1

All optical reconstruction of a crystals band structure

Online Supplementary Material

3
4

Detailed description of the procedure

In the first step, we simulate the spectrogram (analogous to that of Fig. 2a) for each band in the

trial set and extract sosc (see OSM).

The interaction of the fundamental laser with the solid is simulated by a 1D integration of the

Semiconductor Bloch Equations. The procedure is detailed in Ref. 14. To speed up the

calculation, only the spectrum of the interband polarization is considered. The high harmonic

10

spectrum is recorded for each delay between the weak second harmonic field and the strong

11

fundamental laser pulse. The delay is scanned over 2 cycles of the fundamental, in 20 steps. This

12

level of accuracy guarantees that the simulated sosc converges well within the experimental

13

uncertainty. The parameters of the simulation are detailed in Table S2.

14

We choose the trial bands to have the same functional form as the target: (k) = g + A [1

15

cos(ka)] + B [1 cos(2ka)]. The spectrograms are calculated by spanning the (A,B) parameter

16

space; the target band is not included in the set. A set of 257 trial bands are considered. The

17

simulated spectrogram for one trial band is reported in Fig. S2a. To extract sosc, each harmonic

18

order is spectrally integrated over the whole width. The oscillations of the 14th harmonic order so

19

obtained are shown if Fig. S2b (red dashed line). Fitting of these oscillations to a cosine function

20

still results in a non-zero uncertainty. The oscillation at twice the fundamental frequency is

21

isolated in the frequency domain (Fig. S2c), transformed back into the temporal domain, where

22

the amplitude is renormalized to 1 and fitted to a cosine function f(t) = cos(2t + sosc). sosc is
1

23

the only parameter of the fit. We find that, although the uncertainty becomes nominally zero, the

24

optimum phase doesn't vary appreciably if the unfiltered or filtered oscillations are considered

25

(the variation is < 10-3 rad). Therefore, the limit on the accuracy of the simulation is mostly

26

dictated by the number of delay points and the delay range.

27
28

In a second step we fit osc to sosc and quantify the quality of the fit 2i:

2N

2N, i

29

where the sum runs over the even harmonics, i labels one trial band and is the uncertainty on

30

osc. Finally, we use the 2 probability distribution to reject candidate band structures that badly

31

fit the experimental phase: a trial i is rejected if the probability of measuring 2 > i2 is lower

32

than the threshold p = 0.1.

33
34

Effect of the functional form of the trials

35

As in any variational method, the functional form of the trial dictates how accurately the

36

reconstructed bandgap approximates the real band structure. In the main text we have shown

37

that, when the target band has the same functional form as the trials, the reconstructed bands lie

38

very close to the target. In this section, instead, we generalize the method to the real ZnO target

39

band structure, obtained from Density Functional Theory calculations in the Local Density

40

Approximation (labelled LDA in the following), expanded up to the 5th spatial frequency and

41

to two sets of trial bands: one set has functional form (k) = g + A [1 cos(ka)] + B [1

42

cos(2ka)] (as in the main text, labelled NNN in Fig. S1); the other set has functional form (k)
2

43

= g (g/2)2B + [(g/2)2 + (A k a)2]B (labelled SQ in Fig. S1). In both cases only two

44

parameters (A,B) need to be scanned. The parameters of the target band are reported in Table S1.

45
46

Figure S1 shows the result of the reconstruction. The target phase and target bandgap are

47

reported as red lines (labelled LDA) in panels (a,c) and (b,d) respectively. The ZnO split-off

48

valence and first conduction bands have small effective masses and strong linearity across the

49

Brillouin Zone, a shape not accurately reproduced by the short expansion used in the main text.

50

The results with the SQ bands are shown in panels (a) and (b). All trial bands but one are

51

rejected with a confidence level of 90%, because the probability of measuring 2 > i2 is less than

52

the threshold p-value of 0.1. The only plausible band has p = 0.2 and it well approximates the

53

target bandgap, except at the edge of the Brillouin Zone, where the derivative of the target

54

vanishes as expected from the Fourier expansion. The functional form of the trials cannot

55

reproduce this feature. On the contrary, if only two spatial frequencies are used in the Fourier

56

expansion (the NNN trials, shown in panels (c) and (d)), all bands must be rejected with a 90%

57

confidence level. The two best guesses are reported in the figures, with a p-value of 6 x 10-4 and

58

8 x 10-4. They fail to reproduce both the high curvature around and the total width of the band

59

gap. Clearly, the strong linearity of the target bands cannot be captured with a Fourier expansion

60

up to only two spatially defined frequencies.

61

In conclusion, the good sensitivity of osc to the functional form of the bandgap implies a strong

62

selectivity of the method to the band structure.

63
64

Effect of the measured high harmonic range on the accuracy of the reconstruction

65

Accurate reconstruction of the band gaps requires measurement of many high harmonics. In the

66

situation presented in Fig. 2 of the main text, the measured harmonic range extends up to the

67

30th harmonic order, which is very close to the maximum bandgap of the target band structure.

68

The reconstructed band structures lie very close to the target, as shown in Fig. 2c and reported

69

below in Fig. S3a. However, many more bands are plausible when the measured harmonic

70

range extends only up to the 18th order, as shown in Fig. S3b.

71
72

The loss in the accuracy of the reconstruction can be interpreted as a loss of momentum

73

resolution. This can be taken as the difference between the maximum and minimum momenta

74

that correspond to a given harmonic photon energy in the set of plausible bands. The black dash-

75

dotted lines in Fig. S3 mark the momentum resolution k at the 20th harmonic order. Although it

76

may seem strange at first, a heuristic explanation of the dependence of the resolution on the

77

accuracy of the reconstruction is the following. Consider the trajectories depicted by the two

78

wiggled arrows of Fig. 1. Although they are mapped to two different harmonic photon energies,

79

both propagate along the lower part of the band. Therefore, there is redundant information about

80

the bottom part. The higher part is however only sensed by the top trajectory. Here, the

81

redundancy is lost. When many harmonics are measured, there is plenty of redundant

82

information about the lower part of the bands, and the resolution is therefore high. On the

83

contrary, when few harmonic orders are measured there is less redundancy and the momentum

84

resolution is lower.

85
86

Effect of the absolute phase on the reconstruction

87

The absolute optimum phase is a measurable parameter that, when used, improves the accuracy

88

of the reconstruction. Figure S4 shows the effect that varying the absolute phase of the target by

89

0.015 rad (equivalent to about half the uncertainty on the optimum phase) has on the

90

reconstruction. Only one band is plausible and has p = 0.11, just above the acceptance threshold.

91

The associated band gap, grey line of panel b, closely matches the target (red line). When the

92

absolute phase is offset by more than ~0.03 rad (equivalent to the uncertainty on the optimum

93

phase), no bands are plausible.

94
95

The sensitivity to the optimum phase can be mitigated by neglecting it. Figure S5 shows the

96

results when the average phase is removed from both the target and the trials. This corresponds

97

to a situation where the absolute phase in not measured in the experiment. The lack of the extra

98

constraint represented by matching the absolute phase results in slightly lower accuracy (panel

99

a). High precision can be restored by measuring the optimum phase more accurately (from 0.025

100

rad to 0.020 rad), as shown in panel b.

101
102

In the simulations, a variation of the minimum bang gap Eg results in an almost rigid shift of the

103

absolute optimum phase. In our model target solid of Fig. 2, for example, a variation of Eg by 0.2

104

eV results in a variation of 0.1 rad of the absolute phase. This dependence can be understood in

105

the following way. Each harmonic order is linked to a specific trajectory of the electron-hole pair

106

when it is referenced to the minimum bandgap. A harmonic photon with energy exactly equal to

107

Eg corresponds to electron-hole pairs that do not travel in the solid: they recombine as soon as

108

they are created. If the reference is lowered, the same harmonic order corresponds to emission

109

above the minimum bandgap. Therefore, the associated trajectory and its optimum phase are
5

110

different. Owing to the shape of the bands, the variation in the optimum phase differs between

111

harmonic orders. However, we find that when the variation in Eg is less than the fundamental

112

photon energy, the variation of the optimum phase is similar for all harmonic orders.

113
114

As shown in Fig. S5, a rigid shift in the optimum phase is rather unimportant, since the

115

reconstruction can also be performed by neglecting it. However, it is important to investigate the

116

effect of deviations from a rigid shift as Eg is varied. In an experiment, it is the Eg of the trials

117

that has to adapt to a given experimental measurement of osc. For simplicity, in Fig. S6 we

118

compare the same set of trials with Eg = 3.4 eV (as for Fig. 2) to three different target bands with

119

the same shape as in Fig. 2, but with Eg decreasing from 3.4 eV to 3.2 eV. As the gap is

120

decreased, the convergence to the target bandgap worsens. For the reconstruction we removed

121

the average phase, since target and trials differ by more than 0.03 rad. We find that good

122

convergence is achieved for variations < 0.2 eV of Eg between the target and the trials,

123

corresponding to ~ 60% of the fundamental photon energy.

124
125

The fundamental energy gap of solids is typically well known. Uncertainty can arise from a lack

126

of knowledge of which valence band mostly contributes to high harmonics generation (the top-

127

most valence bands can differ in the maximum energy by < 1 eV), and to bandgap

128

renormalization following strong excitation. The latter has been measured to be ~ 0.2 eV in

129

Silicon, excited by intense visible pulses [13]. The analysis presented above proves that for

130

deviations in the minimum gap between target and trials is < 0.2 eV the method is sensitive to

131

bandgap renormalization if the offset phase is measured, otherwise it returns the field-free band

132

structure. For stronger variations, either the minimum bandgap has to be measured and replaced
6

133

in the simulations, or the trials need to be tuned to correctly reconstruct the experimental band

134

structure.

135
136

Effect of the accuracy on the optimum phase for the reconstruction

137

In the current experiment, the accuracy on osc is =0.050 rad, twice as much as that used for the

138

reconstruction presented in Fig. 1. In Fig. S7 we show that the quality of the reconstruction

139

significantly deteriorates if the experimentally obtained is used. The current experiment was

140

not designed to minimize this error. A better control on the delay jitter, on the intensity

141

fluctuations of the laser source, and scans over longer delays should lower the uncertainty in

142

future experiments.

143
144

Effect of Stark shift

145

The analytical solution of the Semiconductor Bloch Equations for negligible depletion of the

146

valence band shows electron-hole pairs moving in the field-free band structure (k) (Ref. 14). As

147

found for atomic high harmonic generation [20], depletion automatically includes the effect of

148

the strong field on the energy states. In solids, this can be obtained as follows. Consider the first

149

order differential equations for the probability amplitudes (equations 4a-b of the Supplementary

150

Material in Ref. 14), formally equivalent to the Semiconductor Bloch Equations. A second order

151

equation can be obtained by deriving one of the two and inserting the other one in the resulting

152

equation. The resulting second order equation can be integrated using WKB approach, and the

153

semiclassical action results in:

154

Where d is the transition dipole moment between the bands, t and t are the emission and

155

recollision time of the trajectory of the electron-hole pair and A(t) is the laser vector potential.

156

Here, the electron propagates in the Stark shifted band structure. The effect is maximum at the

157

minimum bandgap, (k)= g, and when the field is strongest. For the parameters of Table S2, the

158

dressed bandgap energy differs from the field-free bandgap by only 2%.

159

References

160
161
162

20. M. Lewenstein, P. Balcou, M. Y. Ivanov, A. LHuillier and P. B. Corkum, Phys. Rev. A 49,

2117 (1994).

163
164

165
166

Figure S1 Reconstruction with different functional forms. Panels (a) and (b) show the result of

167

the reconstruction with the functional form of the trial bandgaps labelled SQ in the text. osc is

168

plotted in panel (a) for the target LDA band (red dots) and for the retrieved band. The

169

bandgaps are plotted in panel (b), for the target (red lines) and for the SQ retrieved band (gray

170

line). Panels (c) and (d) show the result with NNN bands. Here, no bands have sufficient

171

confidence. The two most confident ones are plotted as light grey lines. The corresponding p-

172

values are reported and labelled p.

173
174
175
176

Spatial frequency

Parameter values (eV)

(k) = g +

g = 3.4

+A+

A = 4.1714

+ B [ 1 cos(ka)] +

B = - 3.6175

+ C [1 cos(2ka)] +

C = - 0.1488

+ D [ 1 cos(3ka)] +

D = - 0.2568

+ E [ 1 cos(4ka)] +

E = - 0.07396

+ F [1 cos(5ka)].

F = -0.07431

177
178

Table S1 Band parameters for LDA ZnO target bandgap. Each term of the functional form of

179

the target ZnO band is reported in the left column; the value of the parameter is in the second

180

column. The bands are obtained from a LDA-DFT calculation and fit to the above functional

181

form.

182
183
184
185
186
187

10

Parameter
al

Value Unit
3.250

a = 3 / 2 al 2.814

3.4

eV

F0

0.285 V/

F1

0.04

F0

3.66

60

fs

T2

1.3

fs

1.8

188
189

Table S2 Parameters of the simulation. al is the lattice constant of wurtzite ZnO, a = /K, where

190

K is half of the reciprocal lattice vector along the direction, g is the minimum bandgap (at

191

), F0 and F1 are the peak field strength of the fundamental and second harmonic, is the

192

fundamental wavelength, is the pulse duration, T2 is the dephasing time of the interband

193

polarization and d is the transition dipole moment between the bands.

194

11

195
196
197

Figure S2 Analysis of simulated spectrograms. (a) Typical simulated spectrogram for a trial

198

band. The 14th harmonic order is spectrally integrated across its width and the modulation of the

199

intensity is plotted in panel (b) (red dashed line). The spectrum of the modulation is plotted in

200

panel (c) (red dashed lines). It has some anharmonic components, which are filtered out (the

201

green dashed line is the spectral filter). The filtered oscillation is reported with green dots in

202

panel (b). These oscillations are normalized to 1 and fitted to a cosine function with offset phase

203

(black solid line).

204

205

12

206

Figure S3 Effect of the measured harmonic range on the reconstruction. When only harmonics

207

up to the 18th are measured (panel b), many more bands become plausible (grey lines). Panel a

208

shows the reconstruction of Fig. 2, when harmonics up to the 30th order are measured.

209

210
211

Figure S4 Effect of the absolute phase on the reconstruction. When the target phase of Fig. 2

212

(blue dots in panel a) is offset by 0.015 rad (red dots with error bar in panel a), only one band is

213

plausible, with p-value of 0.11 (therefore lower than in Fig. 2). The retrieved phase is the grey

214

line in panel a, while the associated band structure is the grey line in panel b. The target band

215

structure is the red line in panel b.

216

13

217

Figure S5 Effect of neglecting the absolute phase on the reconstruction. When the average

218

optimum phase is removed from the trials and the target (both panels), the reconstruction returns

219

similar results to those obtained in Fig. 2, although some extra bands become plausible (panel a).

220

The same accuracy as when the absolute phase is included can be achieved by reducing the

221

experimental error on the optimum phase from = 0.025 rad (as in Fig. 2) to 0.02 rad (panel b).

222

223
224

Figure S6 Effect of the minimum bandgap on the reconstruction. The minimum bandgap Eg of

225

the target band structure is progressively decreased from 3.4 eV (as in Fig. 2) to 3.2 eV, through

226

panels a to c. The reconstruction performed with the same set of trials with Eg = 3.4 eV (as in

227

Fig. 2) fails for Eg < 3.3 eV (panel c).

228

229

14

230

Figure S7 How the accuracy on the experimental osc affects the reconstruction. (a) The same

231

data as in Fig. 1, with =0.025 rad. (b) With =0.050 rad, as in the current experiment.

15