Prehistory of FMZVs Michael E. Hoffman Outline Some Prehistory of Finite Multiple Zeta Values Multiple Zeta Values Multiple Harmonic Sums mod p Michael E. Hoffman A New Start: QuasiSymmetric Functions U. S. Naval Academy True Duality at Last Back to Low Weights Kyushu University Workshop on Multiple Zeta Values 22 August 2014 Michael E. Hoffman Prehistory of FMZVs Outline Prehistory of FMZVs Michael E. Hoffman 1 Multiple Zeta Values Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions 2 Multiple Harmonic Sums mod p 3 A New Start: Quasi-Symmetric Functions 4 True Duality at Last True Duality at Last Back to Low Weights 5 Back to Low Weights Michael E. Hoffman Prehistory of FMZVs Multiple Zeta Values Prehistory of FMZVs Michael E. Hoffman Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions For positive integers i1 , i2 , . . . ik with i1 > 1 (to ensure convergence), the multiple zeta value (henceforth MZV) ζ(i1 , i2 , . . . , ik ) is the sum of the k-fold series X 1 . ik i1 i2 n1 >n2 ···>nk ≥1 n1 n2 · · · nk We call k the depth of ζ(i1 , . . . , ik ), and i1 + · · · + ik its weight. Of course, the MZVs of depth one are the values ζ(i) of the Riemann zeta function at positive integers i > 1, for which Euler proved the formula True Duality at Last Back to Low Weights ζ(2i) = (2π)2i |B2i |, 2(2i)! but he also studied depth-two MZVs. In the late 1980’s several people starting studying MZVs of general depth. Michael E. Hoffman Prehistory of FMZVs My Early Work on MZVs Prehistory of FMZVs Michael E. Hoffman Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights In the summer of 1988, my colleague Courtney Moen interested me in what he called the “sum conjecture”: for n > k ≥ 2, X ζ(a1 , a2 , . . . , ak ) = ζ(n). a1 >1, a1 +···+ak =n The case k = 2 had been proved by Euler in his 1776 paper. Courtney eventually came up with a very long proof of the case k = 3. I plunged into the search for a general proof, but instead discovered two other properties of MZVs: the duality and derivation theorems. Neither gave a quick proof of the sum conjecture, though the derivation theorem reduced the k = 3 case to a three-line proof. Michael E. Hoffman Prehistory of FMZVs Duality of MZVs Prehistory of FMZVs Michael E. Hoffman Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights Because of the influence of the duality theorem on my later work, I state it here, in the form I gave it in my 1992 paper. Let I = (i1 , . . . , ik ) be a composition (finite sequence of positive integers) with i1 > 1. Let S(I ) = (i1 , i1 + i2 , . . . , i1 + · · · + ik ) be the corresponding sequence of partial sums. If In is the set of strictly increasing sequences consisting of integers from {1, 2, . . . , n}, then there are two involutions on In : R given by R(a1 , . . . , ak ) = (n + 1 − ak , . . . , n + 1 − a1 ), and C which takes (a1 , . . . , ak ) to its complement in {1, 2, . . . , n}, arranged in increasing order. Then I defined τ (I ) = S −1 CRS(I ), and the duality theorem was the result that ζ(τ (I )) = ζ(I ). Michael E. Hoffman Prehistory of FMZVs Duality of MZVs cont’d Prehistory of FMZVs Michael E. Hoffman Outline Actually I couldn’t prove the duality theorem in general, but only in the “height 1” case where I = (a + 1, 1, . . . , 1) for a positive integer a. In this case the duality theorem reads Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights ζ(a + 1, 1, . . . , 1) = ζ(b + 1, 1, . . . , 1). | {z } | {z } (1) a−1 b−1 Of course now we know that the general duality theorem follows immediately from the iterated-integral representation of MZVs. But in those early days duality was still pretty mysterious to me: I had guessed it from doing extensive calculations in weight ≤ 6. Michael E. Hoffman Prehistory of FMZVs Truncated Multiple Harmonic Sums Prehistory of FMZVs Michael E. Hoffman The first phase of my joint work with Courtney on MZVs was done by about 1992. We then spent some time trying to understand the truncated multiple harmonic sums (MHSs) Outline ζn (i1 , . . . , ik ) = Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights X i1 i2 n≥n1 >n2 ···>nk ≥1 n1 n2 1 · · · nkik which are simply rational numbers. They still had the “stuffle” multiplication, e.g., ζn (2)ζn (3) = ζn (2, 3) + ζn (3, 2) + ζn (5), but now any composition could be the exponent string since convergence wasn’t an issue. In particular, 1 n+1 ζn (1, . . . , 1) = , | {z } n! r + 1 r where n k is the unsigned Stirling number of the first kind. Michael E. Hoffman Prehistory of FMZVs Mod p Finite Sums Prehistory of FMZVs Michael E. Hoffman Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights If p is prime, all the denominators in ζp−1 (I ) are invertible in Z/pZ, and so ζp−1 (I ) makes sense as an element of that ring. Courtney and I were already thinking about mod p values of ζp−1 (I ) when Y. Matiyasevich published a note in the “Unsolved Problems” section of the American Mathematical Monthly (January 1992) titled “What divisibility properties to generalized harmonic sums have?”. It presented several conjectures about ζp−1 (1, . . . , 1) without any awareness of the Stirling number formula above. (The May 1992 issue carried a sort of retraction, giving some references the author should have been aware of.) Courtney and I had submitted a note proving the conjectures, but all such submissions were rejected. Nevertheless, we continued studying the mod p values of the sums ζp−1 (I ). Michael E. Hoffman Prehistory of FMZVs Wolstenholme’s Theorem Prehistory of FMZVs Michael E. Hoffman Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights The most obvious result concerning such sums was Wolstenholme’s theorem that ζp−1 (1) ≡ 0 mod p for prime p > 2, and Courtney and I had no trouble generalizing it to ζp−1 (r ) ≡ 0 mod p for prime p > r + 1. So the single zeta values were trivial in this theory. Michael E. Hoffman Prehistory of FMZVs What Ring Were We Working In? Prehistory of FMZVs Michael E. Hoffman A New Start: QuasiSymmetric Functions Let me digress for a a moment to say that early on, we didn’t worry too much about restrictions on p in our formulas: we regarded two expressions as the same if they were congruent mod p for p sufficiently large. We never formalized this by working in the ring Q Z/pZ Pp prime p prime Z/pZ True Duality at Last as is usual nowadays, but it was the same idea. Outline Multiple Zeta Values Multiple Harmonic Sums mod p Back to Low Weights Michael E. Hoffman Prehistory of FMZVs Symmetric Sums Prehistory of FMZVs Michael E. Hoffman Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights In my 1992 paper on MZVs I gave a result expressing symmetric sums of MZVs as rational polynomials in the single zeta values, e.g., 1 1 ζ(3, 2, 2) + ζ(2, 3, 2) + ζ(2, 2, 3) = ζ(3)ζ(2)2 − ζ(3)ζ(4) 2 2 − ζ(2)ζ(5) + ζ(7) The result carries over to MHSs, and in view of the Wolstenholme-type results this meant symmetric sums of MHSs were simply 0 mod p. The first weight where one has an asymmetrical sum is 3, and ζp−1 (2, 1) is not 0 mod p; in fact we saw quickly that ζp−1 (2, 1) ≡ Bp−3 mod p for p > 3. Michael E. Hoffman Prehistory of FMZVs Motivation Prehistory of FMZVs Michael E. Hoffman Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights This gave us some motivation to pursue the study of mod p MHSs. We thought of them as a sort of “toy model” for MZVs. A basic question about MZVs is whether they are really something beyond single zeta values. It certainly seems that they are, since (for example) nobody has been able to write ζ(6, 2) as a rational polynomial in the single zeta values; but we also don’t know how to prove ζ(6, 2) isn’t such a polynomial (or for that matter that it isn’t rational!). For MHSs all the “single zetas” are zero mod p, and yet we had multiple sums that were clearly nonzero. It was easy to write programs to produce the mod p values of MHSs, and thus easy to test conjectures about them. Michael E. Hoffman Prehistory of FMZVs Depths 2 and 3 Prehistory of FMZVs Michael E. Hoffman Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights For depth 2 we had the general result (−1)i i + j Bp−i−j ζp−1 (i, j) ≡ i +j i mod p, (2) where p > i + j + 1 and Bn is the nth Bernoulli number, and we had a similar but more complicated result describing ζp−1 (i, j, k) in terms of binomial coefficients and Bernoulli numbers. Of course congruence (2) implies ζp−1 (i, j) ≡ 0 mod p if i + j is even, and our depth-3 result implied that ζp−1 (i, j, k) was congruent mod p to a multiple of ζp−1 (i + j + k − 1, 1) if i + j + k was odd. Michael E. Hoffman Prehistory of FMZVs My Search for Duality Prehistory of FMZVs Michael E. Hoffman Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights I was eager to find a mod p counterpart of the duality theorem for MZVs. In fact, I observed early on that ζp−1 (a, 1, . . . , 1) ≡ ζp−1 (b, 1, . . . , 1) | {z } | {z } b−1 mod p, (3) a−1 and, using mod p properties of (both kinds of) Stirling numbers, was eventually able to prove it. This led me to believe I was on the verge of getting a general mod p duality result. I just had to modify my definition of τ so that it produced (3) instead of (1). Michael E. Hoffman Prehistory of FMZVs A Disappointment Prehistory of FMZVs Michael E. Hoffman Outline Multiple Zeta Values Multiple Harmonic Sums mod p Actually this wasn’t hard to do. In the partial-sum step of the definition of τ I took all but the last term: S̄(a1 , . . . , ak ) = (a1 , a1 + a2 , . . . , a1 + · · · + ak−1 ) so that the partial sums are in {1, . . . , n − 1} for n = a1 + · · · + ak . Then I considered the function A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights τ̄ (I ) = S̄ −1 CR S̄(I ) which takes (a, 1, . . . , 1) to (b, 1, . . . , 1). But I was in for a | {z } | {z } b−1 a−1 disappointment: when applied to compositions I not of height 1, the congruence ζp−1 (τ̄ (I )) ≡ ζp−1 (I ) mod p simply isn’t true. Michael E. Hoffman Prehistory of FMZVs A Discovery Delayed Prehistory of FMZVs Michael E. Hoffman I had actually missed the real duality result, and wouldn’t find it for another five years. In retrospect, the problem was that I was wedded to using the sums Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights 1 · · · nkak X ζp−1 (a1 , . . . , ak ) = n a1 n a2 p−1≥n1 >n2 >···>nk ≥1 1 2 rather than ? ζp−1 (a1 , . . . , ak ) = X 1 . · · · nkak n a1 n a2 p−1≥n1 ≥n2 ≥···≥nk ≥1 1 2 This was understandable given my experience with MZVs (where the only “clean” duality result involves the unstarrred sums). But for the mod p MHSs, it is the latter kind of sums that have the nice duality. Michael E. Hoffman Prehistory of FMZVs Low-Weight Calculations Prehistory of FMZVs Michael E. Hoffman Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights Courtney and I kept up our joint work on mod p multiple harmonic sums through 1995, when I gave a talk about it at a sectional meeting of the American Mathematical Society in Greensboro, North Carolina. We did some calculations in low weights, using our results for double and triple series mod p. This allowed us to find, correctly, that in weight 3 everything is a multiple of ζp−1 (2, 1), that everything in weight 4 is zero mod p, and that everything in weight 5 is a multiple of ζp−1 (4, 1). But the calculations got bogged down in weight 6 and higher. The problem was that we didn’t know enough congruences for sums of higher depth. Michael E. Hoffman Prehistory of FMZVs A Shift of Focus Prehistory of FMZVs Michael E. Hoffman Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last In 1996 I shifted my attention back to ordinary MZVs. I spent most of that year struggling with the algebra of MZV multiplication, and eventually discovered that the quasi-shuffle or “stuffle” algebra was essentially the algebra of quasi-symmetric functions. This was a solo effort, as Courtney wasn’t as interested in algebra as I was. After I wrote my 1997 paper on MZVs, I broadened my focus somewhat and again thought about the mod p MHSs, but now from a more advanced algebraic perspective. Back to Low Weights Michael E. Hoffman Prehistory of FMZVs Quasi-Symmetric Functions Prehistory of FMZVs Michael E. Hoffman Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights I started thinking of the algebra of mod p MHSs as just another homomorphic image of QSym, the algebra of quasi-symmetric functions. The algebra QSym can be thought of as a subalgebra of the formal power series ring in x1 , x2 , . . . , where each xi has degree 1. A formal power series f of bounded degree is quasi-symmetric if, for any i1 < i2 < · · · < ik , the coefficient in f of xia11 xia22 · · · xiakk is the same as the coefficient in f of x1a1 x2a2 · · · xkak . Any quasi-symmetric function is a linear combination of the monomial quasi-symmetric functions X M(a1 ,a2 ,...,ak ) = xia11 xia22 · · · xiakk , i1 <i2 <···<ik which are indexed by compositions (a1 , . . . , ak ). Michael E. Hoffman Prehistory of FMZVs MZVs, MHSs, and Quasi-Symmetric Functions Prehistory of FMZVs Michael E. Hoffman Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights One can think of MZVs as the images under a homomorphism ζ : QSym0 → R, where QSym0 is the subalgebra generated by the M(a1 ,...,ak ) with ak > 1, given by ζ(M(a1 ,...,ak ) ) = ζ(ak , . . . , a1 ); the homomorphism simply sends xj to 1j . Similarly, for each prime p there is a homomorphism ζp−1 : QSym → Z/pZ sending ( 1 mod p, if j < p, xj → j 0, otherwise, so that ζp−1 (M(a1 ,...,ak ) ) = ζp−1 (ak , . . . , a1 ). Michael E. Hoffman Prehistory of FMZVs Multiple Harmonic Sums as Images Prehistory of FMZVs Michael E. Hoffman Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights For a composition I = (i1 , . . . , ik ), we write `(I ) = k (the length of I ) and |I | = i1 + · · · + ik (the weight of I ). Also, we write Ī for the reverse (ik , . . . , i1 ) of I . Then our homomorphism ζp−1 : QSym → Z/pZ sends MI to ζp−1 (Ī ). ? (Ī ) is the image of Also, ζp−1 X EI = MJ , (4) JI where is the refinement order on compositions, i.e. J I if J can be obtained by combining adjacent parts of I . Now while the “fundamental quasi-symmetric functions” X FI = MJ JI were well studied, the EI were apparently new. Michael E. Hoffman Prehistory of FMZVs QSym as a Hopf Algebra Prehistory of FMZVs Michael E. Hoffman Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights Actually the EI weren’t entirely absent from the literature on QSym (though they hadn’t been named). QSym has a Hopf algebra structure given by X ∆(MI ) = MI1 ⊗ MI2 , I1 I2 =I where the sum is over all expressions of I as the juxtaposition of compositions I1 , I2 . Then the antipode S of the Hopf algebra QSym is given by S(MI ) = (−1)`(I ) EĪ . Michael E. Hoffman Prehistory of FMZVs (5) QSym as a Hopf Algebra cont’d Prehistory of FMZVs Michael E. Hoffman Since QSym is a commutative Hopf algebra, S is an automorphism of QSym as an algebra, and S 2 = id. Thus any formula in the MI , e.g., Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights M(2) M(1,3) = M(2,1,3) + M(1,2,3) + M(1,3,2) + M(3,3) + M(1,5) has a corresponding version in the EI by applying S: −E(2) E(3,1) = −E(3,1,2) − E(3,2,1) − E(2,3,1) + E(3,3) + E(5,1) . There was no reason to think of the MI as more fundamental than the EI , and by Möbius inversion of equation (4) X MI = (−1)`(I )−`(J) EJ . JI Michael E. Hoffman Prehistory of FMZVs QSym as a Hopf Algebra cont’d Prehistory of FMZVs Michael E. Hoffman Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Also, by comparing (5) with another formula for S one has X (6) (−1)`(I ) EI = (−1)k MI1 · · · MIk , I1 I2 ···Ik =Ī where the sum is over all expressions of Ī as a juxtaposition of compositions. For example, − E(2,1,3) = −M(3,1,2) + M(3) M(1,2) + M(3,1) M(2) − M(3) M(1) M(2) . Back to Low Weights Michael E. Hoffman Prehistory of FMZVs Operators on Sequences Prehistory of FMZVs Michael E. Hoffman I also played around with operators on sequences {a(n)}∞ n=0 . In particular, one has the partial-sum operator Σ given by Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions Σa(n) = a(n) i=0 and the operator ∇ given by ∇a(n) = True Duality at Last Back to Low Weights n X n X i=0 (−1)i n a(i). i It is easy to see that ∇ and Σ generate an infinite dihedral group under composition, i.e, ∇2 = id and ∇Σ = Σ−1 ∇. Michael E. Hoffman Prehistory of FMZVs True Duality at Last Prehistory of FMZVs Michael E. Hoffman Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights If for each composition I we define a sequence ζI? by ζI? (n) = ζn? (I ), then I noticed that Σ∇ζI? = −ζI?∗ , where I ∗ = S̄ −1 C S̄(I ) in the notation introduced above (so (3, 1, 2)∗ = (1, 1, 3, 1)). Also, it is easy to see that Σ∇a(p) ≡ a(p) mod p for prime p. So now I had my true duality theorem: ? ? ζp−1 (I ) ≡ −ζp−1 (I ∗ ) mod p. I don’t know exactly when I proved this, but it was prior to September 2000 (when I gave a talk about it). Michael E. Hoffman Prehistory of FMZVs True and False Duality Prehistory of FMZVs Michael E. Hoffman It is interesting to see how I was led astray by the result (3), which is essentially a coincidence. It is a general (and easily proved) fact that Outline Multiple Zeta Values ? ? ζp−1 (ik , . . . , i1 ) ≡ (−1)i1 +···+ik ζp−1 (i1 , . . . , ik ) mod p. (7) Multiple Harmonic Sums mod p Since (a, 1, . . . , 1)∗ = (1, . . . , 1, b), if we apply the duality | {z } | {z } A New Start: QuasiSymmetric Functions theorem to I = (a, 1, . . . , 1) and then use the congruence (7), | {z } True Duality at Last we have Back to Low Weights a−1 b−1 b−1 ? ? (−1)a ζp−1 (a, 1, . . . , 1) = (−1)b ζp−1 (b, 1, . . . , 1) | {z } | {z } a−1 b−1 Michael E. Hoffman Prehistory of FMZVs True and False Duality cont’d Prehistory of FMZVs Michael E. Hoffman Now applying ζp−1 to equation (6) gives X ? (−1)`(I ) ζp−1 (I ) = (−1)k ζp−1 (I1 ) · · · ζp−1 (Ik ). I1 I2 ···Ik =Ī Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights (8) If I = (a, 1, . . . , 1), then mod p the right-hand side collapses to | {z } b−1 a single term (since ζp−1 (1, . . . 1) ≡ 0 mod p) and we have ? (a, 1, . . . , 1) ζp−1 (a, 1, . . . , 1) ≡ (−1)a ζp−1 | {z } | {z } b−1 mod p, (9) b−1 from which congruence (3) follows. But congruence (9) depends very strongly on the special features of the composition (a, 1, . . . , 1). For I not of this form the right-hand side of (8) can have many terms that are nonzero mod p. Michael E. Hoffman Prehistory of FMZVs A Useful Byproduct Prehistory of FMZVs Michael E. Hoffman Outline Nevertheless, my pursuit of “false duality” left me with one byproduct that was of some use later. As a result of my investigation of the MHSs ζn (a, 1, . . . , 1), I proved that | {z } b−1 Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights ? (−1)a ζp−1 (a, 1, . . . , 1) ≡ ζp−1 (a, 1, . . . , 1) ≡ | {z } | {z } b−1 p−b X b−1 j (−1) (−j) j=1 p−a p−b (j − 1)! j mod p, where kn is the Stirling number of the second kind. This allowed me to compute these MHSs much more effectively than doing it directly from the definition. Michael E. Hoffman Prehistory of FMZVs A Barrier Removed Prehistory of FMZVs Michael E. Hoffman Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights Now that I had a valid duality result, the barrier to computing MHSs in low weights was removed. For fixed weight, small-depth MHSs are dual to large-depth MHSs because of the relation `(I ) + `(I ∗ ) = |I | + 1. Nevertheless, though I included a section on finite sums mod p in my survey article published in 2005 (but written two years earlier), I didn’t really have the opportunity to renew my computational attack on low weights until my sabbatical in 2003-4. I made use of both the duality theorem and some results derived from equation (6) ? (I ) in terms of ζ ? (J) for J of smaller that expressed ζp−1 p−1 depth. At that time I worked out generators through weight 9. The results were as follows. Michael E. Hoffman Prehistory of FMZVs Results for |I | ≤ 9 Prehistory of FMZVs Michael E. Hoffman Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights |I | ≤ 2: all zero ? (2, 1) ≡ B |I | = 3: mod p all MHSs multiples of ζp−1 p−3 mod p. |I | = 4: all zero ? (4, 1) ≡ B |I | = 5: mod p all MHSs multiples of ζp−1 p−5 mod p. ? (4, 1, 1) ≡ − 1 B 2 |I | = 6: mod p all MHSs multiples of ζp−1 6 p−3 ? (6, 1) ≡ B |I | = 7: mod p all MHSs combinations of ζp−1 p−7 ? (4, 1, 1). mod p and ζp−1 ? (6, 1, 1) and |I | = 8: mod p all MHSs combinations of ζp−1 ? ? ζp−1 (2, 1)ζp−1 (4, 1) ≡ Bp−3 Bp−5 mod p. ? (6, 1, 1, 1), |I | = 9: mod p all MHSs combinations of ζp−1 ? ζp−1 (8, 1) ≡ Bp−9 mod p, and ? (2, 1)ζ ? (4, 1, 1) ≡ − 1 B 3 ζp−1 p−1 6 p−3 mod p. Michael E. Hoffman Prehistory of FMZVs Missing Congruences? Prehistory of FMZVs Michael E. Hoffman Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights This gives counts of wt. gens. 1 0 2 0 3 1 4 0 5 1 6 1 7 2 8 2 9 3 In weights 7,8,9 I had the nagging suspicion that I might have missed some congruences. The only way I could think of to test independence was to find a prime that divided one Bernoulli number. In weight 7 this failed because for no known primes p does p|Bp−7 . In weight 8 I was reassured because for ? (4, 1) ≡ 0 mod p but ζ ? (6, 1, 1) 6≡ 0 mod p. In p = 37, ζp−1 p−1 weight 9 I looked at the only known p’s for which ? (8, 1) ≡ 0 mod p, namely p = 67 and p = 877, and at ζp−1 p = 16, 843, for which p|Bp−3 . This gave inconclusive results. Michael E. Hoffman Prehistory of FMZVs Missing Congruences? cont’d Prehistory of FMZVs Michael E. Hoffman Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights Later I learned that Jianqing Zhao had done similar computations, and he informed me that he suspected a congruence 1 ? ? ? (2, 1)ζp−1 (4, 1, 1) ζp−1 (6, 1, 1, 1) ≡ − ζp−1 9 1889 ? + ζ (8, 1) mod p (10) 648 p−1 in weight 9. When I saw that the congruence (10) was consistent with the values I’d computed at the three primes p = 67, p = 877, and p = 16, 843, I was strongly inclined to believe it. Michael E. Hoffman Prehistory of FMZVs Recent Developments Prehistory of FMZVs Michael E. Hoffman Outline Multiple Zeta Values Multiple Harmonic Sums mod p A New Start: QuasiSymmetric Functions True Duality at Last Back to Low Weights In June 2012 I saw the preprint “New properties of multiple harmonic series modulo p and p-analogues of Leschinger’s series,” by Kh. Hessami Pilehrood, T. Hessami Pilehrood and R. Tauraso; it has since appeared in the AMS Transactions. The authors not only proved the congruence Zhao suspected, but also found another congruence ? ζp−1 (4, 1, 1, 1) ≡ 27 ? ζ (6, 1) 16 p−1 mod p in weight 7. With these results, the revised counts are wt. gens. 1 0 2 0 3 1 4 0 5 1 6 1 7 1 8 2 9 2 and satisfy the Padovan recurrence Pn = Pn−2 + Pn−3 . Michael E. Hoffman Prehistory of FMZVs