Madhyamaka Wiki

Atiśa–Tsongkhapa Pramāṇa Divide

12 min read

The claim

Tsongkhapa positions himself as the systematiser of the Kadampa transmission stemming from Atiśa. But on the question of pramāṇa (valid cognition, ཚད་མ་), the Atiśa documented by apple-jewels-middle-way-2018 holds a position that Tsongkhapa explicitly rejects: that pramāṇa — direct perception (pratyakṣa) and inference (anumāna) in their formal Diṅnāga–Dharmakīrti senses — is useful only for refuting opponents and cannot serve as a means of realising ultimate reality. Tsongkhapa, by contrast, integrates pramāṇa into the realisation of emptiness: ultimate truth IS an object of knowledge, obtained by rational cognition of suchness (tsongkhapa-illuminating-intent-1418 Ch 11).

The argument has two parts. First: this is not a minor scholastic disagreement but a structural divide on the question of how the Two Truths relate to the means of realisation. Second: Nāgārjuna’s Vigrahavyāvartanī vv. 30–51, primary-grounded via westerhoff-vigrahavyavartani-2010, is the Indian primary-text base of the anti-pramāṇa line — and it sits with Atiśa, not with Tsongkhapa. Tsongkhapa’s integration of pramāṇa into the realisation of emptiness has prestigious Indian-Tibetan defenders (Bhāviveka in qualified form, the Diṅnāga–Dharmakīrti corpus as a whole), but on the specific question of whether pramāṇa can reach the ultimate, the Indian primary text he most needs to engage is a primary text he does not read against himself.

Evidence for

Atiśa’s position (primary-grounded via Apple 2018)

apple-jewels-middle-way-2018 recovers Atiśa’s Satyadvayāvatāra vv. 10–13 from Kadampa manuscripts:

  • Pramāṇa is useful for refuting opponents, not for realising emptiness.
  • Direct perception and inference cannot reach ultimate reality.
  • Conventional realities are “mere appearances” (snang ba tsam) — dependent designations without any real basis (avastuka).
  • Atiśa explicitly classifies the pebbles-and-sticks illusion analogy (Śāntarakṣita/Kamalaśīla) as applicable only to “Mind Only and below” — i.e. as a Yogācāra-Madhyamaka analogy unsuitable for what Atiśa calls Great Madhyamaka.
  • A buddha has no continuum of wisdom (jñāna) — buddhahood is nondual fusion with the dharmadhātu bereft of all mind and mental factors. (This is the strongest form of the position: there is no pramāṇa-bearer at the level of buddhahood.)

The position is recovered from the Kadampa manuscript record before Tsongkhapa’s Geluk reformulation. Atiśa is not engaged via the later Geluk doxography that converts him into a Prāsaṅgika in the Tsongkhapa-codified sense.

Tsongkhapa’s position

tsongkhapa-ocean-of-reasoning-1408 Homage .1.1.2 is the earlier (1407–08) and load-bearing primary statement of Tsongkhapa’s position: conventional things are “merely posited by nominal convention” and “established by authoritative cognition” without contradiction. The “merely” precludes inherent existence; it does not preclude pramāṇa. Important refinement of the divide. Ocean’s reading of VV in Preliminary specifies what Tsongkhapa thinks VV actually shows: that “agent and action of establishing and denying authoritative cognition and object make sense in the system asserting essencelessness,” and that for a school positing essence, such things as authoritative cognition do not make sense at all. On Tsongkhapa’s reading, VV vv. 30–51 refute foundationalist (Nyāya) pramāṇa in order to make room for desubstantialised conventional pramāṇa. The Atiśa–Tsongkhapa divide on the Indian-primary-text reading is therefore not whether VV refutes Naiyāyika pramāṇa (both agree), but whether VV’s positive corollary leaves room for Buddhist conventional pramāṇa (Tsongkhapa: yes — VV is precisely the demonstration that essenceless agents-and-actions-of-knowledge work; Atiśa: no — VV’s anti-foundationalism extends to all pramāṇa-realism).

tsongkhapa-illuminating-intent-1418 eleven years later integrates pramāṇa into the realisation of emptiness:

  • Ultimate truth IS an object of knowledge, obtained by rational cognition of suchness, “not established through its own essence” (Ch 11 on MA 6.23).
  • The two-senses distinction (the rational cognition characterised as “ultimate” does establish phenomena; existence-through-intrinsic-mode-of-being does not) is engineered to allow pramāṇa into the system without reifying its objects.
  • Phenomena are posited “through mere conceptualisation in dependence on their bases of designation” — and conventional phenomena, unlike the rope-snake illusion, do exist and are capable of effective functions, with the test being susceptibility to invalidation by pramāṇa (tsongkhapa-illuminating-intent-1418 Ch 9).
  • Three criteria for conventional existence (Ch 11 on MA 6.24–28) are the load-bearing primary-text basis for the pramāṇa-integrationist side. A fact is conventionally existent iff (a) it is acknowledged within unanalysed conventional cognition (tha snyad pa’i shes pa); (b) it is not invalidated by other conventional valid cognition; (c) it is not invalidated by analysis probing the ultimate nature of reality. The criteria operationalise the pramāṇa-integration: pramāṇa is what determines invalidation in (b) and (c). Tsongkhapa is explicit that “deferring to the world” appeals to the unanalysed perspective shared by reflective philosophers and ordinary persons — not a defeatist appeal to cowherds. The three criteria are also the primary-text base for criticising the Atiśa–Gorampa–Karmapa–Mipham line as failing to specify a positive account of conventional warrant. (The wider claim — that those critics owe a positive account of conventional truth — is itself contested, since on the Karmapa’s three-stage view the “no-analysis” level is a positive account that does not require pramāṇa foundationalism.)

jinpa-tsongkhapa-qualms-1999 reports a striking move: Tsongkhapa argues even the non-affirming negation must presuppose the law of excluded middle to be effective, citing Vigrahavyāvartanī 26b: “If the absence of intrinsic being is reversed, intrinsic being becomes established.” The argument is directed against “no-thesis” proponents who reject all logical argumentation — i.e. against the Atiśa–Sakya–Kagyü–Nyingma reading. Tsongkhapa here uses Nāgārjuna against the no-thesis position.

The Indian primary-text base

Vigrahavyāvartanī vv. 30–51 (primary-grounded via westerhoff-vigrahavyavartani-2010) systematically refutes Nyāya pramāṇa theory:

  • Pramāṇas established by other pramāṇas — generates a regress.
  • Pramāṇas establishing themselves on the analogy of fire illuminating itself (vv. 34–39 dismantle the fire analogy).
  • Pramāṇas established by their objects — pramāṇa and prameya turn out mutually dependent, neither has intrinsic pramāṇa-hood.

The conclusion is that no pramāṇa possesses intrinsic nature. This is the exact textual base for Atiśa’s later position. Westerhoff reads it as a positive anti-foundationalist epistemology rather than a global anti-knowledge claim — which is also the structural shape of Atiśa’s position (pramāṇa as polemical tool, not as means of realisation).

Tibetan convergent witnesses to Atiśa’s side

  • Gorampa at MA 6.27 (gorampa-removal-wrong-views) refuses the Geluk pramāṇa-reading of the “stainless mind.”
  • Mipham in Word of Chandra Supplementary Discussion (mipham-introduction-middle-way-2002): “Valid cognition operating on the relative level inevitably establishes the pot as a truly existing pot.”
  • Ninth Karmapa (karmapa-feast-fortunate-1578): pramāṇa operates at the level of “no analysis” only; the three stages of analysis pass beyond it.
  • Gendün Chöpel (twentieth-century, Geluk-trained, ¶14–¶76 of Klu sgrub dgongs rgyan, primary-grounded via lopez-madmans-middle-way-2006). The most concentrated single Tibetan anti-pramāṇa polemic in the wiki: paragraphs ¶14–¶55 dismantle the pramāṇa-grounding of conventional decisions across multiple registers — universal-consensus arguments (the jaundiced-conch analogy ¶15, the king-and-crazing-rainwater story ¶17), lineage-authority circularity (¶19 “an insect is made the final voucher for them all”), the pramāṇa / prameya mutual-dependence regress (¶21), and the sūtra-level rejection of sensory pramāṇa citing Samādhirāja IX.23 (¶21). The polemic culminates in twenty-one refrain-verses (¶56–¶76) each ending in tha snyad tshad grub ‘jog la blo ma bde (“I am uncomfortable about positing conventional validity”), each identifying a specific structural reason why pramāṇa cannot ground itself. GC’s anti-pramāṇa line goes beyond Atiśa’s: Atiśa rejects pramāṇa as a means of realising the ultimate (it remains polemically useful at the conventional level); GC rejects the coherence of conventional pramāṇa itself for the unenlightened mind. The two positions are continuous in direction but GC is the more radical case. GC further supplies three named Geluk-internal cross-witnesses for the narrower anti-qualifier corollary (¶41): lCang skya Rinpoche, Gung thang bstan pa’i sgron me, and Pan chen Blo bzang chos rgyan — the same cross-witness pattern the Ninth Karmapa documents in karmapa-feast-fortunate-1578 with the Karmapa-tradition figures (Changkya is on both lists). The convergent witness count is now four Tibetan recoveries (Sakya / Nyingma / Kagyü / twentieth-century Geluk-internal) plus the eleventh-century Atiśa root — five voices against Tsongkhapa on this question, not three.

Four independent Tibetan recoveries (Sakya, Nyingma, Kagyü, twentieth-century Geluk-internal) converge with Atiśa against Tsongkhapa. The convergence makes the position something other than a Sakya/Geluk sectarian dispute. The Geluk-internal GC convergence is especially diagnostic: it suggests that even within the Geluk academy, the pramāṇa-integration has been a contested institutional commitment rather than a settled philosophical conclusion.

A modern, lineage-external convergent witness — Garfield’s “Farabi’s fallacy”

Garfield, in garfield-causality-2001, questions the prior axiom on which Tsongkhapa’s integration rests: that Dharmakīrti’s pramāṇavāda and Nāgārjuna’s śūnyavāda must be mutually consistent in the first place. He names the hermeneutic error “Farabi’s fallacy” — assuming two great philosophers must be reconcilable and then welding them — and identifies Tsongkhapa as its paradigm Tibetan instance, who “devoted much of the rest of his philosophical life” to demonstrating the consistency of the two vādas. Garfield admires Tsongkhapa (“the titan of the Tibetan philosophical tradition”) but finds this particular influence “less than salutary,” and cautions against “a doxography that takes as axiomatic the consistency of Dharmakīrti’s pramāṇavāda and Nāgārjuna’s Madhyamaka.” This is hostile-witness-valuable in the same way as Kalupahana 1976 on the sp-hierarchy-is-tibetan-construction page: a contemporary Western philosopher, with no lineage-political stake, reaches the Atiśa-side scepticism about the pramāṇa-integration by a wholly independent route. Two calibrations. (i) Garfield’s target is narrow — the rebirth-bodhicitta argument’s reliance on a substantial causal basis (rGyal tshab on Dharmakīrti) — not a wholesale rejection of valid cognition; do not over-read him into Atiśa’s strong “no pramāṇa reaches the ultimate” thesis. (ii) His point is at the level of the consistency axiom (should the two corpora be harmonised at all?), which is logically prior to, and distinct from, this page’s question (can pramāṇa reach the ultimate?). He widens the divide rather than re-stating it.

Śāntarakṣita / Mipham — a distinct middle position that splits the divide cleanly

The MA commentary (Madhyamakālaṅkāra / shantarakshita-madhyamakalankara) shows that the Mipham cited above on the Atiśa side actually holds a two-sided position that neither pole captures, and that maps the divide onto its natural seam. Śāntarakṣita is the Indian Mādhyamika who most fully integrates Dharmakīrtian pramāṇa at the conventional level — the five propositions of his tradition make the causally-efficient thing “the only authentic object of valid cognition,” install reflexive awareness as “the sine qua non of valid cognition on the conventional level,” and hold the two kinds of valid reasoning (conventional pramāṇa + ultimate analysis) without contradiction (MA General Introduction, raw pp. 102–103). That is the Tsongkhapa-adjacent half. Yet he explicitly bars pramāṇa from the actual ultimate: to think intellectual cognition can reach the ultimate is “as foolish… as thinking that a newborn baby can look directly at the sun” (raw p. 29), since the actual ultimate is “the object of neither thought nor word” (MA v. 71) and only the self-cognising wisdom of equipoise reaches it. That is the Atiśa-adjacent half. So Śāntarakṣita/Mipham resolve the apparent binary into its real seam: robust conventional pramāṇa, no pramāṇa at the ultimate. This is exactly the “softening” the Śūnyatāsaptati v. 11 / Vaidalyaprakaraṇa data points (below) already recommend — the Indian base supports pramāṇa doing real work conventionally (with Tsongkhapa) while not reaching suchness (with Atiśa). The page’s Word-of-Chandra citation of Mipham (“valid cognition on the relative level inevitably establishes the pot as a truly existing pot”) is the same position seen from the ultimate side: conventional pramāṇa is robust precisely because it cannot see through to emptiness — which is why it cannot be the means of realising the ultimate.

Evidence against / objections

  • Tsongkhapa’s two-senses distinction. Tsongkhapa is aware of the objection that he reifies emptiness. The Ch 11 distinction between rational-cognition-as-ultimate and existence-through-intrinsic-mode-of-being is engineered precisely to allow pramāṇa in without producing the bden grub the Geluk med dgag of bden grub is supposed to negate. The objection from Atiśa–Gorampa–Mipham is that this rescue is verbal rather than structural; the Geluk response is that the critics have not understood what Tsongkhapa actually says. The dispute is real and not yet decisively settled.
  • Tsongkhapa’s Vigrahavyāvartanī 26b citation (jinpa-tsongkhapa-qualms-1999). Tsongkhapa reads VV 26b as committing Nāgārjuna to the law of excluded middle, contra the no-thesis reading. If this is right, then Nāgārjuna is a logical-foundationalist of a sort, and the vv. 30–51 critique of pramāṇa is a critique of bad pramāṇa (Naiyāyika foundationalism) rather than of pramāṇa as such. This is a coherent reading; whether it survives a verse-by-verse engagement with vv. 30–51 is a further question on which Tsongkhapa’s commentary is less explicit.
  • Bhāviveka’s qualified svatantra use. ames-prajnapradipa-ch2-1995 establishes that Bhāviveka’s actual practice is a paramārthatas-qualified autonomous syllogism, not unqualified pramāṇa-foundationalism. This is closer to Tsongkhapa than to Atiśa. The Tibetan Svātantrika doxography reads Bhāviveka as the authoritative Indian precedent for a Madhyamaka use of inference. Atiśa’s anti-pramāṇa line tracks Candrakīrti against Bhāviveka, not the Indian Madhyamaka tradition as a whole.
  • Śūnyatāsaptati v. 11 — Indian primary-text datum on the Tsongkhapa side. Per komito-seventy-stanzas-1987 / Śūnyatāsaptati, v. 11 reads: “Because ignorance and karmic formations are interrelated as cause and effect so these two are known by a valid cognizer (tshad ma) not to exist inherently.” Nāgārjuna himself, in a Yukti-corpus text, uses pramāṇa / tshad ma terminology positively to establish emptiness — not just to refute opponents. The grammar matches Tsong Khapa’s integrationist reading rather than Atiśa’s polemical-tool-only reading. Combined with Westerhoff’s 2018 note that the Vaidalyaprakaraṇa cannot be reduced to paraprasiddhānumāna — Nāgārjuna’s actual practice of refutation does not fit the later Tibetan Prāsaṅgika–Svātantrika typology — the Indian primary-text base is now visibly split between the VV vv. 30–51 anti-pramāṇa programme (which sits with Atiśa) and these two Tsong-Khapa-leaning data points (Śūnyatāsaptati v. 11, VP’s actual argumentative practice). The polarisation should be softened in of this wiki.
  • The Indian primary-text base is not univocal. Nāgārjuna’s Vigrahavyāvartanī is anti-pramāṇa; but Diṅnāga and Dharmakīrti — both recognised as Buddhist ācāryas — develop a sophisticated Buddhist pramāṇa theory. Tsongkhapa’s integration of pramāṇa into Madhyamaka is at least a candidate harmonisation of the two Indian Buddhist textual streams. Atiśa’s position requires either rejecting Diṅnāga–Dharmakīrti as outside Madhyamaka or assigning their pramāṇa theory to a strictly conventional register that does not bear on emptiness.
  • Path-dependence and pedagogy. The Geluk debate-yard tradition needs formal pramāṇa tooling for its pedagogy to function. Whether the institutional commitment to pramāṇa is a consequence of Tsongkhapa’s philosophical position or a cause of it is itself a question. The pedagogy may have philosophical force regardless of whether the textual base supports it.

Linked pages

Graph View

Backlinks