Martingale Transforms and Subsequence Estimates in Noncommutative L p-Spaces

arXiv:2610.18008 · math.OA, math.FA · Submitted 2026-10-13 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Mathematics. The day in mathematics: new theorems, new proofs, and what they open.

Ada: I'm Ada.

Emmy: And I'm Emmy, here with Kolya.

Ada: We're looking at the paper: "Martingale Transforms and Subsequence Estimates in Noncommutative L p-Spaces".

Paper summary: Ada: So we're looking at this paper titled "Martingale Transforms and Subsequence Estimates in Noncommutative L p-Spaces". It deals with von Neumann algebras, which are these really big mathematical structures that show up everywhere in operator theory. The main point of this work is proving that for any of these algebras, every bounded sequence you put into the Haagerup space L1(M), which is basically the dual space M star, it has to have either a subsequence that converges in norm or a specific type of sequence called a two co-lacunary one.

Emmy: That's right. The core thesis here is Theorem one point one, which states that for an arbitrary von Neumann algebra M, every bounded sequence in the Haagerup space L1(M) = M star has either a norm-convergent subsequence or a two co-lacunary subsequence <ref:2610.18008#pg1,bounded sequence in the Haagerup space L1(M) = M>. It’s addressing an open problem from Parcet and Randrianantoanina twenty-eight Remark three point one one and it manages to remove the requirement for hyperfiniteness or even semifiniteness in the algebra M itself <ref:2610.18008#pg1,Parcet and Randrianantoanina 28, Remark 3.11 and>.

Kolya: So what does this mean for us? The abstract says it answers that open problem because it removes those restrictions on the von Neumann algebra M, which is important because many of these structural results rely on those assumptions. I wonder if removing those constraints makes the proof significantly harder to manage in practice.

Ada: That’s a good question, Kolya. Well, the paper outlines a reduction strategy first. It starts by looking at seminormalized weakly null sequences and then it passes through several steps involving corners and different types of cases like one-sided compression or biequiintegrable cases before landing on the finite-trace setting.

Emmy: The structure of the proof seems to rely heavily on tools from noncommutative geometry, specifically using things like polar decompositions to build elements with orthogonal supports in the one-sided case, which is how they try to get that lower l2 estimate needed for their argument.

Kolya: I'm curious about those estimates. When they apply the L1-infinity duality inequality there, what kind of lower l2 estimate are they actually getting? Is it enough to control the tail behavior of the sequence correctly?

Ada: The paper uses that lower l2 estimate to control off-diagonal pairings, which is a key part in managing how these sequences interact within the algebra. Then, for the biequiintegrable case, they use a uniform two-sided approximation and introduce a compression map that goes from L1(M) to L1 of some specific projection space, which lets them reduce it to the finite-trace setting.

Paper summary: Emmy: The reduction to the finite-trace case is where things get really concrete. They establish a pruning lemma there, and then they use Mazur's lemma to construct a finite martingale difference sequence from that, which then leads to contradictions using both the non-commutative Lp cotype two inequality and some weak type one one martingale transform inequality.

Kolya: So the contradiction comes from combining those two different ways of bounding the sequence—one related to how well it behaves in terms of its Lp properties, and the other related to its martingale structure. That sounds like a solid way to force a conclusion about whether it converges or not.

Ada: It is. The implication is that this property holds for every von Neumann algebra, which is pretty broad since you can't always assume hyperfiniteness there, but the proof has these specific steps involving the sigma-finite corner and the four-corner decomposition to handle those tricky seminormalized weakly null sequences.

Emmy: And thinking about the structural consequences listed in section four, like that statement about a bounded linear operator between preduals being Dunford-Pettis if and only if it is l2-strictly singular, that shows how this subsequence property connects to the behavior of operators on these spaces.

Kolya: That link between the sequence property and operator theory seems very deep. It suggests that having this specific subsequence structure is essentially equivalent to having a certain kind of singularity in the operator itself, which is a structural statement about M star.

Ada: Right. And it also leads to Corollary four point five, which shows that for Haagerup Lp-spaces with one less than or equal to p less than or equal to two, you get the same subsequence property <ref:2610.18008#pg2>. That's a nice extension because it moves beyond just L1 and looks at these broader function spaces too.

Emmy: It also connects back to Theorem three point nine, which states that every seminormalized weakly null sequence in L1(M) has a two co-lacunary subsequence, and Corollary four point seven extends that to arbitrary semifinite von Neumann algebras when you look at the space L1 plus L2 with a trace.

Kolya: So to summarize for someone who only listens: this paper confirms that no matter what kind of von Neumann algebra you pick, if you have a bounded sequence in its dual space, it's either going to settle down into convergence or it has this specific two co-lacunary structure. It removes those structural restrictions that used to be necessary for these kinds of results.

Ada: Exactly. And the title itself points to the technique they are using—martingale transforms—which is central to how they manage those sequences in the final finite-trace setting before reaching the contradiction. The authors, Huang and Zhu, built a reduction from seminormalized weakly null cases down through these different structural decompositions.

Paper summary: Emmy: The implication for algebraic geometry or structure is that this sequence property behaves predictably across all types of von Neumann algebras without needing extra assumptions about their trace structure. It shows a more universal control over sequences in these spaces than was previously known.

Kolya: It's interesting how they manage to use the L1-infinity duality inequality specifically for the one-sided case; that’s where most of the technical machinery seems to be focused on ensuring those lower bounds stay controlled. I wonder if there are any limitations they flag in their work regarding the constants involved in that inequality.

Ada: They do mention some limitations, and it's important to note that the paper focuses on proving this for sequences in L1(M) itself, but extending it perfectly to all bounded sequences in these spaces requires careful handling of the different cases they discussed. They state that the result is established under certain conditions related to seminormalized weakly null cases first.

Emmy: So, looking at the title "Martingale Transforms and Subsequence Estimates in Noncommutative L p-Spaces", it really captures the flavor of what they're doing: using these transforms to get those estimates for different p values, which is what Corollary four point five explores later on with the optimal range being one less than or equal to two <ref:2610.18008#pg2>.

Kolya: And looking ahead, I think the next step might be seeing if this subsequence property holds for larger p, say p greater than two, because they specifically noted that for p greater than two, you don't necessarily get that two-co-lacunary subsequence <ref:2610.18008#pg1>.

Ada: That is a crucial caveat. The paper is very precise about where the result stops applying and why those specific constraints on the sequence or the space are necessary for their proof to work as intended. It’s a lot of machinery to get there, but it gives us a very solid floor on what we can expect regarding boundedness in these noncommutative settings.

Emmy: So, this paper is really about proving universality—showing that this specific type of sequence behavior is robust across the entire class of von Neumann algebras without needing those strong structural assumptions on M. It's a statement about the inherent structure of L1(M) itself.

Kolya: And it suggests that these martingale transforms are a powerful way to probe that structure, moving from general cases down to the finite-trace setting where they can actually construct the contradiction.

Ada: That’s pretty much it. The main point is establishing this subsequence dichotomy for L1(M) and then using those results to build up structural consequences about operators and other related spaces. This gives a strong tool for analyzing boundedness in noncommutative analysis.

Conclusion: Ada: So we've been looking at how this paper proves that in any von Neumann algebra, every bounded sequence in its dual space has either to converge or have this specific two co-lacunary structure.

Emmy: Yeah, and the title of the paper is "Martingale Transforms and Subsequence Estimates in Noncommutative L p-Spaces". It really tells you exactly what's going on here.

Kolya: It’s about using these martingale transforms to get those estimates, which is how they manage the sequence behavior across different p values.

Ada: The authors are Huang and Zhu, and this work shows that they can establish this result for a much wider range of algebras than before.

Emmy: It moves beyond just the standard ones where you know the structure of the algebra, showing this holds more generally.

Kolya: And it points to some structural consequences for operators, like how Dunford-Pettis operators relate to being l2 strictly singular.

Ada: So basically, they’re giving us a tool that helps us understand boundedness in these noncommutative settings by looking at how sequences behave under these martingale operations.

Emmy: It gives you a more universal way to handle those complex sequence problems without needing those extra assumptions on the algebra's trace structure.

Kolya: But it also sets clear boundaries, especially since they noted that for larger p values, you don't necessarily get that two co-lacunary subsequence.

Ada: That’s a key caveat there; it means we need to be careful about where this specific type of sequence behavior actually applies.

Emmy: And the paper sets up some interesting connections to other spaces, like those symmetric function spaces, showing how similar results apply there too.

Kolya: It suggests that these martingale tools are a really powerful way to probe the fundamental structure of L p-spaces in these algebras.

J. Huang, Y. Zhu

NNSF of China

math.OA, math.FA

Submitted: 2026-10-13

Updated: 2026-11-18

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

The gist: For every von Neumann algebra M, we prove that each bounded sequence in the Haagerup space L1(M) = M∗ admits either a norm-convergent subsequence or a 2-co-lacunary subsequence Main Result Theorem…

Key concepts

Haagerup space $L^1(M)$
$L^1(M)$ is the space of bounded linear functionals on a von Neumann algebra $M$. It is a fundamental Banach space in noncommutative analysis. The paper studies sequences within this specific space to determine their structural properties, specifically whether they converge or exhibit 2-co-lacunarity.
2-co-lacunary subsequence
A sequence is 2-co-lacunary if it satisfies a specific structural property related to its distribution. The proof shows that any bounded sequence in $L^1(M)$ must either have a convergent subsequence or possess this 2-co-lacunary structure, which is key to the main result.
Dunford–Pettis operator
An operator is Dunford–Pettis if it maps weakly null sequences to sequences with certain desirable properties. The paper connects the property of a bounded linear operator being Dunford–Pettis to being l2-strictly singular, providing structural characterizations for operators between preduals of von Neumann algebras.
Compression map $J_N$
A compression map $J_N$ is a specific type of operator that maps elements from the larger space $L^1(M)$ to a finite-trace space. This map is used in the proof to reduce complex problems involving arbitrary von Neumann algebras to simpler, more manageable finite-trace settings.

Terminology

Summary

For every von Neumann algebra M, we prove that each bounded sequence in the Haagerup space L1(M) = M∗ admits either a norm-convergent subsequence or a 2-co-lacunary subsequence

Main Result

Theorem 1.1 states that for an arbitrary von Neumann algebra M, every bounded sequence in the Haagerup space L1(M) = M∗ has either a norm-convergent subsequence or a 2-co-lacunary subsequence

Proof Outline

The proof reduces the main result to the seminormalized weakly null case (Theorem 3.9) by utilizing Rosenthal’s l1-theorem [33], the weak sequential completeness of M∗ [35, Chapter III, Corollary 5. p. 12], and the translation stability of 2-co-lacunarity

The proof proceeds through a series of reductions involving various cases:

  1. For seminormalized weakly null sequences, it passes to a σ-finite corner (Lemma 3. p. 1 p. 18) and then uses the Raynaud–Xu four-corner decomposition ([32, Corollary 4. p. 1 p. 112]) to reduce the problem to the one-sided compression case (Lemma 3. p. 1 p. 17) and the biequiintegrable case (Corollary 3. p. 1 p. 16)

  2. For the one-sided case, it uses polar decompositions to construct elements with orthogonal supports, controlling off-diagonal pairings and applying the L1–L∞ duality inequality to obtain a required lower l2-estimate

  3. For the biequiintegrable case, it uses the uniform two-sided approximation (Lemma 3. p. 1 p. 14) and introduces compression maps JN: L1(M) −→ L1(PN CPN, Tr), where PN = 1[e−N,eN) (D), allowing reduction to the finite-trace case (Corollary 3. p. 1 p. 13)

  4. In the finite-trace setting, it establishes a pruning lemma (Lemma 3. p. 1 p. 11) and uses Mazur’s lemma to construct a finite martingale difference sequence, yielding contradictory estimates from the non-commutative Lp cotype-2 inequality and the weak-type (1, 1) martingale-transform inequality

Structural Consequences

The paper establishes several structural consequences derived from Theorem 1.1:

((4.3) A bounded linear operator between the preduals of arbitrary von Neumann algebras is Dunford–Pettis if and only if it is l2-strictly singular. p. 1

((4.5) The subsequence property for Haagerup Lp-spaces, 1 ≤ p ≤ 2, associated with arbitrary von Neumann algebras. p. 1

((4.7) Every bounded sequence in (L1 + L2)(M, τ) has either a norm-convergent subsequence or a 2-co-lacunary subsequence when M is a semifinite von Neumann algebra. p. 0

Subsequence Characterizations

The paper provides characterizations related to compactness and Dunford–Pettis operators:

((4.1) A bounded subset K of M∗ is relatively norm compact if and only if it contains no 2-co-lacunary sequence. p. 0

((4.2) A bounded linear operator T: X → M∗ is Dunford–Pettis if and only if, for every weakly null sequence (xn) in X, the sequence (T xn) has no 2-co-lacunary subsequence. p. 1

((4.3) A bounded linear operator T: M∗ → N

is Dunford–Pettis if and only if T is l2-strictly singular.

Haagerup Lp-Spaces

Corollary 4.5 shows that for arbitrary von Neumann algebras M, every bounded sequence in the Haagerup space Lp(M) has either a norm-convergent subsequence or a 2-co-lacunary subsequence for 1 ≤ p ≤ 2

Seminormalized Weakly Null Sequences

Theorem 3.9 states that every seminormalized weakly null sequence in L1(M) = M∗ has a 2-co-lacunary subsequence

Corollary 4.7 extends this to arbitrary semifinite von Neumann algebras, showing that every bounded sequence in (L1 + L2)(M, τ) has either a norm-convergent subsequence or a 2-co-lacunary subsequence

Spreading Models

Corollary 4.4 states that for an arbitrary von Neumann algebra M, every seminormalized weakly null sequence in X isomorphic to a closed subspace of M∗ has a 2-co-lacunary subsequence, and every such infinite-dimensional X contains a basic 2-co-lacunary sequence

Corollary 4.5 also states that every spreading model generated by a seminormalized weakly null sequence in X dominates the unit vector basis of l2

Symmetric Banach Function Spaces

Theorem 4.8 provides an equivalence for a symmetric Banach function space E = E(M, τ) such that E ⊂ (L1 + L2)(0, ∞), stating that every bounded sequence in E(M, τ) has either a norm-convergent subsequence or a 2-co-lacunary subsequence

Optimal Range

Remark 4.6 notes that the range 1 ≤ p ≤ 2 in Corollary 4. p. 25 is optimal, and for p > 2, no subsequence is necessarily 2-co-lacunary

Extension to Arbitrary Algebras

Corollary 4.7 extends the result of Sukochev and Zhou [34, Corollary 3. p. 04 and Theorem 1. p. 01] to arbitrary semifinite von Neumann algebras

Final Result on (L1 + L2)

Corollary 4.7 proves that every bounded sequence in (L1 + L2)(M, τ) has either a norm-convergent subsequence or a 2-co-lacunary subsequence

Reference Summary

[1] D. J. Aldous and D. p. 0 H. Fremlin, Colacunary sequences in L-spaces, Studia Math. p. 6 71 (1982), no. 3, 297–304

[2] D. Alspach, R. Judd, and E. Odell, The Szlenk index and local l1-indices, Positivity 9 (2005), no. 1, 1–44

[3] G. Androulakis, E. Odell, Th. Schlumprecht, and N. Tomczak-Jaegermann, On the structure of the spreading models of a Banach space, Canad. p. 0 J. Math. 57 (2005), no. 4, 673–707

[4] S. V. Astashkin, Sequences of independent functions and structure of rearrangement invariant spaces, Russian Math. p. 4 Surveys 79 (2024), no. 3, 375–457

[5] C. Bennett and R. Sharpley, Interpolation of operators, Pure and Applied Mathematics, vol. p. 0 129, Academic Press, Boston, MA, 1988

[6] H. Brezis, Functional analysis, Sobolev spaces and partial differential equations, Universitext, Springer, New York, 2011

[7] A. Brunel and L. Sucheston, On B-convex Banach spaces, Math. p. 0 Systems Theory 7 (1974), no. 4, 294–299

[8] R. M. Causey and C. Doebele, A Ramsey theorem for pairs in trees, Fund. p. 7 Math. 248 (2020), no. p. 0 2, 147–193

[9] V. I. Chilin, P. G. Dodds, and F. A Sukochev, The Kadec–Klee property in symmetric spaces of measurable operators, Israel J. p. 1 Math. 97 (1997), 203–219

[10] J. Diestel, Sequences and series in Banach spaces, Graduate Texts in Mathematics, vol. p. 0 p. 0 92, Springer-Verlag, New York, 1984

[11] R. Diestel, Graph theory, 5th ed., Graduate Texts in Mathematics, vol. 173, Springer, Berlin, 2017

[12] P. G. Dodds, T. K. Dodds, and B. de Pagter, Noncommutative Banach function spaces, Math Z. p. 0 201 (1989), no. 4, 583–597

[13] P. G. Dodds and B. de Pagter, Normed Köthe spaces: a non-commutative viewpoint, Indag. p. 6 Math. (N.S.) 25 (2014), no. 2, 206–249

[14] P. Dodds, B. de Pagter, F. Sukochev, Sets of uniformly absolutely continuous norm in symmetric spaces of measurable operators, Trans. p. 4 Amer. Math. Soc, 368 (2016), no. 6, 4315-4355

[15] T. Fack and H. Kosaki, Generalized s-numbers of τ-measurable operators, Pacific J. p. 2 Math. 123 (1986), no. 2, 269–300

[16] U. Haagerup, Lp-spaces associated with an arbitrary von Neumann algebra, Algèbres d’opérateurs et leurs applications en physique mathématique (Proc. p. 1 Colloq., Marseille, 1977), Colloq. Internat. CNRS, vol. 274, CNRS, Paris, 1979

[17] U. Haagerup, M. p. 0 Junge, and Q. Xu, A reduction method for noncommutative Lp-spaces and applications, Trans. p. 0 Amer. Math. Soc. 362 (2010), no. 4, 2125–2165

[18] U. Haagerup and J. p. 0 Kraus, Approximation properties for group C∗-algebras and group von Neumann algebras, Trans. p. 0 Amer. Math. Soc. 344 (1994), no. 2, 667–699

[19] U. Haagerup, H. p. 0 P. Rosenthal, and F. p. 1 A Sukochev, Banach embedding properties of noncommutative Lp-spaces, Mem. p. 0 Amer. Math. Soc. 163 (2003), no. 776, vi+68 pp

[20] C. W. Henson, N. J. Kalton, N. T. Peck, I Tereščák, and P Zlatoš, Some Ramsey type theorems for normed and quasinormed spaces, Studia Math. 124 (1997), no. 1, 81–100

[21] J. Huang, M. Junge, F Sukochev, and D Zanin, Embeddings of symmetric operator spaces into Lp-spaces, 1 ≤ p < 2, on finite von Neumann algebras, Israel J. p. 0 Math. 268 (2025), 73–116

[22] J. Huang, M Pliev, and F Sukochev, l2-strictly singular operators on the predual of a hyperfinite von Neumann algebra, Proc. p. 1 Amer. Math. Soc. 150 (2022), no. 4, 1623–1633

[23] J Huang, M Pliev, and F Sukochev, On narrow operators from Lp into operator ideals, Mediterr. J. Math. 19 (2022), Art. 208

[24] J.-P. Kahane, Some random series of functions, 2nd ed. p. 2, Cambridge Studies in Advanced Mathematics, vol. p. 0 5, Cambridge University Press, Cambridge, 1985

[25] N. J. Kalton, Rademacher series and decoupling, New York J. p. 2 Math. 11 (2005), 563–595

[26] E. Nelson, Notes on non-commutative integration, J. p. 9 Funct. Anal. 15 (1974), 103–116

[27] E. Odell and Th Schlumprecht, A problem on spreading models, J. p. 0 Funct. Anal. 153 (1998), no. 1, 249–261

[28] J. Parcet and N Randrianantoanina, Gundy’s decomposition for non-commutative martingales and applications, Proc. p. 0 Lond. Math. Soc. (3) 93 (2006), no. 1, 227–252

[29] G. Pisier and Q Xu, Non-commutative Lp-spaces, in Handbook of the geometry of Banach spaces, vol. p. 0 2, W B Johnson and J Lindenstrauss (eds.), North-Holland, Amsterdam, 2003

[30] N. Randrianantoanina, Embeddings of non-commutative Lp-spaces into preduals of finite von Neumann algebras, Israel J. p. 4 Math. 163 (2008), 1–27

[31] N. Randrianantoanina, Kadec–Pełczyński decomposition for Haagerup Lp-spaces, Math. p. 0 Proc. Cambridge Philos. Soc. 132 (2002), no. 1, 137–154

[32] Y Raynaud and Q Xu, On subspaces of non-commutative Lp-spaces, J. Funct. Anal. 203 (2003), no. 1, 149–196

[33] H P Rosenthal, A characterization of Banach spaces containing l1, Proc Natl Acad Sci USA 71 (1974), no. 6, 2411–2413

[34] F Sukochev and D Zhou, 2-co-lacunary sequences in noncommutative symmetric Banach spaces, Proc Amer Math Soc. 148 (2020), no. p. 0 5, 2045–2058

[35] M Takesaki, Theory of operator algebras. I, Springer-Verlag, New York–Heidelberg, 1979

Page 3

The paper is titled Martingale Transforms and Subsequence Estimates in Noncommutative L p-Spaces and it matters because it proves that for every von Neumann algebra M, each bounded sequence in the Haagerup space L1(M) = M∗ admits either a norm-convergent subsequence or a 2-co-lacunary subsequence

The proof proceeds through a series of reductions involving various cases:

  1. For seminormalized weakly null sequences, it passes to a σ-finite corner (Lemma 3.8) and then uses the Raynaud–Xu four-corner decomposition ([32, Corollary 4.12]) to reduce the problem to the one-sided compression case (Lemma 3.7) and the biequiintegrable case (Corollary 3.6)

  2. For the one-sided case, it uses polar decompositions to construct elements with orthogonal supports, controlling off-diagonal pairings and applying the L1–L∞ duality inequality to obtain a required lower l2-estimate

  3. For the biequiintegrable case, it uses the uniform two-sided approximation (Lemma 3.4) and introduces compression maps JN: L1(M) −→ L1(PN CPN, Tr), where PN = 1[e−N,eN) (D), allowing reduction to the finite-trace case (Corollary 3.3)

  4. In the finite-trace setting, it establishes a pruning lemma (Lemma 3.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Improve sequence analysis capabilities by implementing a formal check for 2-co-lacunary subsequences in arbitrary L1 spaces over von Neumann algebras, allowing systems to distinguish between norm-convergent and 2-co-lacunary behavior for any bounded sequence.

  2. Enhance operator theory applications by enabling the system to characterize Dunford–Pettis operators as being equivalent to l2 strictly singular operators on predual spaces, a generalization that applies beyond hyperfinite von Neumann algebras.

  3. Develop a tool for analyzing infinite dimensional Banach spaces by using the result that every infinite dimensional X contains a basic 2-co-lacunary sequence, improving the ability to find structural elements in such spaces.

  4. Implement a method to determine if an operator T is Dunford–Pettis by testing if it restricts to an isomorphic embedding on some closed subspace isomorphic to l2, providing a concrete operational test for this property.

  5. Create a function that verifies whether every bounded sequence in Haagerup Lp spaces (for 1 less than or equal to p less than or equal to 2) possesses either a norm-convergent subsequence or a 2-co-lacunary subsequence, covering the full range of the theorem.

Related papers