences between the two products at seconds level may not appear significant but do get magnified when a whole hour is displayed in the UI. The better approach is to focus on statements/sessions that are consuming resources and compare the PI metrics to those displayed by data dictionary views. This also implies the data is more accurate when seen at each minute resolution but as time passes and data gets aggregated, the granularity of data becomes coarse and accuracy gets hit due to compression and discarding of data with aggregation moving up to every 6 hours - a day - 1 week and so on ..