# Stabilization decay

**Stabilization decay** is a gradual decrease in the potential to increase **memory stability** with review as **stability** increases. This means that durable memories cannot be easily made more durable. Stability tends to saturate. Once memories approach the definition of permastore, it is hard to make them more stable.

Stability tends to decay along the formula:

**SInc**=(*SInc*-1)*S_{Max}^{decay}+1

where:

**SInc**:**memory stabilization***decay*: stabilization decay constant associated with item difficulty (usually in the range of -0.2..-0.8)- S: memory stability
*SInc*: maximum possible stability increase for memories of given difficulty (usually in the range of 2..7)_{Max}

High *decay* may paradoxically indicate that a student is capable of quickly transferring memories to the high stability storage.

In SuperMemo for Windows, **stabilization decay** can be inspected with **Tools : Statistics : Analysis : Approximations** where stability is expressed as *repetition category* used in SuperMemo as of Algorithm SM-11 (2002). Due to the deployment of anti-cramming boundary conditions in Algorithm SM-8 (and later), stabilization cannot drop below 1.200.

Algorithm SM-2 did not provide for **stabilization decay** due to the fact that E-factor stood for both stabilization and difficulty, which were both presumed constant, and approximated incrementally.

**Stabilization decay** has been known since SuperMemo 7, and wired in the Algorithm SM-8 in 1995. However, it has always been limited to ensure stability increase of no less than 1.1-1.2 (depending on implementation). Algorithm SM-17 is more permissive as it permits stabilization to drop below 1.0. However, for larger samples, this occurs mostly for two reasons:

- spacing effect for high retrievability
- artifact of long interval manipulation

The artifact drop may occur when manually shortening very large intervals. When items survive very long intervals, that information may contribute to the estimate of stability, esp. in cases of substantial underestimate. The same information becomes unavailable when the user shortens the interval, which is a natural occurrence when *recall anxiety* kicks in (*"will I still remember in 10 years? can I trust SuperMemo?"*). This effect of interval on stability estimate will indirectly affect stabilization, and on rare occasions, may lower it below the natural limit of 1.00.

See also: Stabilization curve

This glossary entry is used to explain SuperMemo, a pioneer of spaced repetition software since 1987

decayFigure:Stabilization decayis the drop inmemory stabilizationwith the increase inmemory stability. The picture taken from SuperMemo shows the impact of stability on stabilization for items of difficulty estimated at 0.37. 25,686 repetition cases have been used to plot the graph. Thein stabilization is -0.529. The maximum possible increase in stability (SIncMax) is 3.102 (at Stability=1). The blue dots show data points (S:SInc), where the bigger the circle the bigger the sample of repetitions. The yellow line is a power fit described by the formula:SInc=(3.102-1)*S^{-0.529}+1. This fit provides a deviation of 0.4235