In a sandhi sequence like devān+tatra, according to P 8.3.7 (naśchavyapraśān), the word final n changes into ru, which changes into visarga and ultimately into s. The vowel before the n becomes optionally nasalized by P 8.3.2 (atrānunāsikaḥ pūrvasya tu vā), and in the alternative that it does not become nasalized, it is followed by an anusvāra by P.8.3.4 (anunāsikāt paro 'nusvāraḥ). The rule P 8.3.36 (vā śari) says that a word-final visarga optionally remains a visarga before ś, ṣ, and s, while the Vārttika (kharpare śari vā lopo vaktavyaḥ) stipulates that if the following ś, ṣ and s are followed by a khaR sound (i.e. kh, ph, ch, ṭh, th, c, ṭ, t, k, p, ś, ṣ, or s, then the visarga is optionally deleted. Your example अमुष्मिं स्तवे, and the observations of Whitney, MacDonell and Coulson seem to relate to the variation noted by these rules of Pāṇini and the Vārttika. There are quite a few options involved, and they reflect the variation in the usage. Best,
Madhav
Madhav M. Deshpande
Professor Emeritus
Sanskrit and Linguistics
University of Michigan
[Residence: Campbell, California]