Mantroddhara: the key printed above the lock

The verse that is a set of coordinates

In Kashmir, around the year 1000 of the Common Era, Abhinavagupta sat down to explain a line and a half of scripture. He produced sixteen explanations. Then he counted them for the reader, so there could be no mistake about how many there were.

The lines come from the Parātriṃśikā, whose title names thirty verses on the Supreme, a short Trika scripture that travels with his long commentary, the Vivaraṇa. The passage reads, in the electronic text of the root verses held at Göttingen, where all three lines are numbered verse nine (the preceding verse has just said that this is so “of all mantras and vidyās”):

iyaṃ yoniḥ samākhyātā sarvatantreṣu sarvadā | caturdaśayutaṃ bhadre tithīśāntasamanvitam | tṛtīyam brahma suśroṇi hṛdayam bhairavātmanaḥ ||

“This is called the womb of them all, always, in all the tantras. Joined with the fourteenth, O gracious one, accompanied by that which ends with the lord of the lunar days, the third brahma, O fair-hipped one, is the Heart of Bhairava.”

Read as devotional address it passes without incident. It is not devotional address. Three of those phrases are not describing anything. They are giving positions. Tṛtīyaṃ brahma, “the third brahma,” points at an item in a list. Caturdaśayutam, “joined with the fourteenth,” gives a count. Tithīśāntasamanvitam, “accompanied by that which ends with the lord of the lunar days,” gives the far end of a series. A reader who knows which list, which count and which series recovers three Sanskrit sounds: SA, the vowel AU, and the breath-mark called visarga, written Ḥ.

Set them in order and you have SAUḤ. The seed-syllable of the goddess Parā, the Supreme, whose name the text carries.

The operation has a name in the tradition’s own vocabulary. Mantroddhāra, from ud-dhṛ, to draw out or lift up: the extraction of a mantra. It is the reason a great many tantric scriptures, read by someone without the tradition, appear to describe cosmology when they are in fact spelling.

The study that opened the subject

The man who made this a field was the French Indologist André Padoux, in a paper published in the Bulletin de l’École française d’Extrême-Orient in 1978, running from page 65 to page 85, titled simply “the selection of mantras.” It was the first of a series, as Padoux himself sets out in a later footnote: the second, in volume 67 for 1980, took up nyāsa, the ritual installation of mantras on the body, and a third followed in volume 76 for 1987 on japa, recitation. Padoux later reworked the first paper into English, and his definition there is the one to hold onto:

“Mantroddhāra is thus the extraction of the mantra out of the totality of sound or, more precisely, out of the totality of the Sanskrit phonemes, the varṇasamāmnāya (the alphabetical form of vāc). It is at the same time the selection and composition of the mantra of a particular deity, or of the mantra to be given to a particular person — the notion of extraction being, however, always essential.”

Two things in that sentence do real work. The first is varṇasamāmnāya, the collected sounds, which Padoux insists “is not a mere collection of letters, but the phonetic form of the absolute, of vāc, the Word.” The alphabet, in this tradition, is not a filing system. It is a map of what exists, and the levels of speech descend from the undivided to the spoken. The second is that final clause. Padoux grants that extraction is also composition, and then immediately subordinates the composing to the extracting. The mantra is being taken out of something that was already whole.

Thirteen years later the Indologist Gudrun Bühnemann published a study in the Bulletin of the School of Oriental and African Studies whose title said the other half of the thing out loud: “Selecting and Perfecting Mantras in Hindu Tantrism.” Selecting is mantroddhāra. Perfecting is something else, and we will come back to it, because it turns out to be where the whole question actually lives. The sentence she sets down near the start is worth having in front of you:

“A mantra, a combination of letters thought to be of divine origin, is transmitted by a preceptor to a disciple in an initiation ritual whereby the disciple is authorized to practise its ritual repetition and strive for its perfection in order to attain supernatural powers.”

Transmitted. Authorized. Hold those two words.

The machinery

The only systematic modern survey of how the coding actually works is recent and comes from an unexpected direction: a 2020 article by A. Kavaleuskaya, then at Hamburg, published in the Russian ethnographic journal Etnografia and freely available. It is a taxonomy, and a useful one, and we should say plainly what it is. Twenty-one pages, by an early-career scholar, in a journal outside the field, which we have not been able to find cited or reviewed anywhere. Its apparatus is loose in places; we have re-checked every reference taken from it and corrected several. Much of what follows in this section, and the dating in the next, rests on it alone. Where it is the only support for a claim, we say so.

The devices fall into families. There are lettered grids, the prastāra or spreading-out and the gahvara, which Padoux glosses as cavern or hiding place, where the sounds are laid into cells and a verse names a cell by number. There is plain ordinal position in the alphabet. There is neighbour-addressing, where a sound is identified as the one before or after another. There is substitution, in which a deity, a sacred seat, a body part or an element stands for a sound. There are numerical codes and codes based on a letter’s written shape. And there is deliberate scrambling, most often simple reversal, called viloma or sometimes krodha, the angry order.

There is also more than one alphabet. Alongside the standard mātṛkā, the mother-series running from A to KṢA that the tradition calls the garland of letters, there is the mālinī, also called the nādiphāntakrama, the sequence beginning with na and ending with pha. These are not two parallel systems. Kavaleuskaya shows the mālinī is itself extracted from a grid laid out in mātṛkā order, so the second alphabet is a product of the first. Its precise order varies between texts, and this article will not print it. Somadeva Vasudeva has catalogued the variants, and a string that differs between witnesses is not a string to reproduce casually.

It was not there at the beginning

The practice is not original to Śaiva scripture, which is worth saying because the opposite is widely assumed.

The Niśvāsa, the oldest surviving Śaiva tantra, whose earliest layers Dominic Goodall, Alexis Sanderson and Harunaga Isaacson place between about 450 and 550 of the Common Era, is not cited for a single mantra extraction in Kavaleuskaya’s survey. It appears there only for points of doctrine about the alphabet, and from a later layer of the text at that. The earliest text she knows to describe an alphabet grid is the Vīṇāśikhatantra, which gives the grid and then uses it, identifying one seed-syllable as the occupant of cell twenty-five. The Vīṇāśikha is a left-current text, outside the Siddhānta mainstream.

Then it thickens. Grid use, on her reading, “increased and became well established by the ninth century,” and multiplied after Abhinavagupta across the later Krama, Kaula and Śrīvidyā material. Look at where her examples come from and the pattern is hard to miss: the Tantrasadbhāva, the Siddhayogeśvarīmata, the Mālinīvijayottara, the Kubjikāmata, the Jayadrathayāmala, the Vāmakeśvarīmata. These are the tantras of the left and the Kaula streams, not the Saiddhāntika base. The inference is ours and we flag it as ours: dense mantra-coding looks like a later development, and largely a non-Saiddhāntika one.

It has a Buddhist twin, and the direction of traffic there is the reverse of what one might guess. Toru Tomabechi published on mantra extraction in the Sarvabuddhasamāyoga in 2007, in a volume for Ernst Steinkellner, so the term and the practice are at home in Buddhist tantra and not a loan into it. More pointed still is a 2020 paper by Akira Fujii in the Journal of Indian and Buddhist Studies on the two Bhūtaḍāmara tantras, one Hindu and one Buddhist. His argument is that the Hindu text takes its mantras over from the Buddhist one, strips the Buddhist vocabulary out of them, and then applies encoding methods that the Buddhist version does not use. The mantras travelled one way. The habit of coding them was added at the other end.

The key is printed just above the lock

Now the part that breaks the usual story.

Go back to the Parātriṃśikā and read what sits immediately before verse nine. Verses five through eight do not conceal anything. They build the grid, in the open, in order.

Verse five: the vowels are the tithis, the lunar days, running out to the dot called bindu. Verse six: the principles from earth up to the person occupy the five consonant classes beginning with KA and ending with MA. Verse seven: after the four dhāraṇās, the retentions of wind, fire, water and Indra, comes the series beginning with ŚA, and before them the brahmapañcaka, the set of five called the brahmas. Verse eight: from A to KṢA, in that order, the emanation is declared.

That is the key. Every coordinate verse nine will use has just been defined. “The third brahma” counts into the ŚA-series from verse seven and lands on SA. “The fourteenth” counts the vowels of verse five and lands on AU. For the third coordinate, Abhinavagupta’s commentary does the arithmetic out loud: tithīnāṃ pañcadaśānām īśo visargaḥ, the lord of the fifteen lunar days is the visarga.

The text publishes its own codebook immediately above the cipher that uses it.

One caution about the numbering, since it carries weight here. The verse divisions of the Parātriṃśikā differ between editions, and the Göttingen corpus holds a second file, the commentary with the verses embedded, and it divides them differently: it closes a verse after “this is called the womb” and carries the coded phrases across what it marks as nine and the opening of ten. Kavaleuskaya, working from a printed edition, cites the same verse seven for the five brahmas that we have cited, so the editions agree about the thing that matters even where they disagree about the numbering. Nothing in the argument turns on the count. What every division agrees on is that the key comes first and the cipher follows it.

And this is not a slip. Kṣemarāja, Abhinavagupta’s pupil, says the order is standard. Padoux quotes him on it: “the secret essence of the mantras is none other than the blissful [Mātṛkā] […]. And this is the reason why in every āgama the presentation of the extraction of mantras is preceded by the spreading out (prastara) of the matrka or of the malini.” Every āgama. First the grid, then the extraction. By design.

The cipher is a bad cipher

Once you start looking, the coding is leaky nearly everywhere.

Some code-words do not hide a sound, they describe how it is written. Tribinduka, “having three dots,” is I. Trikoṇa, “triangle,” is E. Kavaleuskaya traces those shapes back toward Brāhmī. Others describe how a sound is made: dīrghādya, “the long one first,” is Ā; ūrdhvanāda, “the upper resonance,” is Ū. Others simply give a position: kādyam, “beginning with KA,” is Ḥ. And a great many are the sound’s own first syllable wearing a costume. Haṃsa, the goose, is HA. Soma is SA. In her own words, these “do not enhance obfuscation but, on the contrary, they help to identify a hidden letter.”

Sometimes a text drops the pretence inside the same passage and names a phoneme outright: a-kāra, the letter A, or repha, the R-sound, sitting among the coded ones.

Reversal, the commonest scrambling, is trivially undone. Kavaleuskaya says so herself, and finds it genuinely puzzling: reversed mantras are “quite easy to identify and to read in the intelligible forward order,” so, she asks, “why at all one should use such an easy method?” She offers an answer herself, and it is not concealment: the reversal, she suggests, may mark the mantra as dangerous, something not to be approached casually. The Kubjikāmatatantra answers her in a way that makes the puzzle worse. Having given a long coded mantra, it says at 10.33:

pāramparyakramāyātam upadeśasamanvitam | vilomavihitaṃ sarvaṃ khādakāstre ‘py ayaṃ vidhiḥ ||

“Come down through the sequence of the lineage, accompanied by oral instruction. All of it is set down in reverse. This same rule applies to the devouring-weapon mantra too.”

The text tells you it is written backwards. In the text.

Then there are the dictionaries. Syllable-codebooks exist as independent works, and they also exist inside the scriptures: the Jayadrathayāmala carries a chapter called varṇanāma, the names of the letters, which is simply a glossary of the substitutions. A tradition that prints its codebook in its own revelation is not, we would say, running a serious access-control system at the cipher layer.

That “we would say” is doing work, and the reader should see it. Neither of the scholars this article leans on would put it as flatly. Padoux is explicit that a mantra “cannot be freely revealed to just any person” and that it is to be revealed “in a particular, secret (or at least indirect) way,” and he describes the gahvaras as theoretically secret diagrams in an order known only to initiates, caverns in which the mantras are hidden. Kavaleuskaya, writing about a Tantrāloka derivation of this same mantra, says it would be impossible to unpin the encoded mantra in the verse “without the commentary of Jayaratha and without a broad knowledge of all doctrinal links and identifications.” Both of them hold that the coding conceals. What follows is our reading of why concealment is the smaller half of what it does, and it is a reading, not a report.

Sixteen readings of the same line and a half

Which brings us back to a scholar in Kashmir and his sixteen answers.

Abhinavagupta’s commentary on the coded lines gives the phonemic reading, the one that yields SAUḤ, in a single clause, and surrounds it with fifteen others. It never writes the mantra out. It tells you, at one point, that the third brahma is the one lying between ṣa and ha, and that the fourteenth is the one standing between O and AṂ, and leaves you to assemble it. Kavaleuskaya quotes that same clause from page 84 of the printed Sanskrit text of the commentary, where it carries a subject word the electronic text has lost, and glosses the brahmas as the five final letters ŚA ṢA SA HA KṢA. It reads the same words as a statement about will, knowledge and action. It reads tṛtīyam, “the third,” as icchā, will. It reads caturdaśa, “fourteen,” as forty. Elsewhere in the same commentary it runs a numerological pass over the same words: fourteen is forty, doubled it is eighty, the lunar days are fifteen, the lords are the eleven Rudras, the ends are the three times. It reads the line as a description of the void, and as a description of emission, and as several things in between. The edition numbers sixteen of them. The count is not the editor’s. The commentary announces it before it begins, evaṃ ṣoḍaśadhā hṛdayam etat, thus this Heart is sixteenfold, and closes the series by totting it up again: evaṃ ṣoḍaśātmikā bījavyāptir uktā, thus the sixteenfold pervasion of the seed has been stated.

A cipher has one plaintext. That is what makes it a cipher. This passage behaves as though it has many, and the tradition’s greatest reader worked through them one after another and told you how many he had found.

We should be careful here, because it is easy to overstate. Abhinavagupta is not denying that SAUḤ is in the verse. SAUḤ is well attested as the mantra of Parā, derived independently in the Tantrāloka, in the Mālinīvijayottara and in the Vāmakeśvarīmata, and we have seen no source that assigns her a different one. The point is narrower and harder. The coded lines are not a container with one thing inside them. They are a knot that rewards pulling, and the mantra is one of the things that comes out.

What the extraction is actually for

If concealment is not doing the work, what is?

The clearest case is a passage in Abhinavagupta’s own Tantrāloka, at 4.186b to 189a, where he derives the same mantra SAUḤ from doctrine rather than from a grid. The reasoning runs: sat, the existent, which gives SA; existence is never spoken of apart from the rise of will, knowledge and action, icchā-jñāna-kriyā, and that triad gives the vowel AU; and in the Bhairava-nature, consciousness, it is emitted, visṛjyate, which gives the emission-mark, the visarga. The sentence does not stop at the end of 188. On the Göttingen text it runs on into the next line, where the emission itself, the visarga, is drawn in along with the triple nature of the powers by supreme awareness. The derivation and the thing derived are described in one breath.

That is not a mantra being hidden. That is a mantra being argued for. The three sounds are shown to be the only three sounds it could have been, given what the tradition holds about how reality unfolds. Paul Muller-Ortega, in a 2003 study of encoding in the Tantrāloka which we have not been able to read directly and report here at second hand, makes essentially this case: Abhinavagupta’s intention, on his reading, encompasses much more than the secret embedding of a mantra in a text, and these passages work to supply a theoretical justification for the mantra’s particular sound-shape. Note the “much more than.” He is adding a function, and we have not read him at first hand.

Padoux’s Vāc carries a second, different derivation of the same mantra, keyed to the Mālinīvijayottara, in which SA covers the principles from matter up to illusion, AU covers those from pure wisdom to Sadāśiva, and the visarga is Śiva. Same three sounds. Different argument. Both printed.

Put the doctrine back in and the coding stops looking like a lock. If the sounds of Sanskrit are the principles of the cosmos, then specifying a mantra by its cosmological coordinates is not a disguise. It is the most direct available description. Saying “the third brahma” rather than “SA” is closer to what the tradition thinks SA actually is.

Where the gate actually sits

So the information was not the chief barrier. Something else was, and Bühnemann’s title named it: perfecting.

Mantrasaṃskāra is the set of rites held to make a mantra work. The Śāradātilaka describes ten of them. Padoux relays the image the Kulārṇava uses at 15.71 to 72 for the same set of ten: “As a sword rubbed on a whetstone gets sharpened, in the same way the mantras bloom and radiate thanks to the ten saṃskāras.” The Netratantra gives nine, naming them: dīpana, kindling; bodhana, awakening; tāḍana, striking; abhiṣecana, consecrating; vimalīkaraṇa, cleansing; indhananiveśana, fuelling; saṃtarpaṇa, satisfying; guptibhāva, concealing; āpyāyana, nourishing. The Śāradātilaka’s ten are not simply those nine with extras. It keeps eight of them, drops indhananiveśana, and adds two preliminaries at the outset, janana, birth, and jīvana, vivifying. Kṣemarāja quotes a tantra that counts seven. The number is not fixed, and that it is not fixed tells you these are living ritual repertoires rather than a formula.

A mantra without these, in this view, is a mantra that does not do anything. The syllables are correct and inert.

And here the tidy division we have been building develops a crack, which it is only honest to point at. Padoux notes on the same page that the Śāradātilaka’s janana, the first of its ten, “is simply the extraction of the letters,” which makes it a variant of mantroddhāra itself. On that text’s own accounting the extraction is not outside the ritual gate at all. It is the first thing inside it.

Above that sits initiation itself, and the lineage condition the Kubjikāmata stated in the same breath as it announced its own reversal: pāramparyakramāyātam upadeśasamanvitam, come down through the sequence of the lineage, accompanied by oral instruction. The Mālinīvijayottara says the same of the SAUḤ extraction. The Svacchanda puts the point negatively, that the phonemes do not yield liberation absent realised union with Śiva, which is a statement about efficacy and not about secrecy at all. The Kulārṇava, the text Bühnemann works from, makes the guru the whole hinge.

There is a tempting next step here, and we will not take it. One could argue that the coding is an institutional device, that it makes initiation non-substitutable and so protects the standing of the guru. The argument is coherent. Hugh Urban has made versions of it about tantric secrecy more broadly. But nobody we can find has argued it for mantroddhāra specifically, and Kavaleuskaya, who had the evidence in front of her, holds both things at once. Secrecy, she writes, serves “not only the purpose of protecting information from those uninitiated, but also facilitates” what she calls “the authorization and sacralization” of the teachings. She does not drop the protective function. She adds a second one beside it. Our reading is that the cipher marks the mantra as something that must be received rather than taken. That is an inference, and it is ours.

What was lost was the commentary

The most striking evidence that this was never a working security system is that the tradition’s own experts kept failing to decode it.

Kṣemarāja, commenting on the Svacchanda, rejects an earlier scholar’s reading of one code-word as the R-sound and reads it instead as the vowel Ū. On a line in the Jayadrathayāmala, a commentator takes the single word “first” two different ways, as RA and as YA, and Kavaleuskaya adds a third parse incompatible with both. Code-words are not stable across a text, so there is often no clean one-to-one mapping to recover. Sometimes a grid silently shrinks from the full alphabet to seventeen letters to five syllables with no notice given, at which point, in her blunt phrase, “a reader is bound to random guessing.” Witnesses of the same tradition disagree over whether SAḤ is one unit or two.

Jayaratha, Abhinavagupta’s commentator two centuries later, had a procedure for when a code defeated him: search other texts of the same canon, then the sub-canon, then down into the most general Saiddhāntika layer, until the information turns up somewhere. That is a scholar’s workaround, and it tells you the keys were distributed across the corpus rather than held in a vault.

And sometimes it simply ran out. Kavaleuskaya notes cases “when all the literary sources of a died-out tradition fail to assist in retrieving the correct mantric form.” Her verdict on the commentators is the honest one: even authorised exegetes “might not have had access to detailed initial instructions.” What was lost was not a secret. It was the teaching that came with the text, and once that went, the text alone could not always be made to give the mantra up.

When the codebook went to press

In 1913, in London and Calcutta, the first volume of the Tantrik Texts series appeared under the general editorship of John Woodroffe, writing as Arthur Avalon, edited by Tārānātha Vidyāratna. It contained the Tantrābhidhāna, together with the dictionary of seeds and the dictionary of hand-gestures. The syllable codebooks were now a printed book. In 1984, in Varanasi, Ram Kumar Rai published them in English. Modern editors do the same thing: Sanjukta Gupta compiled a dictionary of substitute words into her edition of the Lakṣmītantra in 1972, because an editor needs one.

Woodroffe is usually treated as a dated figure. But in the reworked study Padoux quotes him approvingly on this very subject, that in the mother-letters the mantra lies scattered and the extraction forms it by selection, and adds that although Avalon was no expert in Sanskrit the pandits he worked with were, so that his contribution to the field was important and remains useful. The pandits are the part usually left out. Tārānātha Vidyāratna edited the volume.

So the cipher has been open for over a century. Did anything change?

The one piece of well-documented present-day testimony we have found comes from Mani Rao, who did fieldwork at Devipuram in Andhra Pradesh and published it in Religions in 2019. The syllables of the fifteen-syllable Śrīvidyā mantra, she notes, are publicly known. And yet: “unless the mantra has been given by a guru, its practice is considered ineffective by traditional practitioners.” One of her informants, Gopichand Balla, came to the practice, in her words, by “learning how to draw the yantra from a pdf file” that his teacher had put on the internet, and even so, she observes, it is the guru of a community of practitioners who approves what is done with it.

That is the whole argument in one village. The information went public and the gate did not move, which is what you would expect if the gate was not chiefly made of information.

We should be plain about a limit here. We have found no documented comment, from a practitioner or a scholar, on what printing the codebooks in particular did to mantroddhāra. Rao’s testimony is about a mantra, not about a cipher. The parallel is strong and we are drawing it; the direct evidence does not exist that we can see.

What the coding was

Call it encryption and almost everything about it looks incompetent. The key is in the text. The dictionaries are in the canon and in the bookshops. The reversals announce themselves. The code-words sometimes draw you a picture of the letter they are hiding. The tradition’s best readers argued with each other about what the codes said, and the greatest of them, handed a line and a half of code, found sixteen things in it and stopped only when he had made his point.

Call it a derivation and it comes into focus. The Parātriṃśikā lays out the alphabet as a map of the cosmos and then specifies a mantra by its position on that map. The Tantrāloka takes the same mantra and shows that its three sounds are the three stages by which the real becomes the uttered. Both are arguments that this mantra, and no other, belongs to this goddess. That the argument can be run backwards to recover the syllables is a consequence of its being a good argument.

What the tradition withheld, it withheld elsewhere, and it never pretended otherwise. The Kubjikāmata says it in the same sentence as it admits the reversal: through the lineage, with instruction. A reader holding the manuscript in 1100, or the printed edition in 1913, or the PDF in 2019, could get the syllables. He could not get the authorisation, and the rites that were held to make the syllables live were not written down for him either.

The code was a lock. It was never the main one, and the main one was somewhere else.