29.1. DIVERGENCE AND CURL OF A VECTOR FIELD 569

This establishes the first formula. The second formula is done similarly. Now consider thethird.

(∇× (∇×f))i = ε i jk∂ j (∇×f)k = ε i jk∂ j (εkrs∂r fs)

=

=ε i jk︷︸︸︷εki j εkrs∂ j (∂r fs) = (δ irδ js −δ isδ jr)∂ j (∂r fs)

= ∂ j (∂i f j)−∂ j (∂ j fi) = ∂i (∂ j f j)−∂ j (∂ j fi)

=(

∇(∇ ·f)−∇2f)

i

This establishes the third identity.Consider the fourth identity.

∇ · (f ×g) = ∂i (f ×g)i = ∂iε i jk f jgk

= ε i jk (∂i f j)gk + ε i jk f j (∂igk)

=(εki j∂i f j

)gk −

(ε jik∂igk

)fk

= ∇×f ·g−∇×g ·f.

This proves the fourth identity.Consider the fifth.

(∇× (f ×g))i = ε i jk∂ j (f ×g)k = ε i jk∂ jεkrs frgs

= εki jεkrs∂ j ( frgs) = (δ irδ js −δ isδ jr)∂ j ( frgs)

= ∂ j ( fig j)−∂ j ( f jgi)

= (∂ jg j) fi +g j∂ j fi − (∂ j f j)gi − f j (∂ jgi)

= ((∇ ·g)f +(g ·∇)(f)− (∇ ·f)g− (f ·∇)(g))i

and this establishes the fifth identity.

29.1.2 Vector PotentialsOne of the above identities says ∇ ·(∇×f) = 0. Suppose now ∇ ·g= 0. Does it follow thatthere exists f such that g = ∇×f ? It turns out that this is usually the case and when suchan f exists, it is called a vector potential. Here is one way to do it, assuming everythingis defined so the following formulas make sense.

f (x,y,z)=(∫ z

0g2 (x,y, t) dt,−

∫ z

0g1 (x,y, t) dt +

∫ x

0g3 (t,y,0) dt,0

)T

. (29.1)

In verifying this you need to use the following manipulation which will generally holdunder reasonable conditions but which has not been carefully shown yet.

∂x

∫ b

ah(x, t) dt =

∫ b

a

∂h∂x

(x, t) dt. (29.2)

The above formula seems plausible because the integral is a sort of a sum and the deriva-tive of a sum is the sum of the derivatives. However, this sort of sloppy reasoning willget you into all sorts of trouble. The formula involves the interchange of two limit opera-tions, the integral and the limit of a difference quotient. Such an interchange can only beaccomplished through a theorem. The following gives the necessary result.

29.1. DIVERGENCE AND CURL OF A VECTOR FIELD 569This establishes the first formula. The second formula is done similarly. Now consider thethird.(Vx (Vx f)); €1jkOj (VX fy = Ein} (Exrs Ars)Fhip €krs9j (Orfa) = (SirS js — 8is5 jr) O; (pfs)= 0; (d:f;) — 9; (Ajfi) = 9 (Aj fj) — 0; (O;fi)(viv-f)-V’F),LThis establishes the third identity.Consider the fourth identity.V-(f xg) = A(F XQ); = %UEijKfigeEijk (fj) Bk + Eiji (Asx)(€xij0ifj) 8k — (€ jk Bk) Se= Vxf-g-Vxg-f.This proves the fourth identity.Consider the fifth.(Vx(fxg)); = eipOi(F XG) = EijcOjEtrsfrBs= Eni jEkrsOj (fs) = (66 js _ dis6 jr) 0; (fr8s)= 9;(fig;) — 9; (fisi)= (djgj) fit 8jOjfi—(Ojf;) 81 — Fj (Oj81)= (V-g)f+9-V)(f)-(V-fig—(F-Y) (9);and this establishes the fifth identity. J29.1.2 Vector PotentialsOne of the above identities says V-(V x f) =0. Suppose now V-g = 0. Does it follow thatthere exists f such that g = V x f ? It turns out that this is usually the case and when suchan f exists, it is called a vector potential. Here is one way to do it, assuming everythingis defined so the following formulas make sense.Zz v4 x TFuna=(/ g2 (x, y,t) at,— | 81 (x,y,t) a+ | g3 (t, y,0) aro) ; (29.1)In verifying this you need to use the following manipulation which will generally holdunder reasonable conditions but which has not been carefully shown yet.d [? > Oh=| h(x,t) a= | 9, 1) dt. (29.2)The above formula seems plausible because the integral is a sort of a sum and the deriva-tive of a sum is the sum of the derivatives. However, this sort of sloppy reasoning willget you into all sorts of trouble. The formula involves the interchange of two limit opera-tions, the integral and the limit of a difference quotient. Such an interchange can only beaccomplished through a theorem. The following gives the necessary result.