Here are sixteen lines from the Aeneid:
I.1 Arma virumque cano Troiæ qui primus ab oris
I.119 Arma virûm tabulæque et Troïa gaza per undas
II.668 Arma viri ferte arma; vocat lux ultima victos
IV.495 Erige, et arma viri thalamo quæ fixa reliquit
VI.233 Imponit suaque arma viro remumque tubamque
VI.489–90 Phalanges / ut videre virum fulgentiaque arma per umbras
VI.651 Arma procul currusque virûm miratur inanis
VI.814–5 Tullus in arma viros et iam desueta triumphis / agmina
IX.56–7 Non obvia ferre / arma viros sed castra fovere
IX.462–3 Turnus in arma viros armis circumdatus ipse /suscitat
IX.620 Sinite arma viris et cedite ferro
IX.777 Semper equos atque arma virûm pugnasque canebat
X.423 Hæc arma exuviasque viri tua quercus habebit
XI.696–8 Tum validam perque arma viro perque ossa securim … congeminat
XI.746–7 Volat igneus æquore Tarchon / arma virumque ferens
XII.425–6 ‘Arma citi properate viro, quid statis?’ Iapyx / conclamat
Arma virum is a catchphrase in the Aeneid; no-one could deny that. It occurs very frequently, and to the complete exclusion of synonymous equivalents. Nevertheless it is not a formula, like τὸν δ᾽ ἀπαμειβόμενος προσέφη or pius Æneas. There is no fixed grammatical relationship between arma and virum. Sometimes, for instance, it’s arma virum, ‘arms [and a] man’, and sometimes it’s arma virûm ‘the men’s arms’. And sometimes arma and virum seem to be completely unrelated, as at VI.651, which I think is to be translated ‘he wonders at the far-off arms and the chariots empty of men’. These words are not joined by any consistent logical or grammatical relationship, but by the mere fact of their frequently appearing in each other’s company.
To this we can contrast the Catullan æquora–vectum formula, which appears twelve times in Vergil. (It was described by Robert Schmiel, ‘A Vergilian Formula’, Vergilius XXV [1979]: pp. 37–40). Here there is a consistent grammatical relationship between the two words. One is always carried across the seas, whether one is trans æquora vectus, per æquora vectus, or simply æquora vectus. And because the basis of this formula is grammatical rather than lexical, the word æquora can even be swapped out for metrical equivalents like maria or tot vada without making the formula any less recognizable. The collocation arma virum is not like this: it depends for its existence on the very words it is made up of, and not on their syntactical relationship to each other, or the sense which they jointly express.
The equivalent to arma virum in medieval and modern poetry is the epic rhyme. Lyric poets like Marot and Housman were usually capable of avoiding stock rhymes, just as ancient lyric poets tended to avoid repeated collocations. In a lyric poem you sometimes even get the sense that some lines were chosen for their prettiness before the sense of the poem itself was determined. (For instance, maybe Leuconoë was only a made a woman in order to allow for the words credula postero). But you can’t achieve such minute elegance in a long and narrative poem, where a certain pre-determined story must be told; and so Chaucer wrote day and lay, honour and conquerour, wyf and lyf etc. whenever it helped him move his tale along. The words wyf and lyf bear no consistent relationship to each other in grammar or sense, but they make a harmonious jangle.
Now, Vergil’s epic diction has certain lyric tendencies that distinguish it from Homer’s. Lest any two lines be exactly alike he used fixed formulæ in much greater moderation. Still, in a long narrative poem there was no reaching the studied uniqueness of Horace’s lyric lines, or even of Horace’s wandering hexameters. In the Aeneid Vergil had a pre-ordained story to tell. So at certain junctures he sacrificed lyric preciousness to the requirements of that story, and resorted to his toolbox of collocations that could be guaranteed to be both metrical and elegant.
