The Wolfram Mathworld has the following statement here:

A derivative identity for expressing higher order modified Bessel functions in terms of $I_0(x)$ is $I_n(x)=T_n(d/(dx))I_0(x)$

However, I'm not sure how to translate that into Mathematica syntax. My simple minded attempts have failed:

BesselI[l, x] /. {l -> 0.5, x -> 0.100}
D[ChebyshevT[l, x], x]*BesselI[0, x] /. {l-> 0.5, x -> 0.100}
ChebyshevT[l, D[BesselI[0, x], x]]*BesselI[0, x] /. {l -> 0.5, x -> 0.100}
ChebyshevT[l, D[BesselI[0, x], x]] /. {l -> 0.5, x -> 0.100}

Apologies for being obtuse if I've missed something obvious


5 Answers 5


You can make use of my DifferentialOperator paclet to do this. Install with:


and load with:


Then (I had to use an image since the partial derivative in the input doesn't translate well to MSE):

enter image description here


Waiting for a more elegant solution, you can try this:

Creates a differential operator from a polynomial in x:

DiffOp[poly_] /; PolynomialQ[poly, x] := Block[{Dop},
  Dop[c_, {ord_}] := c*(D[#, {x, ord - 1}] &);
  Return[Total[MapIndexed[Dop, CoefficientList[poly, x]]]];

For instance:

op = DiffOp[ChebyshevT[4, x]]

FullSimplify[op[BesselI[0, x]]]

BesselI[4, x]

Update: maybe slightly better (and without Return[], see Carl Woll remark):

DiffOp[poly_, x_, f_] /; PolynomialQ[poly, x] := 
 CoefficientList[poly, x].Table[D[f, {x, i}], {i, 0, Exponent[poly, x] }]

For instance:

DiffOp[ChebyshevT[4, x], x, f[x]]

prints: $$ 8 f^{(4)}(x)-8 f''(x)+f(x) $$


FullSimplify[DiffOp[ChebyshevT[4, x], x, BesselI[0,x]]]



  • $\begingroup$ It seems this is peculiar to the ChebyshevT or some such Std function. If I try to use (1/2*x^2*((Sqrt[1 - 1/x^2 + 1])^n + (Sqrt[1 - 1/x^2])^n)) inplace of ChebyshevT, the function is not evaluated. I expect I've forgotten some other MMA fu? $\endgroup$
    – Hedgehog
    Dec 18, 2017 at 12:17
  • $\begingroup$ @Hedgehog to define the differential operator the function expects a polynomial x^n -> d^n. In your example, you have a 1/x^2 term, thus not a polynomial. The function does not accept it due to the /; PolynomialQ[poly, x] test. If you want to manage negative x powers this is certainly possible, but you will get integro-differential operators $\endgroup$ Dec 18, 2017 at 12:27
  • $\begingroup$ If you want to compute ChebT, the formula I have is: 1/2 ((x - I Sqrt[1 - x^2])^n + (x + I Sqrt[1 - x^2])^n) and there is no 1/x^2 term. Do not forget to FullSimplify the expression to let Mathematica recognize a polynomial before calling DiffOp[] (I will be back in one hour, sorry) $\endgroup$ Dec 18, 2017 at 12:34

Assuming the idea is to translate the derivative operator (which must make this Q&A a duplicate, no?), this was what I came up with...

(* assumes op is a polynomial with constant coefficients and the variable in f is x *)
CircleDot[op_, f_] /; PolynomialQ[op, Derivative[1]] && FreeQ[op, x] :=
   With[{c = CoefficientList[op, Derivative[1]]},
    c.NestList[D[#, x] &, f, Length@c - 1]

ChebyshevT[3, Derivative[1]] \[CircleDot] BesselI[0, x] // Expand
(*  BesselI[3, x]  *)

Mathematica graphics

  • $\begingroup$ Interesting: Nested D[#, x] simplifies by Expand, but direct D[#, {x, k}] gives derivatives that do not simplify to the desired form. $\endgroup$
    – Michael E2
    Dec 17, 2017 at 19:46

Operate[] combined with Through[] are suitable for this task:

k = 5; (* example *)
FullSimplify[Through[Operate[Composition[Through, ChebyshevT[k, δ]], BesselI[0, x]]] /.
             {(c_ /; FreeQ[c, δ])[BesselI][0, x] :> c BesselI[0, x],
              (c_. δ^j_.)[BesselI][0, x] :> c Derivative[0, j][BesselI][0, x]}]
   BesselI[5, x]

Like Picaud Vincent, I create a differential operator to represent $T_n(d/(dx))$. Only I do it somewhat differently.

chebDOp[n_] :=
  Module[{rules, opExpr},
    rules = CoefficientRules[ChebyshevT[n, \[FormalU]]];
    opExpr = 
       Plus @@ 
         (#[[2]] D[BesselI[0, \[FormalU]], {\[FormalU], #[[1, 1]]}] & /@ rules);
    Function[\[FormalU], \[FormalV]] /. \[FormalV] -> opExpr]


And @@ Table[FullSimplify[chebDOp[n][x] == BesselI[n, x]], {n, 4}]



  • $\begingroup$ Thanks for the contribution. Your answer gives the same as Picaud, which, ironically is what I wanted. But his seems simpler and possibly more general? in that I can pass in any polynomial and looks like it should work - I acknowledge this is not I requirement of the question. Thanks again. $\endgroup$
    – Hedgehog
    Dec 18, 2017 at 12:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.