In the Albert-Loewer explanation of why we have abundant reliable records of the past but none of the future, the low-entropy initial macrostate (the Past State) is supposed to induce a 'density asymmetry': the probability distribution over the universe's macrohistories, conditional on the present macrostate, is sharply peaked over few past macrohistories but diffuse over future ones. Stradis (2024) argued that the only entropic derivation on offer -- the bottleneck argument, which traces this peaking to the Past State's tiny phase-space volume -- fails, because the low-entropy macrostates of the past are themselves small targets in phase space, and that justifying the density asymmetry by appeal to present records is circular. He proposed instead to derive the record asymmetry from the fork asymmetry (the common-cause structure of present correlations), itself grounded in the absence of conspiratorial correlations in the initial state. The open problem is threefold: give a non-circular derivation of the density asymmetry from the initial state, or show none is possible (settling the bottleneck argument's final status); make the fork-asymmetry derivation rigorous for generic systems, replacing the plausibility argument that conspiratorial initial microstates have negligible Lebesgue measure; and settle whether the density asymmetry reduces to the fork asymmetry.

Background

A long-standing puzzle in the foundations of statistical mechanics is that the microscopic dynamical laws are essentially time-symmetric, yet we reliably find records -- fossils, footprints, photographs, memory traces -- of the past and never of the future. This 'record asymmetry' is the target problem.

The framework in which the problem is posed is the Mentaculus of David Albert. Its ingredients: (i) time-symmetric, deterministic microscopic dynamics; (ii) the Past Hypothesis (PH), the posit that the universe began in a particular macrostate of very low entropy, called the Past State; and (iii) the Statistical Postulate (SP), a uniform probability measure (the Lebesgue, i.e. phase-space volume, measure) over the microstates compatible with the Past State. A microstate is a complete instantaneous state of the universe; a macrostate is its specification only up to macroscopic variables; a microhistory (macrohistory) is a trajectory of microstates (macrostates) through time. Following Stradis, conditionalizing the Mentaculus additionally on the present macrostate (the Current State) yields the 'Updated Mentaculus', a probability distribution over the world's macrohistories 'from today's standpoint'.

Albert and Barry Loewer explain the record asymmetry within this framework as follows. A recording device functions reliably only if it starts in an appropriate 'ready state' (blank paper, unexposed film, flat sand), and the fact that devices were in the right ready states is made overwhelmingly probable by the Past State. To use this story we must know which states are ready states, which seems to require further records, threatening a regress; Albert holds that the regress terminates because we are 'indirectly acquainted' with the Past State through the accumulated statistical impact of past frequencies on our record-forming apparatus. For this to work, the Past State must have a time-asymmetric statistical impact: it must ascribe higher probabilities to past events than to future ones. Albert and Loewer claim it does, through a 'density asymmetry': the Updated Mentaculus is sharply peaked over relatively few past macrohistories but thinly spread over many future macrohistories.

How is the density asymmetry supposed to follow from the Past State? Loewer's answer, which Stradis dubs the 'bottleneck argument', is that because the Past State occupies a tiny region of phase space while the Current State occupies a large one, microhistories must converge as one traces them from the present toward the past, and, since the SP assigns microhistories equal probabilities, the distribution over past macrohistories becomes sharply peaked. Mathias Frisch objected by analogy with a partially expanded gas: knowing only that a system started in some low-entropy macrostate leaves its earlier history wide open, which would predict a reversed density asymmetry. Stradis's own objection is subtler: the association between low entropy and small phase-space volume 'cuts both ways'. The early universe's low entropy means the macrostates entered toward the past are themselves small targets in phase space, less likely to be visited over and over, so that the corresponding macrohistories become more heterogeneous -- contrary to what the bottleneck argument needs.

Stradis's positive alternative invokes the 'fork asymmetry' of Hans Reichenbach's Common Cause Principle: whenever two events A and B are correlated (e.g. a footprint and a beach towel), there is typically an earlier event C (a stroller) that raises both their probabilities and 'screens off' their correlation (A and B are conditionally independent given C and given not-C), while structurally reversed 'backward forks' do not occur. Events A and B are the fork-tips; C is the fork-point. Interpreting fork-tips as records (or as sub-components of one record) explains both the collective and individual informativeness of records, and why only some things (photos, not random chunks of cloud) are records. The derivation of this structure from the initial state proceeds through determinants (Stradis's term for a set of physical circumstances that necessitates an event): for an instance of C there are background conditions XAX_A such that (C∧XA)(C \wedge X_A) necessitates A, and spurious-production circumstances YAY_A that exhaust the other ways of producing A (similarly for B). Since C appears in the necessary and sufficient conditions of both A and B, a forward fork exists whenever the background conditions are statistically independent, (XA∨YA)⊥ ⁣ ⁣ ⁣⊥(XB∨YB)(X_A \vee Y_A) \perp\!\!\!\perp (X_B \vee Y_B). Backward forks would require 'conspiratorial' correlations among highly specific early-universe circumstances, and the SP (being uniform) assigns such conspiratorial microstates vanishingly small probability -- so the fork asymmetry is claimed to follow from the initial state, non-circularly. On this basis Stradis re-orders the explanation: initial state, fork asymmetry, record asymmetry, density asymmetry -- the reverse of Albert and Loewer's 'initial state, density asymmetry, record asymmetry', in which the density asymmetry explains records and records can then not be used to justify the density asymmetry on pain of circularity.

Three gaps in this picture define the open problem. The bottleneck argument's failure leaves the density asymmetry without a non-circular derivation from the Past State. The fork-asymmetry derivation rests on a plausibility argument about the small Lebesgue measure of conspiratorial initial microstates, not a rigorous result for generic systems. And the claim that the density asymmetry is explained via the fork asymmetry plus records is asserted rather than established, with the rival orderings left undecided. Related but distinct foundational issues -- the justification of the Lebesgue measure as a probability measure, the privileging of thermodynamic macrovariables in the macrostate partition, and the underdetermination of the Past State's physical characteristics beyond 'very low entropy' -- are acknowledged open problems in the same literature but carry independent research programs of their own.

Problem Statement

Within the (Updated) Mentaculus -- time-symmetric deterministic dynamics plus the Past Hypothesis (the universe began in a particular very-low-entropy macrostate, the Past State) plus the Statistical Postulate (uniform Lebesgue probability over the Past State's compatible microstates), conditionalized on the Current State -- give a non-circular explanation of the record asymmetry, i.e. of why there are many reliable records of the past and none of the future, grounded in the initial state rather than in the records to be explained. This requires settling three linked questions.

(1) The density asymmetry and the bottleneck argument. Either exhibit a rigorous argument that the Past State's small phase-space volume (together with the Statistical Postulate and the absence of a low-entropy final constraint) entails a density asymmetry -- a sharply peaked distribution over past macrohistories and a diffuse distribution over future macrohistories of the Updated Mentaculus -- without anywhere appealing to facts about present records; or establish that no such derivation is available (e.g. by showing that the countervailing effect Stradis identified -- past low-entropy macrostates being small targets in phase space -- cannot be overcome by any phase-space or dynamical consideration, or by exhibiting, within a specified class of Mentaculus-like systems, a system satisfying the Past Hypothesis whose induced distribution over macrohistories fails to be past-peaked, which would establish unavailability only relative to that class rather than in full generality). A counterexample confined to a specified class of Mentaculus-like systems settles this negative horn only if the class is argued to be generic; otherwise a general no-go argument is required. The final status of the bottleneck argument must be explicitly settled.

(2) The fork asymmetry derivation. Provide a derivation, for generic systems and not merely by plausibility argument, of the fork asymmetry from the initial state: show that the statistical independence of the background conditions and spurious determinants associated with putative fork-tips, (XA∨YA)⊥ ⁣ ⁣ ⁣⊥(XB∨YB)(X_A \vee Y_A) \perp\!\!\!\perp (X_B \vee Y_B), holds with overwhelming probability under the Statistical Postulate restricted to the Past State -- in line with Stradis's own formulation, that microstates rife with the 'conspiratorial' correlations needed for backward forks have extremely small Lebesgue measure and are sparsely scattered across the Past State -- or else show that this independence fails for some generic class of systems, undermining the fork-asymmetry account. The derivation must state precisely which dynamical assumptions (mixing, coarse-graining, isolation of subsystems, etc.) it uses and verify they do not covertly presuppose the record asymmetry.

(3) The explanatory ordering. Determine whether the density asymmetry reduces to the fork asymmetry: whether the ordering 'initial state -> fork asymmetry -> record asymmetry -> density asymmetry' can be completed so that the last step (explaining the peaked distribution over past macrohistories by tallying what present records tell us) is non-viciously circular, or whether the density asymmetry instead plays an indispensable, independently grounded role. An answer must make explicit its criterion for benign versus vicious circularity and apply it uniformly to both orderings.

These three deliverables jointly constitute one objective: deciding which probabilistic asymmetry -- density or fork -- carries the explanatory burden for records, and whether the initial state can carry that burden at all without circularity. An answer that resolves only one prong while leaving the others untouched does not complete the problem as stated.

References

  1. Athamos Stradis, 'Present records of the Past Hypothesis', Synthese 203(6), article 214 (2024). DOI: 10.1007/s11229-024-04643-7, https://doi.org/10.1007/s11229-024-04643-7 (source paper)
  2. Athamos Stradis, 'Memory, the fork asymmetry, and the initial state', Synthese 199(3-4), 9523-9547 (2021). DOI: 10.1007/s11229-021-03214-4, https://doi.org/10.1007/s11229-021-03214-4 (predecessor paper developing the fork-asymmetry account of memory)
  3. Athamos Stradis, 'Global and Local Imperialism in Statistical Mechanics', Erkenntnis 91(1), 311-323 (2024). DOI: 10.1007/s10670-024-00892-7, https://doi.org/10.1007/s10670-024-00892-7
  4. David Z. Albert, 'Time and Chance', Harvard University Press (2000). DOI: 10.4159/9780674020139, https://doi.org/10.4159/9780674020139 (origin of the Mentaculus, the Past Hypothesis, and the ready-state account of records)
  5. Barry Loewer, 'Counterfactuals and the Second Law', in H. Price and R. Corry (eds.), 'Causation, Physics, and the Constitution of Reality: Russell's Republic Revisited', Oxford University Press, pp. 293-326 (2007). DOI: 10.1093/oso/9780199278183.003.0011, https://doi.org/10.1093/oso/9780199278183.003.0011 (source of the bottleneck argument as reconstructed by Stradis 2024)
  6. Mathias Frisch, 'Counterfactuals and the Past Hypothesis', Philosophy of Science 72(5), 739-750 (2005). DOI: 10.1086/508111, https://doi.org/10.1086/508111 (box-of-gas objection to PH-based explanations of records)
  7. David H. Wolpert and Jens Kipper, 'Memory Systems, the Epistemic Arrow of Time, and the Second Law', Entropy 26(2), 170 (2024). DOI: 10.3390/e26020170, https://doi.org/10.3390/e26020170 (rival formal framework)
  8. David H. Wolpert, Carlo Rovelli, and Jordan Scharnhorst, 'Disentangling Boltzmann Brains, the Time-Asymmetry of Memory, and the Second Law', Entropy 27(12), 1227 (2025). DOI: 10.3390/e27121227, https://doi.org/10.3390/e27121227
  9. Aram Ebtekar and Marcus Hutter, 'Modeling the Arrows of Time with Causal Multibaker Maps', Entropy 26(9), 776 (2024). DOI: 10.3390/e26090776, https://doi.org/10.3390/e26090776 (toy-model derivation of a Pearlean causal structure and past-only records)
  10. Thomas Blanchard, 'Causation and the Time-Asymmetry of Knowledge', Australasian Journal of Philosophy 102(4), 959-977 (2024). DOI: 10.1080/00048402.2024.2328629, https://doi.org/10.1080/00048402.2024.2328629 (causal-Markov derivation of the knowledge asymmetry)
  11. Meir Shimon, 'The Evidential Arrow of Time: Records, Reconstruction, and Empirical Science', Preprints.org preprint (2026). DOI: 10.20944/preprints202606.0839.v1, https://doi.org/10.20944/preprints202606.0839.v1 (top-down reorientation)
  12. Sean Gryb, 'New Difficulties for the Past Hypothesis', Philosophy of Science 88(3), 511-532 (2021). DOI: 10.1086/712879, https://doi.org/10.1086/712879 (gauge/measure difficulties for PH typicality)
  13. Chris Dorst, 'Does the Best System Need the Past Hypothesis?', Philosophy of Science 91(2), 410-429 (2023). DOI: 10.1017/psa.2023.152, https://doi.org/10.1017/psa.2023.152 (PH redundancy in the best-system package)